Executive summary

Business overview

AI inside X: The company has a vision it calls “AI inside X” whereby AI is integrated into every domain. It aims to create an environment where anyone can easily develop AI products and operate them at minimal cost. As of FY03/21, handwritten text recognition AI (AI-OCR) based on deep learning was the company’s main business. FY03/21 revenue was JPY4.6bn (+188.9% YoY) and operating profit JPY2.4bn (+446.1% YoY). The mainstay product is the DX Suite AI-OCR offering.

Key features of the mainstay DX Suite: The company launched DX Suite services in November 2017. It has not disclosed the character recognition rate since its stock market listing. However, when AI inside raised funds on August 3, 2018, it revealed that its technology was able to read roughly 6,300 characters consisting of Japanese logographic characters (kanji; level 1 and 2) and two sets of Japanese syllabic characters (hiragana and katakana) with an average accuracy of 99.91%. As of FY03/21, most customers were using the cloud version of DX Suite, but some needed to use the services on premises for security reasons. The company therefore offers AI inside Cube hardware for edge computing so users can process documents on-site without accessing the cloud. The company has been adapting DX Suite for multiple languages and rolling it out globally since December 2020.

Earnings structure: AI inside uses a software as a service (SaaS) business model for DX Suite. It discloses two categories of revenue: selling model and recurring. Selling model revenue is booked for each specific transaction (initial costs or fixed-term contract). Recurring revenue is recognized continuously as long as the customer keeps using the service, and comprises pay-as-you-use costs based on the number of requests (readings) to convert forms into data by the customer and monthly fixed charges based on basic functions and options. The share of Lite plans (no initial costs) in the sales mix is increasing, and in FY03/21, recurring revenue comprised 87.7% of the total (47.1% in FY03/20). Annual recurring revenue (ARR) can be broken into number of users x ARR per user (ARRPU). As of end-FY03/21, there were 9,596 Lite plan users (+8,423 YoY), 1,034 Standard/Pro users (+334), and 87 AI inside Cube (Note: from FY03/22 on, Cube contract figures will be displayed as Edge) users (+54). The churn rate (average lost monthly usage fees over the past 12 months due to contract cancellations as a percentage of monthly usage fees for existing users) was 1.14%.

External sales channels: After enhancing product quality sufficiently using a direct sales model, from around FY03/19 the company started to make full use of external sales channels such as distributors, OEM sales, and product partners. This was aimed at avoiding a labor-intensive sales organization and major advertising expenses, while enabling swift, efficient growth in user numbers. In FY03/21, the breakdown of recurring revenue by sales channel was direct sales 31.1% (JPY1.3bn), partner sales 18.1% (JPY730mn), and OEM sales 50.8% (JPY2.0bn). The use of outside channels had a significant impact. However, there were issues with these. In April 2021, major customer Nippon Telegraph and Telephone West Corporation (NTT West) notified the company of its intention not to renew most of its licenses.

Competing services, market share, target market: AI-OCR involves using AI to read with a high degree of accuracy, but the basic function of reading text on paper is the same as traditional OCR. Accordingly, AI-OCR competes with OCR. It also competes with BPO services whereby people check data written on paper and input it once again, as well as potential customers’ internal workflows (including small companies that use inexpensive part-time personnel for these tasks). Potential customers will decide which method to use after considering cost and accuracy. BPO services could conceivably use lower-priced overseas labor, but checking Japanese-language information requires personnel familiar with the language, so cost benefits are not as easy to find as in the English-speaking world. DX Suite has a domestic AI-OCR market share of 64% (FY2019 estimates of AI-OCR solutions vendor sales and market shares [cloud] in MIC Research Institute’s OCR Solutions Market Survey–Expanding with AI-OCR, 2020 Edition, February 2020). Because the company’s DX Suite uses AI to support manual inputs, the company said its target market is a portion of the non-IT external contracting market (worth roughly JPY1.7tn, Yano Research Institute) which includes data input operations. AI inside estimates simple tasks in core operations (increasingly being performed by AI and RPA) accounted about JPY580.0bn of this segment, which it is targeting.

Earnings trends

FY03/21 results: FY03/21 revenue was JPY4.6bn (+188.9% YoY), operating profit was JPY2.4bn (+446.1% YoY), ordinary profit was JPY2.3bn (+471.9% YoY), and net income was JPY1.7bn (+295.4% YoY). Achievement rates versus the company’s revised forecasts were 102.8% for revenue, 126.9% for operating profit, 125.9% for ordinary profit, and 145.4% for net income. While revenue was generally in line with the company’s revised forecast, operating profit, ordinary profit, and net income exceeded prior forecasts as a result of SG&A expenses coming in lower than previously expected, thanks largely to lower outsourcing costs in line with efforts to improve efficiency in internal operations and the suspension of exhibitions and business trips as a result of a resurgence in new COVID-19 cases in Japan.

FY03/22 forecasts: AI inside forecasts FY03/22 revenue of JPY3.6bn (-21.5% YoY), operating profit of JPY453mn (-80.8% YoY), ordinary profit of JPY455mn (-80.5% YoY), net income of JPY279mn (-83.2% YoY), and EPS of JPY71.37. The company forecasts a recurring revenue ratio of 90%. By sales channel, the company estimates direct sales accounting for 38.3% of recurring revenue (31.1% in FY03/21), partner sales accounting for 42.8% (18.1%), and OEM sales accounting for 18.9% (50.8%). The company expects the number of NTT West “Omakase AI OCR” contracts to decline from 9,284 at the end of FY03/21 to just over 1,000 in Q1 FY03/22 (Note: figures based on notification of non-renewal of a contract from NTT West the company received in April 2021), and remain at a favorable level thereafter. It also expects the number of Lite plan users (excluding “Omakase AI OCR” users) to remain firm (there were 312 such users as of Q4 FY03/21). On the other hand, the company looks for 1,523 Standard/Pro/Edge users, up from 1,034 Standard/Pro users at the end of FY03/21. The company forecasts costs-to-revenue ratios of 24.80% for R&D, 23.24% for S&M, and 23.48% for G&A, with the overall cost-to-revenue ratio at 15.93%.

Medium-term outlook and strategy: AI inside does not disclose medium-term business plans or targets because it thinks these are difficult to formulate against the backdrop of a rapidly changing AI market. It has only disclosed its future roadmap. While the company did not provide updated insights into its roadmap at the FY03/21 results briefing, it said there was no particular change from that released in Q3 FY03/21. The company sees its mission as “To bring AI to every being in the world and contribute to a richer future society.” It has set out a platform strategy to achieve this corporate mission. Moving forward, it aims to extend its business beyond DX Suite to deliver infrastructure as a service (IaaS) that can drive AI, including AI inside Cloud, AI inside Cube, and networks; AI inside Learning Center, a tool that will enable anyone to easily create AI; and a marketplace (platform as a service or PaaS, similar to Apple’s App Store) that will enable the selection and use of AI and apps created by users themselves, other users, or the company. Further, through DX Suite membership (note: the company envisions a membership system such as Amazon Prime), AI inside strives to supply a higher-value-added free use of AI, apps, and services, and continue to enhance the user experience. As announced on April 28, 2021, the company expects that most of the DX Suite Lite plan licenses issued to NTT West will not be renewed in FY03/22. However, there have been no similar situations with other OEM partners or distributors, so it plans to keep using external sales channels.

Strengths and weaknesses

Shared Research believes the company’s strengths are 1) its entry into the AI-OCR market before competitors, which enabled data collection and high accuracy; 2) its privacy control function serving as a barrier to entry; and 3) expansion of sales channels through external partners.

We believe the company’s weaknesses are 1) difficulty in detecting problems before churn rates rise due to OEM supply arrangements; 2) its reliance on domestic AI-OCR for majority of revenue in short term; and 3) the need to distinguish itself to avoid direct competition with massively capitalized global competitors in the AI field.

Key financial data

Recent updates

1H FY03/22 flash update

On November 10, 2021, AI inside Inc. announced earnings results for 1H FY03/22.

Trends and outlook

Quarterly trends and results

1H FY03/22 results

The company’s 1H FY03/22 revenue was JPY1.7bn (-12.6% YoY), operating profit was JPY298mn (-70.5% YoY), ordinary profit was JPY290mn (-71.3% YoY), and net income was JPY192mn (-68.1% YoY). Discontinuation of NTT West “Omakase AI OCR” (equivalent to DX Suite Lite) contracts in Q1 (released on April 28, 2021) lead to the fall in revenue and operating profit. Progress rates versus the company’s FY03/22 forecast were 47.2% for revenue, 65.8% for operating profit, 63.8% for ordinary profit, and 68.7% for net income.

AI inside kept its FY03/22 forecast announced on May 12, 2021 unchanged. The company aims o achieve full-year revenue target through initiatives to increase contracts via sales partners and acceleration of its AI inside Cube series sales. For details on the assumptions behind the company’s forecast, please see the FY03/22 company forecast section later in this report.

Performance indicators

1H FY03/22 recurring revenue was JPY1.6bn (-2.8% YoY) and selling model revenue was JPY145mn (-58.0% YoY). The company has revised service fees to focus on recurring revenue, which led to a sharp reduction in selling model revenue. The recurring revenue ratio has reached 91%. A breakdown of 1H revenue shows NTT West’s “Omakase AI OCR” accounting for JPY334mn, and revenue unrelated to this accounting for JPY1.4bn.

DX Suite and Intelligent OCR contracts numbered 1,906, including 692 Lite contracts (including OEM) and 1,214 Standard/Pro/Edge (DX Suite for AL inside Cube series) contracts (initial target of 1,523 contracts for the full-year). There were 122 Edge contracts. The number of NTT West “Omakase AI OCR” contracts (equivalent to DX Suite Light) fell from 9,284 as of end-March 2021 to 141. The number of Lite plan contract excluding Omakase AI OCR totaled 551. There were 776 Elastic Sorter users.

The number of DX Suite users came to 33,573 (30,539 in 1H FY03/21). The user/account number was 18.8 (2.4 in 1H FY03/21).

The overall churn rate was not released. The Standard/Pro churn rate was 0.55% (monthly average between October 2020 and September 2021 for this and the following), while the churn rate excluding an OEM partner, NTT West’s “Omakase AI OCR” was 0.53%.

In terms of costs, as a ratio to revenue, R&D was 17.2%, S&M was 22.1%, G&A was 19.0%, and the cost-to-revenue ratio was 24.3%.
Cost of revenue in 1H FY03/22 was JPY413mn. Server costs amounted to JPY93mn. In In April 2021, the company officially released Workflows, a successor product to "DX-Suite," and "Learning Center" which enables simple application development with no code. It recognized post-release maintenance fees as cost of revenue, which led to labor costs of JPY105mn and outsourcing costs of JPY177mn. 

The number of sales partners increased to 102, from 99 at the time of announcement of FY03/21 results.

Full-year FY03/22 company forecast

Initial FY03/22 company forecast (out May 12, 2021)

AI inside forecasts FY03/22 revenue of JPY3.6bn (-21.5% YoY), operating profit of JPY453mn (-80.8% YoY), ordinary profit of JPY455mn (-80.5% YoY), net income of JPY279mn (-83.2% YoY), and EPS of JPY71.37. The company maintained that it would continue to advance its sales partner (including OEM) strategies. In the direct sales channel, the company aims to focus on customers with high ACV (annual contract value). The company’s forecasts are calculated by taking into account conditions that are usually predictable. However, there are number of unpredictable factors at present that could affect the company's performance moving forward, and management has accordingly indicated that it could revise its forecasts if any of these factors comes into play.

Revenue

The company forecasts a recurring revenue ratio of 90%. By sales channel, the company estimates direct sales accounting for 38.3% of recurring revenue (31.1% in FY03/21), partner sales accounting for 42.8% (18.1%), and OEM sales accounting for 18.9% (50.8%). In value terms, it forecasts revenue from direct sales of JPY1.2bn (JPY1.3bn in FY03/21), partner sales of JPY1.4bn (JPY730mn), and OEM sales of JPY614mn (JPY2.0bn). The company expects direct sales to be flat, with ongoing growth in partner sales. It expects an impact on OEM sales from the lapse of licenses at NTT West discussed below. Selling model revenue include revenue from the launch of Learning Center.

OEM sales to NTT West

In line with the announcement released on April 28, 2021, the company expects many of the DX Suite Lite plan licenses issued to OEM partner NTT West will not be renewed in FY03/22. As such, the company in FY03/22 does not expect to book the JPY1.8bn portion of the JPY2.1bn in revenue from NTT West DX Suite Lite plan licenses booked in FY03/21. On the other hand, the churn rate for DX Suite licenses supplied to customers by the company and its sales partners has remained at a relatively low level, with the company focused on bolstering revenue by attracting new customers through enhanced sales activities. The company forecasts revenue tied to NTT West’s “Omakase AI OCR” of JPY366mn (JPY2.1bn in FY03/21), with non-related revenue reaching JPY3.2bn (JPY2.5bn in FY03/21). While the company has not made an explicit forecast, it expects OEM sales to account for JPY614mn of FY03/22 recurring revenue and also expects revenue of JPY366mn related to NTT West’s “Omakase AI OCR,” implying that it expects revenue of about JPY250mn from OEM sales to NTT East and NTT Data.

Learning Center

In April 2021, the company released Learning Center, an AI platform that allows non-programmers to create AI applications without coding. This product is expected to play a central role in the company’s AI platform business, which the company sees as a future earnings pillar. It does not expect significant revenue in FY03/22, and targets about 40 use cases (half in OCR and half in other areas). The company sees FY03/22 as a year for strengthening the proposal abilities of its sales staff, enhancing product functions, creating use case examples, and building a foundation for securing profitability from FY03/23. The company has flagged that it expects to move onto marketplace business development after it has achieved 40 use cases.

Case 1: At waste treatment facilities, it is necessary to sort and remove any hazardous materials mixed in with materials not suitable for incineration during the crushing and processing stage. A user developed AI to automate the process of hazardous material detection using Learning Center, with the finished AI installed in AI inside Cube.

Case 2: Waiting for elevators in factories can delay logistics activities. A customer employed Learning Center to develop AI to understand and analyze the operational status in real time.

Case 3: A customer developed AI to inspect the appearance of holograms on manufactured IC cards at the time of shipment. In just 30 days, manufacturing site staff used Learning Center to plan, prepare the training dataset, develop AI, and verify accuracy. Despite the small dataset used for development (about 100 images), it has achieved an accuracy of roughly 90%, and will continue to improve.

Workflows

In April 2021, the company released the alpha version of Workflows, but has not incorporated any revenue from the business in its FY03/22 forecast.

Costs

The company forecasts costs-to-revenue ratios of 24.80% for R&D, 23.24% for S&M, and 23.48% for G&A, with the overall cost-to-revenue ratio at 15.93%. It expects increases in cost of revenue and SG&A expenses in line with an increase in personnel costs tied to the sale, support, and maintenance of not only DX Suite, which has been available for some time, but also Learning Center and Workflowsα, which the company began supplying in April 2021. The company also expects an increase in personnel costs in line with the expected expansion in the company’s business in other areas. The company forecasts for non-personnel expenses are calculated by estimating increases tied to the expansion in business based on actual results, and then adding depreciation on software assets and tangible fixed assets such as the company’s data center servers as well as AI inside Cube rental assets.

The company has forecast cost of sales reflecting declining contract numbers at NTT West and growth elsewhere. Personnel expenses comprise most of SG&A expenses, and are fixed. In FY03/21, the company incurred higher R&D expenses for developing new products such as Learning Center, Workflows, and Digital Input. The company will focus on updates in FY03/22, so expects the YoY growth rate to be moderate. It expects sharp growth in sales & marketing and general & administration expenses as it increased headcount in 2H FY03/21.

Extraordinary gains and losses

The company has not included estimates for extraordinary gains and losses in its initial forecast due to the difficulty of calculating such at the beginning of the term.

Contract count

The company expects the number of NTT West “Omakase AI OCR” contracts to decline from 9,284 at the end of FY03/21 to just over 1,000 in Q1 FY03/22 (Note: number based on the notification of non-renewal from NTT West received in April 2021), and remain at a favorable level after this impact has wound down. It thinks it will be possible to reduce the churn rate to levels in line with direct sales and other OEM partners if it can succeed in onboarding (post-installation support to promote ongoing usage) and other support initiatives. The company also expects the number of Lite plan contracts (excluding “Omakase AI OCR” contracts) to remain firm (there were 312 such contracts as of Q4 FY03/21). It expects growth in Lite plan contracts to come mainly from distributors and OEM sales. On the other hand, the company looks for 1,523 Standard/Pro/Edge contracts, up from 1,034 Pro/Standard contracts at the end of FY03/21. Standard/Pro contracts account for a substantial proportion of total revenue, and the company sees Standard/Pro contract count as a KPI, so released contract forecasts.

Price changes

While the company is looking into lowering Lite plan contract prices with several partners, it has not included any impact from price revisions in its forecast. It said it would decide whether to change prices after estimating the negative impact on revenue from lower prices and the positive impact from increased contract numbers. AI inside said it had no thoughts of lowering prices to levels where it could not make a profit in the pursuit of contract number growth.

Organizational change and strengthened management structure

The company restructured its organization in April 2021. Previously, the R&D and sales divisions had different roles, but the company shifted to a business domain oriented structure where units would collaborate with each other to facilitate flexible, agile decision-making. The company said that its newly appointed chief operating officer (COO) would oversee the sales and R&D divisions and strengthen collaboration between them and develop services to meet customer needs and provide high- value-added solutions. The company also appointed a chief experience officer (CXO) to carry out its platform strategy to improve the customer journey and further optimize their experience with AI.

The company also nominated Yoshiaki Sakito as a candidate for director, Kenichi Hoshi for outside director, and Kyoichiro Suzuki for advisor. The aim of the appointments is to accelerate the rollout of the company’s AI platform strategy and provide a high-value-added platform.

Yoshiaki Sakito: During his time at Apple Japan, he helped the iPod mini become a hit and a market leader in Japan with his unique marketing approach. The company has high hopes for his enterprise sales and marketplace perspectives.

Kenichi Hoshi: At Amazon Japan, he contributed to the construction and expansion of a marketplace by building a mechanism to scale up the business and operating the business from a customer perspective. The company expects that his knowledge will serve it well in its marketplace business, which will be a focus going forward.

Kyoichiro Suzuki: He was in charge of building Microsoft’s developer community and ecosystem and driving its platform strategy. The company hopes his insights will help build the developer community at Learning Center, another focus going forward.

Medium-term outlook and strategy 

AI inside does not disclose medium-term business plans or targets because it thinks these are difficult to formulate against the backdrop of a rapidly changing AI market. It has only disclosed its future roadmap. While the company did not provide updated insights into its roadmap at the FY03/21 results briefing, it said there was no particular change from that released in Q3 FY03/21. Please refer to Company forecast section for FY03/22 progress.

Note: As announced on April 28, 2021, the company expects that most of the DX Suite Lite plan licenses issued to NTT West will not be renewed in FY03/22. However, there have been no similar situations with other OEM partners or distributors, so it plans to keep using external sales channels.

Maintaining competitive advantage in handwritten character recognition

Privacy control strategy

With its privacy controls, the company leaves the choice of whether the AI engine learns or not up to the user (the default position is no learning). Competitors trying to enhance their accuracy in handwritten character recognition need a certain volume of data, but the company’s products already are sufficiently accurate, and any benefit from obtaining more data would be minimal. Meanwhile, concerns about privacy from having AI engines learn have made some users reluctant to adopt it. Privacy controls seem to be an effective strategy for dealing with competitors and users.

Expanded functionality

The company aims to allow the system to support non-paper input methods (digital input). As part of its efforts to bolster digital input operations, the company announced the conclusion of a capital and business alliance with Showcase (TSE1: 3909) and the conversion of that company to an equity-method affiliate in November 2020. The company thinks that applications and contracts will increasingly be processed using web-based forms. Since the company’s DX Suite inputs from paper forms, the user has already determined registrations and settings for what items should be written and what type of output should be used. It thinks that web-based forms can automatically be generated from this information. The company’s focus in digital input support is on creating a web-based form and when the end-user applies using the generated URL, the application can be completed without the need for OCR. Since DX Suite would be able to handle both paper and digital formats within the same workflow process, the advantage of the system is that output data can be consolidated as part of a single process.

The company is also working together with the services offered by other companies and cooperating in areas such as multifunctional equipment. The company on November 11, 2020 launched its product-centered partnership with Fujifilm Holdings (TSE1: 4901) subsidiary Fuji Xerox, which manufactures and sells multifunction equipment (copiers and printers). The company thinks collaborating with multifunction copiers and printers is likely to improve the usability of the system. In particular, since the resolution and other settings deemed appropriate for AI-OCR processing is generally pre-set in multifunction equipment, the user can perform all the necessary operations, from scanning to AI-OCR processing, with a simple touch of a button on the multifunction copier or printer’s display panel, effectively streamlining the conversion of form-based information to digital data. Moreover, electronic forms can be uploaded directly to cloud services and DX-Suite can be used in the centralized management of AI-OCR processed form data.

Geographic expansion

DX Suite development for global markets

AI inside is in the midst of launching its global development efforts for DX Suite in Asia. The company on December 23, 2020 launched an AI engine for reading printed and handwritten characters in traditional Chinese, Thai, and Vietnamese. This makes the cloud version of DX Suite truly multilingual (English is already supported). The company aims to further expand the number of languages supported. As of May 2021, it was marketing DX Suite in China, Taiwan, Vietnam, and Thailand and had completed a pilot study in Taiwan. It is running pilot studies in Vietnam and Thailand. The company said that it selected Asia as the first region for overseas development on expectations for the use of local languages other than English, due to the fact that the region was geographically close to Japan (no significant time difference), and its belief that it was an area in which the company appeared relatively likely to find business partners. In terms of its relationship with its overseas partners, the company expects the same relationships as with its domestic sales partners, OEM partners, and product-centered partners.

Diversifying AI

Platform strategy

The company sees its mission as “To bring AI to every being in the world and contribute to a richer future society.” It has set out a platform strategy to achieve this corporate mission. Moving forward, it aims to extend its business beyond DX Suite to deliver infrastructure as a service (IaaS) that can drive AI, including AI inside Cloud, AI inside Cube, and networks; AI inside Learning Center, a tool that will enable anyone to easily create AI; and a marketplace (platform as a service or PaaS, similar to Apple’s App Store) that will enable the selection and use of AI and apps created by users themselves, other users, or the company. Further, through DX Suite membership (Note: The company envisions a membership system such as Amazon Prime), AI inside strives to supply a higher value-added free use of AI, apps, and services, and continue to enhance the user experience.

Learning Center

In April 2021, the company released Learning Center, an AI platform that allows non-programmers to create AI applications without coding. This product is expected to play a central role in the company’s AI platform business, which the company sees as a future earnings pillar. It does not expect significant revenue in FY03/22, and targets about 40 use cases (half in OCR and half in other areas). The company sees FY03/22 as a year for strengthening the proposal abilities of its sales staff, enhancing product functions, creating use case examples, and building a foundation for securing profitability from FY03/23. The company has flagged that it expects to move onto marketplace business development after it has achieved 40 use cases.

Case 1: At waste treatment facilities, it is necessary to sort and remove any hazardous materials mixed in with materials not suitable for incineration during the crushing and processing stage. A user developed AI to automate the process of hazardous material detection using Learning Center, with the finished AI installed in AI inside Cube.

Case 2: Waiting for elevators in factories can delay logistics activities. A customer employed Learning Center to develop AI to understand and analyze the operational status in real time.

Case 3: A customer developed AI to inspect the appearance of holograms on manufactured IC cards at the time of shipment. In just 30 days, manufacturing site staff used Learning Center to plan, prepare the training dataset, develop AI, and verify accuracy. Despite the small dataset used for development (about 100 images), it has achieved an accuracy of roughly 90%, and will continue to improve.

AI inside Cube, Data center

AI inside is developing more efficient hardware and using it to build a unique infrastructure. In the medium term, it plans to build infrastructure that can use AI inside Learning Center.

Full-scale rollout of AI inside X* (from FY03/22 or later)

In FY03/22 or later, AI inside plans to ramp up its AI inside X vision, which targets the non-IT BPO market; further diversify AI applications that can be generated through the simple use of a UI; and develop infrastructure necessary to support AI software.

Business

Business model

AI inside X: The company has a vision it calls “AI inside X” whereby AI is integrated into every domain. It aims to create an environment where anyone can easily develop AI products and operate them at minimal cost. As of FY03/21, handwritten text recognition AI (AI-OCR) based on deep learning was the company’s main business. FY03/21 revenue was JPY4.6bn (+188.9% YoY) and operating profit JPY2.4bn (+446.1% YoY). The mainstay product is the DX Suite AI-OCR offering.

Key features of the mainstay DX Suite

The company launched DX Suite services in November 2017. It has not disclosed the character recognition rate since its stock market listing. However, when AI inside raised funds on August 3, 2018, it revealed that its technology was able to read roughly 6,300 characters consisting of Japanese logographic characters (kanji; level 1 and 2) and two sets of Japanese syllabic characters (hiragana and katakana) with an average accuracy of 99.91%. As of FY03/21, most customers were using the cloud version of DX Suite, but some needed to use the services on premises for security reasons. The company therefore offers AI inside Cube hardware for edge computing so users can process documents on-site without accessing the cloud. The company has been adapting DX Suite for multiple languages and rolling it out globally since December 2020.

Earnings structure

AI inside uses a software as a service (SaaS) business model for DX Suite. It discloses two categories of revenue: selling model and recurring. Selling model revenue is booked for each specific transaction (initial costs or fixed-term contract). Recurring revenue is recognized continuously as long as the customer keeps using the service, and comprises pay-as-you-use costs based on the number of requests (readings) to convert forms into data by the customer and monthly fixed charges based on basic functions and options. The share of Lite plans (no initial costs) in the sales mix is increasing, and in FY03/21, recurring revenue comprised 87.7% of the total (47.1% in FY03/20). Annual recurring revenue (ARR) can be broken into number of users x ARR per user (ARRPU). As of end-FY03/21, there were 9,596 Lite plan users (+8,423 YoY), 1,034 Standard/Pro users (+334), and 87 AI inside Cube (Note: From FY03/22, the number of Cube contracts will be disclosed as Edge) users (+54). The churn rate (average lost monthly usage fees over the past 12 months due to contract cancellations as a percentage of monthly usage fees for existing users) was 1.14%.

External sales channels

After enhancing product quality sufficiently using a direct sales model, from around FY03/19 the company started to make full use of external sales channels such as distributors, OEM sales, and product partners. This was aimed at avoiding a labor-intensive sales organization and major advertising expenses, while enabling swift, efficient growth in user numbers. In FY03/21, the breakdown of recurring revenue by sales channel was direct sales 31.1% (JPY1.3bn), partner sales 18.1% (JPY730mn), and OEM sales 50.8% (JPY2.0bn). The use of outside channels had a significant impact. However, there were issues with these. In April 2021, major customer Nippon Telegraph and Telephone West Corporation (NTT West) notified the company of its intention not to renew most of its licenses.

Competing services, market share, target market

AI-OCR involves using AI to read with a high degree of accuracy, but the basic function of reading text on paper is the same as traditional OCR. Accordingly, AI-OCR competes with OCR. It also competes with BPO services whereby people check data written on paper and input it once again, as well as potential customers’ internal workflows (including small companies that use inexpensive part-time personnel for these tasks). Potential customers will decide which method to use after considering cost and accuracy. BPO services could conceivably use lower-priced overseas labor, but checking Japanese-language information requires personnel familiar with the language, so cost benefits are not as easy to find as in the English-speaking world. DX Suite has a domestic AI-OCR market share of 64% (FY2019 estimates of AI-OCR solutions vendor sales and market shares [cloud] in MIC Research Institute’s OCR Solutions Market Survey–Expanding with AI-OCR, 2020 Edition, February 2020). Because the company’s DX Suite uses AI to support manual inputs, the company said its target market is a portion of the non-IT external contracting market (worth roughly JPY1.7tn, Yano Research Institute) which includes data input operations. AI inside estimates simple tasks in core operations (increasingly being performed by AI and RPA) accounted about JPY580.0bn of this segment, which it is targeting.

Overview of key services

The company sees its mission as bringing AI to every being in the world and contributing to a richer future society. To this end, it provides a wide range of AI-related services. These include Workflows (such as DX Suite) that will enable the use of AI and apps created by users themselves, other users, or the company; infrastructure that can drive AI, including AI inside Cloud, AI inside Cube, and networks; and AI inside Learning Center, a tool that will enable anyone to easily create AI.

Workflows α version (using AI)

This service builds workflows that can flexibly handle AI and apps. Examples of the service include DX Suite (AI-OCR), FaceAuth (automatic identification services, only provided as an API as of May 2021), and Digital Input (enables simple website form production, launch imminent as of May 2021).

DX Suite (cloud service: Intelligent OCR and others)

The company has developed AI software that can recognize handwritten text through deep learning. It provides various AI-OCR services through its DX Suite, which features an easy-to-use user interface designed to enable anyone to use AI for everyday tasks. DX Suite offers tools to automatically sort multiple types of randomly scanned sheets/documents (regardless of structured or semi-structured sheets/documents), and also comes equipped with tools for managing access privileges.

Service menu

DX Suite comprises three applications: (1) Intelligent OCR (basic application), (2) Elastic Sorter (optional application), and (3) Multi Form (optional application). Users can select the applications they wish to use when signing up.

Intelligent OCR is an application that uses handwriting recognition technology to read and digitize structured sheets/documents, which are defined as documents with a standardized layout, allowing the user to select text areas to be processed in advance. For example, various types of application forms, order forms, and surveys can be digitized through use of the service. Customers set areas to be scanned and read on paper documents and upload files, and the service returns results after reading the documents. The customer checks the results, corrects errors as necessary, and downloads the results through files such as CSV format. Intelligent OCR can read addresses that carry over two lines, automatically complete entries outside boxes, and automatically correct distortions and skewing in captured images. As of May 2021, it could handle Japanese, English, Thai, Vietnamese, and traditional Chinese.

Elastic Sorter is an optional application that complements Intelligent OCR. It leverages AI to sort randomly ordered structured documents by type. For example, multiple personal identification documents (such as driver’s licenses, insurance cards, and resident cards) or application forms can be sorted by type through Elastic Sorter. In this way, customers can prepare data to be processed by Intelligent OCR.

Multi Form is an optional application that complements Intelligent OCR. It reads semi-structured sheets/documents, and structures and digitizes related information. Semi-structured sheets/documents are sheets/documents that differ widely in format (e.g., text fields in different position, different layouts) but contain the same information. Because they cannot be categorized into a limited number of document types, they cannot be sorted using Elastic Sorter. For example, Multi Form can—simultaneously and without prior preparation and setting—read and digitize multiple types of sheets/documents such as invoices, receipts, and resident cards. Multi Form is provided free of cost for a special campaign, but the company plans to integrate it in basic services because it expects the demand to rise for non-standard form reading.

Pricing structure

AI inside uses a software as a service (SaaS) business model for DX Suite. It discloses two categories of revenue: selling model and recurring. Selling model revenue is booked for each specific transaction (initial costs or fixed-term contract). Recurring revenue is recognized continuously as long as the customer keeps using the service, and comprises pay-as-you-use costs based on the number of requests (readings) to convert forms into data by the customer and monthly fixed charges based on basic functions and options. Note: The initial cost is booked as unearned revenue and allocated to revenue over the service period.

The pricing structure for the DX Suite service (direct sales and partner sales) varies based on trial use or full adoption, and also depends on the inclusion of optional applications such as Elastic Sorter (leverages AI to sort scanned and randomly ordered structured sheets/documents by type) and Multi Form (reads semi-structured sheets/documents, and structures and digitizes information). For Intelligent OCR, the basic application of DX Suite, AI inside charges initial fees (selling model revenue) and basic fees and pay-for-use fees (recurring revenue). It offers a one-month trial period for an initial fee of JPY200,000 (JPY30,000 on the Lite plan), during which customers may place 200,000 free use (6,000 on the Lite plan). During the trial period, users generally try to maximize the number of OCR scans (requests), and familiarize themselves with the system. Once they decide to move to full adoption of the service, users have a choice between three payment plans (Pro, Standard, and Lite), depending on the combination of initial and monthly fees. The company said that more than 90% of direct sales customers migrated from the trial version to full service as of FY03/21.

The Standard and Pro plans include the initial fee of JPY200,000, and a different basic monthly fee of JPY100,000 for the Standard plan (includes 50,000 free use) and JPY200,000 for the Pro plan (includes 200,000 free use). This is aimed at large corporations with heavy usage volume. Once users exceed the limit of free use, they are charged an additional usage fee of JPY1 per request (for requests that do not involve text processing, JPY0.5 per request on the Standard plan and JPY0.1 on the Pro plan). Requests are counted by number of items rather than pages. As an illustration of the free use quota, assuming an application form or invoice has 30 items per page, 50,000 uses is equivalent to 1,660 pages. DX Suite Lite is aimed at companies with lower usage volume (SMEs). Customers can start using the DX Suite functions at relatively low cost (no initial cost, basic fee of JPY30,000, and JPY3 per request over free use quota of 6,000). Elastic Sorter is offered at a monthly basic fee of JPY20,000 plus pay-for-use fees of JPY1 per sorting job. For Multi Form, users are charged free monthly basic fee and a pay-for-use fee of JPY1 per sheet/document sorting job (currently free due to campaign).

OEM partners determine their own pricing structure. For example, NTT West offers its “Omakase AI OCR” service for a basic monthly fee of JPY33,000 under a one-year contract, with a minimum contract term of one year. The plan entitles users to 6,000 free use (calculated as number of pages processed per month x number of scanned items per page). Additional requests incur a pay-as-you-use fee of JPY3.3 per request. NTT EAST offers its “AI Yomitoru” service for a basic monthly fee of JPY110,000, which includes 60,000 free processing requests (minimum contract term of one year). Additional requests are charged at a pay-as-you-use fee of JPY0.55 per request for items read that either return blank results or consist of checkboxes or stamps, and JPY1.1 per request for addresses, phone numbers, names, and other items read.

AI-OCR involves using AI to read with a high degree of accuracy, but the basic function of reading text on paper is the same as traditional OCR. Accordingly, AI-OCR competes with OCR. It also competes with BPO services whereby people check data written on paper and input it once again, as well as potential customers’ internal workflows (including small companies that use inexpensive part-time personnel for these tasks). Potential customers will decide which method to use after considering cost and accuracy. BPO services could conceivably use lower-priced overseas labor, but checking Japanese-language information requires personnel familiar with the language, so cost benefits are not as easy to find as in the English-speaking world. The presence of such competitors means there is a limit to how high the company can set prices.

Customer base

As of end-1H FY03/19, AI inside provided services to a range of industries, including systems development, banking, securities, insurance, retail, energy, distribution, pharmaceuticals, real estate, and printing. The company first won customers in the financial industry (particularly, banks, securities companies, and insurers), and used its track record in this field to expand into other industries. The company assumed the financial industry presented the highest barriers to entry due to its demanding security requirements, and therefore believed a favorable reputation in the financial industry would open the door to other industries. The company said that AI inside Cube, the on-premises version (discussed later) is suited to heavy users with a single location (for example BPO companies), whereas the cloud version is more appropriate for customers with small usage volume and multiple locations (such as branches). The company is collaborating with NTT Data to deliver DX Suite to data centers for the Local Government Wide Area Network (LGWAN), a secure government network that connects internal networks of local government bodies.

Leveraging AI-OCR technology to recognize handwritten text

The company’s technology combines optical character recognition (OCR) technology with AI to enable highly accurate recognition and digitization of handwritten text. Digitization of handwritten sheets/documents has long relied on human (visual and manual) input. The company’s technology not only helps to speed up data input and curb related personnel expenses, but it can also process sheets/documents without relying on human vision, as well as break up and anonymize^* personal information to such an extent that it can no longer be traced back to original sources.

AI inside started offering an early version of its AI-powered handwritten text recognition service soon after it was established in August 2015, and developed DX Suite—its AI-OCR service for handwritten forms—into a full-fledged product and business by November 2017. The company’s customers were already using DX Suite when competitor Cogent Labs rolled out its AI-OCR solution Tegaki in August 2017. AI inside has secured access to large amounts of data by increasing corporate customers (mainly major companies), and this has allowed it to improve its text recognition accuracy. The company has not disclosed its recognition rate since it went public. However, when it raised funds on August 3, 2018, it revealed its technology was able to read a roughly 6,300 characters consisting of Japanese logographic characters (kanji; level 1 and 2) and two sets of Japanese syllabic characters (hiragana and katakana) with an average accuracy of 99.91%. AI inside has developed algorithms that can generate artificial images of handwritten text, and this has dramatically expanded the training datasets, which is a key factor behind its high text recognition accuracy. Its technologies can not only read handwritten text, but also ignore characters and dotted lines that are not intended to be read. The company had obtained six patents related to text recognition technology as of end-FY03/20.

This high text recognition accuracy is realized with three distinct types of algorithms the company has developed over time: inference algorithms, learning algorithms, and data-generating algorithms. Inference algorithms apply zone-based text recognition technology, as well as algorithms used in the fields of object and speech recognition. AI inside has developed algorithms that can generate artificial images of handwritten text, which has dramatically expanded the training datasets (data from which computers can learn) at its disposal. This has been a key factor in achieving its high text recognition accuracy. The company’s technologies can ignore characters and dotted lines that are not intended to be read, in addition to reading handwritten text.

Simple operation

CEO Toguchi believes customers will not adopt solutions that cannot be integrated in business workflows. Much of the development of DX Suite has therefore concentrated on creating a user-friendly user interface and adding security functions, in addition to improving the text recognition rate. As a result, DX Suite has become a service that can be easily operated by anyone, with OCR and workflow settings only requiring a few mouse clicks. The same applies to checking read data. AI inside has already obtained three technology patents for its security features, which are indispensable for services provided to leading financial institutions and other companies. (The company had acquired six patents related to text recognition technology as of end-FY03/20; see the Career summary table in the Top management section.)

The company is making progress in tie-ups with the services of other companies and in integration with products such as multifunctional equipment. The company in November 2020 launched its product-centered partnership with Fujifilm Holdings (TSE1: 4901) subsidiary Fuji Xerox, which manufactures and sells multifunction equipment (copiers and printers). The company believes integrating its products with multifunction copiers and printers is likely to improve the usability of the system. In particular, since the resolution and other settings deemed appropriate for AI-OCR processing is generally pre-set in multifunction equipment, the user can perform all the necessary operations, from scanning to AI-OCR processing, with a simple touch of a button on the multifunction copier or printer’s display panel, effectively streamlining the conversion of form-based information to digital data. Moreover, electronic forms can be uploaded directly to cloud services and DX-Suite can be used in the centralized management of AI-OCR processed form data.

Privacy control functions

Privacy control functions are incorporated in Intelligent OCR used on the DX Suite platform. The user chooses whether or not the AI engine learns. Setting to the ON position permits AI learning, and data is used to increase the OCR reading accuracy. When it is set to OFF, the AI does not learn the data. Regardless of which setting the customer uses, they receive unlimited AI technology updates from the company. Competitors trying to enhance their accuracy need a certain volume of data, but the company’s products already are sufficiently accurate, and any benefit from obtaining more data would be minimal. Meanwhile, concerns about privacy from having AI engines learn have made some users reluctant to adopt it. Privacy controls seem to be an effective strategy for dealing with competitors and users.

FaceAuth

In February 2021, the company announced the launch of a facial recognition API for ID verification. Combined with the company’s DX Suite AI-OCR service, it enables the reading of text on identity documents and facial recognition of IDs to streamline ID verification processes. The API has no direct relationship with the business and capital tie-up with Showcase Co., Ltd. (TSE1: 3909) announced in November 2020. AI inside developed it independently. While its use will likely be limited to certain industries, the company thinks there is considerable potential in specific ones such as finance. It hopes its use in such industries would improve the convenience of DX Suite, and have a positive impact on contract numbers and usage volume. FaceAuth for ID verification is provided under a strict policy to protect privacy and respect the interests of individuals. Only facial images obtained in accordance with personal information protection laws are used, and attributes such as race, nationality, religion, age, and gender are not determined. Use is limited to verifying the identification of a particular person, and there is no search function to identify one person from an image of many individuals. 

Digital Input

This is a service to easily create web-based forms, and the company plans to release it in the near future (as of May 2021). The company thinks that applications and contracts will increasingly be processed using web-based forms. Since the company’s DX Suite inputs from paper forms, the user has already determined registrations and settings for what items should be written and what type of output should be used. The company thinks that web-based forms can automatically be generated from this information. The company’s focus in digital input support is on creating a web-based form and when the end-user applies using the generated URL, the application can be completed without the need for OCR. Since DX Suite would be able to handle both paper and digital formats within the same workflow process, the advantage of the system is that output data can be consolidated as part of a single process.

Company concludes a capital and business alliance agreement and makes Showcase Inc. (TSE1: 3909), an equity-method affiliate

At the Board of Directors meeting held on November 30, 2020, the company passed a resolution to participate in a third-party allotment of shares by Showcase and to conclude a capital and business alliance agreement.

The company will acquire 1,771,000 new shares (20.71% shareholding after new share issue) in Showcase for JPY1.2bn and make it an equity-method affiliate. The date of payment and start of the business alliance will be December 16, 2020. A third-party allotment was deemed appropriate as Showcase was at a point in its development where it required funds to ensure further growth.

Showcase’s Form Assist is an Entry Form Optimization tool used to improve input rates so as to maximize results. More specifically, it is used to cut down on customer opportunity losses by reducing user disengagement during digital input, including by facilitating changes in the appearance of entry forms, providing analysis for conversion upgrades, and establishing conversion improvement protocols, including for customization and A/B testing. Form Assist has a strong market share in the financial industry, accounting for 65.5% (an estimate for FY2019) of vendor sales in that industry’s entry form optimization market according to the ITR Market View survey of the e-mail/Internet marketing market in 2020. Showcase has stated that the number of forms in use now stands at more than 5,000.

According to AI inside, Showcase’s Form Assist is highly compatible with the company’s DX Suite digital input services. Linking these two services will enhance customer experience and improve service quality for AI Inside users. In addition, Showcase's eKYC (electronic Know Your Customer) tool, ProTech ID checker, will improve service quality by linking with the company’s own ID verification AI capabilities. Business performance is expected to improve through leveraging the sales networks and customer bases of both companies to capitalize on high demand from financial institutions, insurance companies and government agencies, which are users and prospective customers of the company.

AI inside Cube (driving AI)

The AI inside Cube series is an edge computer designed to treat hardware and software as one system. Mainly, it will be provided as a service comprising hardware and an on-premises version of DX Suite as a set for customers that need an on-site environment to manage data (for example government agencies and local governments). The company said that some customers with locations concentrated in one area and no strong reason to choose the cloud version of DX Suite use AI inside Cube as it makes sense from a cost perspective. Company policy is to have customers use whichever offering is better for them, be it the cloud version of DX Suite or AI inside Cube. For the company, AI inside Cube is significant from a strategic perspective as it will form infrastructure for Learning Center (discussed later). The company has made AI inside Cube versatile, enabling it to have various AI functions depending on the software installed.

Features

Heat management and noise control are needed to run powerful GPU. AI inside Cube’s seamless construction, an integrated cooling device, and internal airflow control keep heat and noise levels low. Just as with cloud-based products, the AI inside Cube series features simple plug and play functionality. One key advantage is that under the subscription-based model, there is no need to choose or purchase hardware, or go through processes such as systems integration, customization, AI learning, or pilot studies. Furthermore, it is designed so that customers control their data. Without sending data to the company, the customer can process using AI at any time while still receiving AI technology updates from the company with no particular limits.

Pricing and contract period

The company has not disclosed service pricing for AI inside Cube, but running DX Suite as an on-premises solution rather than a cloud service offers multiple advantages. Customers who use the on-premises solution pay only a fixed fee per month for unlimited use of DX Suite’s AI-OCR applications (up to 15mn processing requests possible per month for a non-mini AI inside Cube). The solution does not include a machine learning function and it is geared toward customers who digitize a large number of sheets/documents as a processing cost per document could be lower than using cloud version. In order to respond flexibly to fluctuations in demand, AI inside Cube is provided under fixed-term and open-ended contracts. Fixed-term contracts come under selling model revenue, while open-ended contracts are included in recurring revenue. The minimum contract length is one month. From FY03/22, the number of AI inside Cube contracts will be disclosed as Edge contracts.

Hardware reuse

Hardware is typically depreciated over five years, but due to rapid obsolescence of hardware along with the rapid advances in GPU semiconductor and other technologies, the company replaces it with a new model roughly every two years. It plans to recover older AI inside Cube units from customers when models are updated or replaced or when rental periods end, and reuse them as internal business servers after thoroughly removing personal data from the devices.

Specifications (as of May 2021)

GPU: General-purpose GPU from Nvidia Corp. (NASDAQ: NVDA)

AI inside Cube: External dimensions (height x width x depth) are 308 x 300 x 300mm, weight roughly 11kg, storage 1TB, processing capacity (single format without Multi Form function installed) roughly 25,000 items/hour.

AI inside Cube mini: External dimensions (height x width x depth) are 156 x 150 x 154mm, weight roughly 2.5kg, storage 512GB, processing capacity (single format without Multi Form function installed) roughly 7,200 items/hour.

Learning Center (developing AI)

From April 2021, the company started providing Learning Center, which enables the development of highly accurate AI models using no code.

Features

Traditionally, the AI development flow required support from AI vendors or developers to assist with the process from planning, data collection, preparation for learning, R&D, learning, evaluation, and systems integration. Using AI vendors or developers to install AI requires communication regarding workflow issues and workplace expertise, in addition to research into systems configuration and pilot studies, involving long time periods. For these reasons it was not easy to install and use AI. Learning Center has a Japanese language UI that is intuitive and easy to operate. Anyone can easily develop highly accurate AI models without coding. After planning and data collection, with simple inputs and a few clicks, Learning Center can manage the data and automatically generate and deliver AI for users without programming knowledge. It is also possible to increase accuracy with additional training from the customer and link with other systems. The company aims to enable low-cost, swift AI adoption and operations at the company, division, or individual level without relying on AI vendors or developers.

Examples of Learning Center in use (company envisioned scenarios)

The company envisages the system being used to read texts on uniquely formatted forms (non-standard form AI-OCR), visual inspection (facilities maintenance), quality control (inspection and detection of faults), safety management (detection of abnormalities), and support for image diagnosis in healthcare settings. The company expects existing DX Suite users to be the first customers.

Case 1: At waste treatment facilities, it is necessary to sort and remove any hazardous materials mixed in with materials not suitable for incineration during the crushing and processing stage. A user developed AI to automate the process of hazardous material detection using Learning Center, with the finished AI installed in AI inside Cube.

Case 2: Waiting for elevators in factories can delay logistics activities. A customer employed Learning Center to develop AI to understand and analyze the operational status in real time.

Case 3: A customer developed AI to inspect the appearance of holograms on manufactured IC cards at the time of shipment. In just 30 days, manufacturing site staff used Learning Center to plan, prepare the training dataset, develop AI, and verify accuracy. Despite the small dataset used for development (about 100 images), it has achieved an accuracy of roughly 90%, and will continue to improve.

Pricing

There are no charges from account issuance through training data creation. Once the model has been designed, there are set fees for training and inference only in months when the model is used. Charges for training (making) are JPY100,000 per month and for inference (using) are JPY30,000 per month (depending on use of the AI, charges may vary). Support services at the time of installation incur separate charges.

Future development

The company plans to set up a marketplace (similar to Apple’s App Store) where customers can share and sell AI created through Learning Center as well as obtain or purchase AI produced by other users. The company expects that the marketplace will serve as a forum where customers can earn revenue from AI they have created themselves and users can obtain AI while constraining development resources and costs.

Sales channels

After enhancing product quality sufficiently using a direct sales model, from around FY03/19 the company started to make full use of external sales channels such as distributors, OEM sales, and product partners. This was aimed at avoiding a labor-intensive sales organization and major advertising expenses, while enabling swift, efficient growth in user numbers. In FY03/21, the breakdown of recurring revenue by sales channel was direct sales 31.1% (JPY1.3bn), partner sales 18.1% (JPY730mn), and OEM sales 50.8% (JPY2.0bn). The use of outside channels had a significant impact. However, as discussed below, there are some issues with using external sales channels. In April 2021, major customer Nippon Telegraph and Telephone West Corporation (NTT West) notified the company of its intention not to renew most of its licenses.

The company started a partner sales system in December 2017. As of FY03/21 the company had concluded alliances with 99 companies, including NTT Data, Hitachi Systems, NEC Networks & System Integration (TSE1: 1973), Dai Nippon Printing (TSE1: 7912), Ricoh Japan (a subsidiary of Ricoh; TSE1: 7752), and Pasona (TSE1: 2168). Its OEM partners are NTT East (product name: “AI Yomitoru,” from January 2019), NTT West (“Omakase AI-OCR,” from December 2019), and NTT Data (“NANATSUAI-OCR,” from December 2019). Product-centered partnerships, including in multifunctional printers and copiers, as well as RPA, include Fujifilm Holdings (TSE1: 4901) subsidiary Fuji Xerox, and RPS Holdings (TSE1: 6572) subsidiary RPA Technologies.

While the number of sales partners is growing, they differ in size and services handled, so growth in their numbers is not directly correlated with future growth in user count. In order to utilize its network of sales partners, the company aims to increase the floor on sales for all sales partners, over 100 companies, not just specific OEM partners. In terms of tangible initiatives, in an effort to shift from direct sales by internal personnel to partner sales, the company is proactively onboarding staff to transfer its sales know-how to sales partners so that they can effectively use materials and demos of company products in their sales activities. The company states that some time will be needed to see the impact of these onboarding efforts for sales partners be reflects in higher contracts through partners.

In direct sales, as of FY03/22 the company sold the Standard/Pro DX Suite plans mainly to large corporations. It was also concentrating on Learning Center proposals.

Major customer does not renew licenses (factor in higher FY03/22 churn rate)

On April 28, 2021, AI inside announced that it had received notification from OEM partner Nippon Telegraph and Telephone West Corporation (NTT West) that NTT West did not intend to renew the unused portion of licenses for AI-OCR services the company provides to NTT West when they expired, in line with the partnership agreement. NTT West’s sales approach of getting end users to try the service before encouraging them to move on to full adoption succeeded in generating a sharp increase in contract numbers in a short time span in FY03/21. However, NTT West was not well-equipped to handle the sudden increase in customer numbers and the migration from trial to full adoption did not proceed as expected, leading to expectations of large-scale contract cancellations in FY03/22.

Impact on FY03/22 results

In line with the announcement released on April 28, 2021, the company expects many of the DX Suite Lite plan licenses issued to OEM partner NTT West will not be renewed in FY03/22. As such, the company in FY03/22 does not expect to book the JPY1.8bn portion of the JPY2.1bn in revenue from NTT West DX Suite Lite plan licenses booked in FY03/21.

Risk of other customers cancelling contracts

The average monthly churn rate in FY03/21 for the Standard/Pro churn rate was 0.55%, while the churn rate for OEM NTT West’s “Omakase AI OCR” was 1.54%. The churn rate excluding “Omakase AI OCR” was 0.47%. Because the Standard/Pro churn rate was 0.55%, and that excluding “Omakase AI OCR” was much the same at 0.47%, we estimate that in FY03/21 the churn rate for Lite plan excluding “Omakase AI OCR” was little different than that for Standard/Pro plans. The company said that churn rates for OEM sales to NTT East Japan and NTT Data were largely in line with its own direct sales. It explained that the mass cancellations of OEM contracts for NTT West did not reflect problems with the Lite plan or the OEM structure overall, but was confined to a specific OEM customer. The company has not provided detailed explanations for the mass cancellations of OEM contracts for NTT West, citing contractual confidentiality obligations. However, AI inside plans to work with NTT West to strengthen its onboarding arrangements and provide it with access to its customer and product support expertise. This suggests that the company sees the problems as lying with onboarding (installation support to promote ongoing use) and support.

Background

Overview of partnership agreement

On December 1, 2019, AI inside entered a partnership agreement with NTT West, enabling NTT West to offer the company’s DX Suite Lite plans as what NTT West calls “Omakase AI OCR” service to businesses (users) that the company itself provides services to. “Omakase AI OCR” is an OEM product of the DX Suite Lite plans AI inside provides to NTT West. The company grants NTT West licenses to use the DX Suite Lite plans, and NTT West provides “Omakase AI OCR” to users via a sublicense. Under the agreement, AI inside receives license fees from NTT West in line with the number of licenses it has provided to NTT West, regardless of how many NTT West users are provided with “Omakase AI OCR.” The company said that NTT West did not acquire licenses in bulk upfront in the hope of sublicensing them to users, but issued them in response to user applications. From an accounting perspective, AI inside recognizes revenue from licensing fees according to the number of DX Suite licenses it has granted to NTT West. The minimum contract period for the company’s DX Suite Lite plans is one year from the contract date.

Overview of notification

As of the date of the notification, NTT West continued to make its best efforts to entrench service use among existing users, and the company also lent support as the OEM provider. However, NTT West has been unable to make proposals more closely tailored to users’ needs, and many have cancelled their “Omakase AI OCR” licenses before the end of the minimum one-year period. In response, NTT West notified AI inside that it would not renew the unused portion of DX Suite Lite plan licenses after expiration, and did not expect to renew 7,636 of the 9,284 licenses it held (as of end-March 2021) in May and June 2021.

Timing of issue awareness

The company said it realized that there were issues with OEM sales to NTT West with the increase in cancellations in January–March 2021 (Q4 FY03/21). OEM products are provided under the partner’s brand, so the company was not involved in sales or support activities. The company said it was possible to get an idea from the system that usage rates were low for NTT West’s OEM product, but migration rates to the main contract for NTT East’s and NTT Data’s OEM offerings were the same as direct sales (over 90%). It was difficult to predict in advance that the migration rate to the main contract would be lower for NTT West. The company started OEM sales to NTT West in December 2019, with sales getting fully underway in January–March 2020 (Q4 FY03/20). Because the minimum contract period was one year, the company was not clearly aware of the low migration rate until the increase in cancellations in Q4 FY03/21.

Future of partnership between the company and NTT West

The notification was not intended to dissolve the partnership between NTT West and the company, but to inform AI inside that NTT West did not intend to renew the unused portion of existing DX Suite Lite plan licenses when they reached their one-year expiry date. AI inside plans to work with NTT West to strengthen its onboarding arrangements to provide users with better support on how to best use the product and to provide NTT West with access to its customer and product support expertise to promote ongoing use.

Key performance indicators (KPI)

KPIs for DX Suite

As of FY03/21, DX Suite was the company’s mainstay product. AI inside uses a software as a service (SaaS) business model for DX Suite. It discloses two categories of revenue: selling model and recurring. Selling model revenue is booked for each specific transaction (initial costs or fixed-term contract). Recurring revenue is recognized continuously as long as the customer keeps using the service, and comprises pay-as-you-use costs based on the number of requests (readings) to convert forms into data by the customer and monthly fixed charges based on basic functions and options. 

The share of Lite plans (no initial costs) in the sales mix is increasing, and in FY03/21, recurring revenue comprised 87.7% of the total (47.1% in FY03/20). Annual recurring revenue (ARR) can be broken into number of users x ARR per user (ARRPU). As of end-FY03/21, there were 9,596 Lite plan users (+8,423 YoY), 1,034 Standard/Pro users (+334), and 87 AI inside Cube (Note: From FY03/2022, the number of Cube contracts will be disclosed as Edge) users (+54). The churn rate (average lost monthly usage fees over the past 12 months due to contract cancellations as a percentage of monthly usage fees for existing users) was 1.14%.

Furthermore, from FY03/2022, the company plans to strengthen its efforts to focus on large customers. Large customers make it possible to increase earnings in various ways as the company can increase the number of users per company (user/account number) rather than just increase the number of contracted companies. It is also possible to increase the use volume (number of requests) per user. While increasing the number of contracts will continue to be important, the importance of indicators such as number of users per company and average number of requests per month is on the rise.

Contract count

Prices are significantly different for Standard/Pro plans aimed at large companies and Lite plans for SMEs, so contract growth rates vary substantially. In the OEM channel, what is counted is the number of contracts between the company and OEM partners. Cumulative contract numbers are disclosed on a quarterly basis. However, while net growth can be estimated from differences in the cumulative figures, it is impossible to know the breakdown between new contracts and contract cancellations from an external perspective.

ARRPU (ARR Per User [annual recurring revenue per user])

Prices are significantly different for Standard/Pro plans for large companies and Lite plans for SMEs, so ARRPU levels depend heavily on the overall sales mix. It is difficult to forecast future revenue based on past ARRPU when a significant change in sales mix is expected. The company has not disclosed ARRPU since FY03/21.

Churn (cancellation) rate

Churn rate is defined as average lost monthly usage fees over the past 12 months due to contract cancellations as a percentage of monthly usage fees for existing users, and is disclosed on a quarterly basis. It is important to note that it is a lagging indicator based on past averages, so the disclosed churn rate will not increase suddenly if there is a spike in cancellations. Further, the churn rate is disclosed monthly, so estimating the annual rate requires a calculation of (1-estimated monthly churn rate) to the power of 12. The churn rate the company discloses is in principle a mix of its products overall, but it does disclose churn rates for specific customers/products at times. 

Number of requests 

The company discloses the total and monthly average number of requests for DX Suite (cloud version). The number of requests is an indicator of how well the company’s product are being used. It does not break down free use and paid request numbers and sometimes only discloses data in graphical form, which means only a rough estimate is possible. The number of requests relating to the on-premises AI inside Cube is not included.

The company discloses the average number of requests per month as a reference indicator to show the status of pay-as-you-use services. In the future, if the marketplace develops and third-party products increase, the importance of the number of average requests will increase as it will rise by the multiplication of the number of contracts, number of users, and number of applications.

User/account number (number of users per company)

In the future the company states that this will become an indicator for the expansion of the number of large customer contracts and users within large customers, areas where the company intends to focus.

Market and value chain

Japanese OCR market

Growing at a high CAGR of 28.5% (from FY2017 to FY2023)

According to research by ITR Corporation, total revenue in the Japanese OCR market reached JPY5.2bn in FY2018, marking a sharp 44% YoY increase. ITR estimates the market expanded another 54% YoY in FY2019. Although OCR solutions have been available for a long time, their adoption by companies has lagged mainly due to their low recognition accuracy. However, AI-driven advances in accuracy and the addition of automatic data extraction functions have led companies to reconsider OCR solutions, and this has fueled the sharp market growth observed in recent years. In addition, a growing number of companies are taking a fresh look at OCR, while considering labor-saving solutions for data input work or further expansion of automation applications when adopting robotic process automation (RPA) systems. Therefore, ITR sees the Japanese OCR market expanding to JPY16.2bn in FY2023, with a high CAGR of 28.5% (FY2017–2023).

Japanese SaaS market growing at CAGR of 90.6% (FY2017–23), driving expansion in Japanese OCR market

Looking at the Japanese OCR market by software delivery method, we observe that the software as a service (SaaS, software delivered through the cloud) market expanded a sharp 5.5x YoY in FY2018, eclipsing the 21% YoY growth in the market for packaged software. The number of companies opting for SaaS solutions—which facilitates improvement of data accuracy through AI technology—is bound to rise further going forward, and ITR believes SaaS will account for roughly 60% of the Japanese OCR market in FY2023.

Rise of AI-OCR vendors

Among vendors, ABBYY continues to drive the market by offering AI-OCR solutions with multi-language support through a wide network of OEM partners (mainly RPA vendors). At the same time, startups such as AI inside and Cogent Labs are enjoying sharp growth (AI inside revenue has shot up 10x in FY03/17–18) by enhancing their handwritten text recognition rate through AI and offering additional functions such as automatic extraction of items to be read. Other well-established OCR vendors also continue to actively enhance their software functionality while strengthening their sales organization.

Affinity with Japanese RPA domestic market

AI-OCR has a strong affinity with RPA. As of end-March 2020, roughly 10,000 Japanese companies had adopted an RPA solution, and, according to AI inside, half of these have expressed interest in using AI-OCR services to the company. The company thinks that companies that have adopted RPA solutions (mainly large corporations) are promising targets for DX Suite Standard and Pro plans. The company said that it thought there was little overlap between the domestic RPA market and its Lite plan offering, whose main presumed customers are SMEs.

The Yano Research Institute expects the Japanese RPA market (total for services and tools) to expand at a CAGR of 35.2% (FY2017–FY2022), while ITR projects the Japanese SaaS OCR market will grow at a CAGR of 90.6% (FY2017–FY2023). In FY2017, the gap in CAGR was only 1.1%, but the difference is expected to widen more than 9x, reaching 10% in FY2022. This outlook suggests that RPA users are eager to adopt SaaS OCR solutions.

AI-OCR market shares

The company has 64% of the JPY4.2bn domestic AI-OCR market share (FY2019 estimates of AI-OCR solutions vendor sales and market shares [cloud] in MIC Research Institute’s OCR Solutions Market Survey–Expanding with AI-OCR, 2020 Edition, February 2020). According to Fuji Chimera Research Institute, the Japanese OCR market (note: not just AI) will expand roughly from JPY16bn in FY2020 to JPY30.3bn in FY2025, and ITR sees the same market growing from JPY3.6bn in 2017 to JPY16.2bn in 2023.

Japanese BPO and AI markets

Non-IT BPO market

The Yano Research Institute projects the Japanese business process outsourcing (BPO) market will expand from JPY4.35tn in FY2019 to JPY4.75tn in FY2024 (CAGR of 2.0%). Within the BPO market, the non-IT BPO segment (which includes data input operations) was worth JPY1.77tn in FY2019, and AI inside estimates simple work tasks in core operations (which are increasingly being performed by AI and RPA) accounted for just over one-third, or JPY0.58tn, of this segment. At the same time, ITR expects the Japanese AI market to grow at a CAGR of 31% from FY2017 to FY2023, with the share for language analysis (includes AI-OCR) rising from roughly 25% in FY2017 to roughly 35% in FY2023.

Main competitors

AI-OCR involves using AI to read with a high degree of accuracy, but the basic function of reading text on paper is the same as traditional OCR. Accordingly, AI-OCR competes with OCR. It also competes with BPO services whereby people check data written on paper and input it once again, as well as potential customers’ internal workflows (including small companies that use inexpensive part-time personnel for these tasks). Potential customers will decide which method to use after considering cost and accuracy. BPO services could conceivably use lower-priced overseas labor, but checking Japanese-language information requires personnel familiar with the language, so cost benefits are not as easy to find as in the English-speaking world.

Comparison of companies offering AI-OCR and AI-related services

AI inside competes with four unlisted companies in the Japanese AI-OCR market: LINE, Net Smile, Cogent Labs, and Cinnamon. These companies were similarly founded around 2015. Research projects and activities involving deep learning have picked up steam since 2012, when research findings revealed through an international research competition^* prompted a global recognition of the effectiveness of deep learning based on image processing. AI-related startups such as PKSHA Technology (TSE Mothers: 3993) and Preferred Networks were also established around this time. As AI-OCR is projected to become the core segment of the OCR field (according to ITR), we have omitted non-AI-OCR Japanese companies from our comparison (in overseas countries with low labor costs, AI inside also competes with companies that digitize handwritten text manually).

AI-OCR competitors (listed companies)

Listed companies (other than AI inside) that engage in AI-OCR operations include IRRC Corporation (TSE Mothers: 7325) and EduLab, Inc. (TSE Mothers: 4427). They both have positioned AI-OCR as a new business, and accordingly, their AI-OCR revenue remains low when compared to AI inside, which is a leading AI-OCR specialist. However, these companies are all promoting AI-OCR through their core businesses (insurance, educational testing services, and social networking, respectively). IRR Corporation has adopted handwritten text recognition technology from AI-OCR specialist Cogent Labs.

IRRC Corporation (TSE Mothers: 7325)

IRRC Corporation has internally developed a one-stop insurance analysis and search system, and it operates walk-in insurance shops under the Insurance Clinic brand across the country. Its wholly owned subsidiary Infordio offers the Smart OCR technology to customers operating in the insurance and other industries. Smart OCR earned revenue of JPY302mn in FY06/21. On September 24, 2019, Infordio concluded a partnership agreement with Cogent Labs, developer of the Tegaki AI-OCR solution. Infordio has been mainly working on development of text recognition tools for structured and semi-structured sheets/documents and handwritten sheets/documents.

EduLab, Inc. (TSE Mothers: 4427)

EduLab has two core businesses: e-Testing and e-Learning (roughly 60% of revenue), and exam operation and subcontracting (roughly 40%). Its services include (1) the development of technologies for measuring academic ability and testing methods, (2) the development, operation, and analysis of ability tests, and (3) the provision of educational services. In addition, EduLab provides EdTech solutions and platforms to support development of next-generation education, and develops new businesses in the EdTech field. Its AI-OCR DEEP READ solution had revenue of JPY552mn in FY09/20. DEEP READ has been adopted by companies such as ORIX Life Insurance Corporation.

Overseas, EduLab has stepped up operations in the enterprise contents management (ECM) field in the US by entering into a capital and business alliance with Ephesoft Inc., which plans to replace the recognition technology for printed and handwritten text used in its next-generation cloud-based ECM products with DEEP READ. EduLab expects the global ECM market to grow to USD80.0bn in 2023, and it sees the global OCR market for ECM products expanding to about JPY260.0bn. It has also successfully developed a speech recognition engine with support for non-native pronunciation and an engine that automatically generates test questions.

AI-OCR competitors (non-listed companies)

LINE Corporation (not listed): Scheduled to merge with Z Holdings (TSE1: 4689)

LINE BRAIN OCR has built a reputation as a highly accurate text recognition solution that can read not only horizontally and vertically written text but also curved and slanted text, which presents a greater challenge. It comes with multi-language support and can also recognize specialized terminology. It was voted the No. 1 OCR solution in four fields by the International Conference on Document Analysis and Recognition (ICDAR; as of March 29, 2019). When used to process structured sheets/documents (order forms, invoices, surveys, and other sheets/documents with specified layouts), users can specify the document layout and sections to be read in advance. Multiple document layouts can be registered, and the software will automatically distinguish between the different types of sheets/documents.

When used to process semi-structured sheets/documents, the software can read and digitize words and sentences embedded in images without prior configuration of settings. It can process various types of text, including vertical, slanted, and curved text. It can detect text from photographs, which can be useful to tag photographs that need to be archived. Based on an electronic Know Your Customer (eKYC) service (tentative name) that combines LINE BRAIN OCR and other AI technology, LINE users can send a photograph of their ID and a selfie to have their identity confirmed on official LINE accounts.

Cogent Labs Inc.

Cogent Labs sells the specialized AI-OCR solution Tegaki. According to Cogent Labs, Tegaki boasts a 99.22% recognition rate for handwritten text (based on in-house research results), and it can therefore be leveraged to optimize data input operations and reduce costs. The service uses proprietary algorithms that combine OCR and deep learning technology to recognize various types of handwritten text including Japanese syllabary characters (hiragana and katakana), Japanese logographic characters (kanji), numbers, alphabetic letters, and symbols. In addition to handwritten text, the service can also process checkboxes and circled characters, saving customers the work of having to sort and categorize sheets/documents by type before they can be processed with OCR. There are many types of handwritten sheets/documents, with some using multiple-choice checkboxes and others containing separate boxes for individual written characters, and this has traditionally been an obstacle for OCR. Tegaki can also use training data to learn to process handwritten text in different languages. Cogent Lab says Tegaki can also be linked to industry-specific linguistic models to improve text recognition accuracy.

Net Smile, Inc.

Net Smile was founded by a graduate of the Faculty of Engineering at the University of Tokyo, who also serves as the president of the company. Its headquarters are situated in front of the University of Tokyo. Its main staff and engineers are mostly foreign nationals. Net Smile sells the AI Scan Robo solution, which is geared toward scanning small volumes of sheets/documents with a large variety, in combination with RPA.

Business description:

Development of, and support for, AI and machine learning simulation and programming

Focus on RPA, business automation, natural language processing, automatic dialog systems, AI for investment

Development of, and support for, computer simulation programs

Development, and support for, mathematical science, technical computing, data analysis software

Cinnamon, Inc.

Cinnamon has offices in Tokyo, Silicon Valley, Hanoi, Ho Chi Minh City, and Taipei. It generates roughly 90% of its revenue in Japan, mainly from sales of the AI-OCR solution Flax Scanner and the speech recognition tool AI solution Rossa Voice.

Overseas company with Japanese subsidiary

ABBYY group

ABBYY is a leading global developer of document recognition, content capture, and language-based technology and solutions. The ABBYY group has head offices in North America (Milpitas, California), Western Europe (Munich), Eastern Europe (Kiev), and Russia (Moscow). It has over 1,000 full-time employees, with the majority consisting of programmers, engineers, and linguists. The group operates 16 offices across the world. In Japan, it offers services through subsidiary ABBYY Japan Co. Ltd.

The group leverages OCR and other technologies that optimize information usage on a global scale, and has developed cutting-edge software solutions that can seamlessly convert sheets/documents with complex layouts (including PDFs) into digital text. ABBYY’s AI-OCR services support 198 languages, and can accurately read sheets/documents with multiple languages by virtue of its high-precision recognition and conversion functionality.

Benchmark companies in AI and IT industries

Amazon: Amazon Web Service (AWS), Microsoft

As the business model he envisions for AI inside in the future, CEO Toguchi points to the Amazon-operated AWS (world’s largest cloud platform at present) and Microsoft’s software business. Both companies have established platform businesses that have become the de-facto standard in their respective fields. AI inside has focused on OEM business, and its users require OEM partners to include the “DX Suite” brand name in the products they sell (a point of difference from rival Cogent Labs). Under its AI inside X platform strategy, the company looks to diversify AI applications that can be generated through deep learning and promote the use of AI-controlled OS’s in an effort to realize a growth scenario driven by scalability and a beneficial synergistic cycle.

Preferred Networks

AI inside regards Preferred Networks (PFN), a Japanese pioneer in AI solutions, as a benchmark company. Preferred Networks was founded in March 2014 as a venture company, and it has developed control technologies based on deep learning. It is the most valuable private company in Japan, and continues to build partnerships with (and attract investment from) companies such as Toyota Motor (TSE1: 7203), Hitachi (TSE1: 6501), Fanuc (TSE1: 6954), Chugai Pharmaceutical Co., Ltd. (TSE1: 4519), DeNA (TSE1: 2432), Mitsui & Co. (TSE1: 8031), and Tokyo Electron (TSE1: 8035). PFN excels in developing cutting-edge technology that combines deep learning with profound expertise in specialist fields.

Preferred Networks is engaged in a wide range of specialized areas from development of Chainer, its core deep learning framework, to the construction of large-scale computation clusters and projects in the robotics or life sciences fields. It has collaborated with Fanuc to enhance industrial robots and machine tools since 2015, and has worked with JXTG Holdings to optimize and automate processes in large-scale oil refinery plants since 2019. In a press release issued in December 2019, PFN announced its intention to shift its focus from development of software (Chainer) to hardware (deep learning processor MN-Core).

Strengths and weaknesses

Strengths

Entry into the AI-OCR market before competitors, enabling data collection and high accuracy

AI inside started offering an early version of its AI-powered handwritten text recognition service soon after it was established in August 2015, and developed DX Suite—its AI-OCR service for handwritten forms—into a full-fledged product and business by November 2017. The company’s customers were already using DX Suite when competitor Cogent Labs rolled out its AI-OCR solution Tegaki in August 2017. AI inside has secured access to large amounts of data by increasing corporate customers (mainly major companies), and this has allowed it to improve its text recognition accuracy. The company has not disclosed its recognition rate since it went public. However, when it raised funds on August 3, 2018, it revealed its technology was able to read a roughly 6,300 characters consisting of Japanese logographic characters (kanji; level 1 and 2) and two sets of Japanese syllabic characters (hiragana and katakana) with an average accuracy of 99.91%. AI inside has developed algorithms that can generate artificial images of handwritten text, and this has dramatically expanded the training datasets, which is a key factor behind its high text recognition accuracy. Its technologies can not only read handwritten text, but also ignore characters and dotted lines that are not intended to be read. The company had obtained six patents related to text recognition technology as of end-FY03/20.

Privacy control function serving as a barrier to entry

AI inside is a frontrunner in text recognition accuracy, and employs privacy control functions (the default position is that its AI does not learn the data unless a user explicitly sets it to ON). In addition to gaining customers whose major concern is data handling, it is difficult for competitors to gather enough data to enable them to close the gap in text recognition accuracy with the company’s AI-OCR offering. This strategy puts the company in a good position to maintain its competitive advantage.

Expansion of sales channels through external partners

After enhancing product quality sufficiently using a direct sales model, from around FY03/19 the company started to make full use of external sales channels such as distributors, OEM sales, and product partners. This was aimed at avoiding a labor-intensive sales organization and major advertising expenses, while enabling swift, efficient growth in user numbers. In FY03/21, the breakdown of recurring revenue by sales channel was direct sales 31.1% (JPY1.3bn), partner sales 18.1% (JPY730mn), and OEM sales 50.8% (JPY2.0bn). The use of outside channels had a significant impact. However, as mentioned in the Weaknesses section below, there are some issues with using external sales channels. In April 2021, major customer Nippon Telegraph and Telephone West Corporation (NTT West) informed the company that it did not intend to renew the majority of its licenses.

Weaknesses

Difficulty in detecting problems before churn rates rise due to OEM supply arrangements

OEM supply involves the OEM partner conducting sales and support activities itself under its own brand, so the company can increase contract numbers while curtailing fixed costs, unlike under the direct sales model. However, it has no control over onboarding or support. Hence, it is more difficult to become aware of any issues in these areas before churn rates actually start increasing than in a direct sales situation. In fact, as announced on April 28, 2021, the company expects many of the DX Suite Lite plan licenses issued to OEM partner NTT West will not be renewed in FY03/22. As such, the company in FY03/22 does not expect to book the JPY1.8bn portion of the JPY2.1bn in revenue from NTT West DX Suite Lite plan licenses booked in FY03/21. This will put upward pressure on the churn rate in FY03/22. The company has not provided detailed explanations for the mass cancellations of OEM contracts for NTT West, citing contractual confidentiality obligations. However, AI inside plans to work with NTT West to strengthen its onboarding arrangements and provide it with access to its customer and product support expertise. This suggests that the company sees the problems as lying with onboarding (installation support to promote ongoing use) and support.

Reliance on domestic AI-OCR for majority of revenue in short term 

As of FY03/21, the company derived most of its revenue from AI-OCR in the domestic market. It is working to expand the territories it operates in and support more languages as part of its global push. It is also expanding services to include FaceAuth (automatic ID services, only provided as an API as of May 2021), and Digital Input (enables digital data input, launch imminent as of May 2021). In April 2021 it released Learning Center, which enables the development of AI using no code, and plans to launch a marketplace which will allow users to sell their AI. However, it does not expect these to make a major contribution in FY03/22, and AI-OCR in Japan will remain the mainstay revenue source for now. 

The need to distinguish itself to avoid direct competition with massively capitalized global competitors in the AI field

The company sees its mission as “To bring AI to every being in the world and contribute to a richer future society.” It has set out a platform strategy to achieve this corporate mission. Moving forward, it aims to extend its business beyond DX Suite to deliver infrastructure as a service (IaaS) that can drive AI, including AI inside Cloud, AI inside Cube, and networks; AI inside Learning Center, a tool that will enable anyone to easily create AI; and a marketplace (platform as a service or PaaS) that will enable the selection and use of AI and apps created by users themselves, other users, or the company. As the company branches out from AI-OCR, there is a mounting possibility that its business territory will overlap with competitors’ in the broader AI marketplace. Competitors include numerous massively capitalized global players such as the so-called GAFA companies. This is undeniably unfavorable for the company given its earnings and balance sheet capacity as of FY03/21, which relies mainly on AI-OCR for revenue. Further, many AI companies use the same Nvidia general-purpose GPU that the company uses in its AI inside Cube hardware. Because AI performance depends on computing power such as the GPU, companies such as Google are developing hardware components in-house. However, AI inside does not have enough funding capacity or expertise to develop hardware on its own.

Historical performance and financial statements

Income statement