Developing and training AI-capable staff
What we cover
Designing and running training on Kanata, producing the materials, and tracking attendance and comprehension
Generative AI consulting
From introducing generative AI to making it stick
Our generative AI consulting helps companies that are rolling generative AI out internally. It covers five areas: developing and training AI-capable staff, implementation support for the major AI platforms, organizing and restructuring internal data, delegating work to AI agents, and generative AI application development.
Third Scope has had AI-powered service development at the center of its business since its founding in 2018. Since 2023, generative AI has accounted for the majority of our client projects, and we have delivered large- and mid-scale generative AI training and implementation programs for organizations including JR Group and TEPCO group companies.
The five areas we cover
Implementation support for major AI platforms
What we cover
Selecting between Claude, ChatGPT and others, designing permissions, drafting internal rules, and making usage visible
Organizing and restructuring internal data
What we cover
Taking stock of the internal documents AI reads, structuring them, and designing and building the RAG setup
Delegating work to AI agents
What we cover
Carving out the target work, designing and running the AI agents, and setting up who checks the output
Generative AI application development
What we cover
Designing and building generative AI applications around your work, including integration with existing systems
Developing and training AI-capable staff with Kanata
Our generative AI training runs on Kanata, our own SaaS product, as the platform for delivering materials and hands-on work. The prompts and materials participants create during training are saved to Kanata's project library, so they remain available on the job after the training ends.
When training is delivered on its own, participants return to their desks and neither the materials nor the prompts they wrote survive. We avoid that gap between training and daily work by leaving the output of the training directly in the platform people use for real work.
An example training programme
All employees
What it covers
The basics of generative AI, where it fits into daily work, and how to judge what information must not be entered
How it runs on Kanata
E-learning plus a comprehension test
Departmental champions
What it covers
Prompt design for the work of each department, and turning AI settings into reusable templates
How it runs on Kanata
Hands-on work, with prompts registered as content types
Executives and managers
What it covers
How to judge the investment, where the risks sit, and how to roll it out to departments
How it runs on Kanata
Group sessions
IT department
What it covers
Permission design, reviewing logs, and managing non-human identities and accounts
How it runs on Kanata
Hands-on work, producing the operating procedures
Kanata's features and pricing are covered on the Kanata product page and on the Kanata service site .
Implementation support for Claude, ChatGPT and other major platforms
We support the implementation of third-party generative AI platforms, including Claude, ChatGPT, and Microsoft 365 Copilot. Adopting Kanata is not a prerequisite. For companies that already hold a contract for a platform, we keep that environment in place and identify why usage is not taking hold.
Implementation support covers four things: selecting the platform that fits the intended use, designing permissions per department, drafting internal usage rules, and making usage visible. Where individual AI subscriptions are already in use department by department, we start by taking stock of actual usage and then produce a plan for migrating to an officially provided environment.
For model selection, we design how models are used for different purposes. Everyday questions go to a fast, low-cost model and important analysis to a high-accuracy one; assigning models per task keeps both cost and quality under control.
Organizing and restructuring internal information and databases for AI
Whether generative AI answers internal questions correctly is decided by the state of the internal documents it reads. We handle taking stock of internal information on the assumption that AI will read it, structuring it, and building the retrieval layer.
The work covers identifying the documents in scope, clearing out duplicates and superseded versions, converting them into a form AI can interpret, designing and implementing the retrieval layer (the RAG setup), and designing how updates are handled. We put this step first because generative AI introduced on top of scattered documents will answer using old versions and incorrect information as its basis.
Database design and API design are handled by the system development team, working inside the same project as the generative AI design.
Delegating work to AI agents
With this service, Third Scope designs and operates the AI agents and processes your routine work on an ongoing monthly basis. You do not need to build a team to construct and run AI agents yourself.
Three things are decided before we start: the scope of the work in question, who checks the results, and how exceptions are reported. Fixing these three first prevents the accident of AI output flowing straight into your business unchecked.
Third Scope has built generative AI applications across three locations — Japan, the UK, and Singapore — and designs agents with a defined role for each piece of work.
Generative AI application development
We build generative AI applications on a contract basis, shaped around the work they support. The areas we cover are AI chatbots, AI content generation, image recognition and expression analysis, predictive analytics models, and proprietary LLM development.
Development areas and selected work
AI chatbots
What it covers
Conversational AI systems that handle customer contact and internal enquiries
Related work
An AI chatbot for tax accountants
AI content generation
What it covers
Engines that generate articles and posts automatically
Related work
Autoland, our trend analysis AI and article generation engine
Image recognition and analysis
What it covers
AI systems for face and facial-expression recognition
Related work
mimik, sales role-play using expression analysis AI
Predictive analytics models
What it covers
Support for demand and revenue forecasting
Related work
A management analytics application
Beyond these, we have built applications with quite different requirements by industry, including AI chatbots for professional practices, AI applications for older users, and AI for real estate management support. Our work is listed in our case studies .
Implementation is handled by our system development team. How requirements definition, database design, API design, and post-launch maintenance and operation work is covered on the system development page.
How a generative AI consulting engagement runs
The five steps from introduction to adoption
-
Understand the current state
What we do
Check how generative AI is actually being used, whether individual subscriptions exist, and how attitudes differ by department
-
Select the work to start with
What we do
Start with work where the effect can be measured and where failure has limited impact
-
Prepare the environment
What we do
Select the platform and design permissions while organizing internal documents in parallel
-
Training and hands-on work
What we do
People in the target departments create prompts and materials and register them in the library
-
Support adoption
What we do
Review usage monthly, find out why unused features are unused, and improve
Steps 1 to 3 can be commissioned on their own over two to four weeks. The deliverables of that stage are a report on actual usage and a prioritized list of the work to target.
Frequently asked questions about generative AI consulting
Can we commission the training on its own?
Yes. Adopting Kanata is not a prerequisite, and we also run training based on the generative AI environment you already use. The number of participants, how many sessions are held, and whether they are online or in person are designed case by case.
We have already deployed ChatGPT or Copilot. Can we still work with you?
Yes. We work with companies that have already deployed ChatGPT or Microsoft 365 Copilot. We keep the existing environment in place and identify why usage is not increasing, organize internal data, and draft usage rules for each department.
Is it safe to let AI read confidential internal information?
When confidential internal information is involved, we operate with settings that limit what can be referenced. In Kanata, the scope of learning data can be set to one of three levels — provided information only, supplemented with general knowledge (internal information takes precedence), or public information may also be referenced (sources required) — and information that cannot leave the company is set to provided information only.
What does delegating work to AI agents mean in practice?
It means Third Scope designs and operates the AI agents and processes your routine work on an ongoing monthly basis. Operation begins only after the scope of the work, who checks the results, and how exceptions are reported have been agreed in advance.
How is the effect of generative AI measured?
We measure three things: the usage rate, the time taken by the work in scope, and how often output has to be reworked. We measure the current values before deployment and compare them with the same indicators afterwards.
