03 — AI Experience Design
We design whose problem it solves, where a human stays in the loop, and how it fits the existing work. Because it will be wrong sometimes, we design the evidence, the correction path and the handover to a person alongside it.
- Typical duration
- 2–4 months
- Indicative price
- From ¥600,000
Adding AI is not the hard part. The hard part is designing how a person notices it was wrong and takes over. Skip that and you ship a feature that is convenient and that nobody trusts.
Whose problem this solves, where a human makes the call, how it fits into the existing product or workflow — we design all of it, including how evidence is shown, how output gets corrected, and when it escalates to a person.
This tends to fit when
- You want to use AI but have not settled where it belongs
- You built a PoC but could not get it into the product or the workflow
- The team does not trust the output, so nobody uses it
- You want to rebuild a workflow around generative AI
Scope
- UX for AI products and AI features
- Conversational and generative UI design
- Workflow design for AI adoption
- Human-in-the-loop design
- Reliability, explanation and error design for AI output
- AI workflow and skills design
What you end up with
- Where AI applies, and in what order
- A map of what AI does and where a human decides
- Conversational / generative UI design and prototypes
- Behaviour on failure, evidence display and escalation paths
- Workflow diagrams and a migration plan
- Workflows and templates for day-to-day operation
How we work
- STEP 01
Pick the work and the problem
Not what AI can do, but whose effort or judgement it removes. Areas where the gain is small are taken out of scope.
- STEP 02
Draw the line
We separate what AI handles from what a person decides, using the cost of being wrong as the criterion.
- STEP 03
Design the experience
Input, output, how evidence is shown, how it gets corrected. We work against real output in a prototype.
- STEP 04
Put it into operation
We connect it to the existing workflow, plan the migration and set the operating rules, then adjust based on how it is actually used.
Duration and team
- Typical duration
- 2–4 months, depending on how much of the workflow is covered and how much validation is needed
- Team
- Led by one AI experience designer, with a design engineer handling prototype validation.
Indicative pricing
- Framing AI adoptionAssumed size：About 1 month. Where AI applies and in what orderFrom ¥600,000
- AI experience designAssumed size：2–4 months. Role split, UI design and prototype validationFrom ¥1,500,000
- Workflow and operationsAssumed size：Continuous. Embedding it in operations and improving the workflowFrom ¥500,000 / month
※ These are indicative. The final figure follows a quote based on the work covered, validation scope and team.
※ Model, API and infrastructure costs are not included. We estimate them during design.
Questions we get about this service
UX design for products with AI features, conversational UI, generative UI, the workflow around AI adoption, human review steps, how errors and uncertainty are communicated, and internal AI workflows.
We also design AI into the making itself — research synthesis, idea exploration, prototyping and engineering.
Yes. Rather than starting from the tool, we work through the problem to solve, the quality required, the users, the data and the operating conditions.
From there we define which parts AI should handle, which parts people should, and what that requires of the technology and the model.
Full automation isn't always the right answer. Decisions with serious consequences, judgements requiring context, and calls about ethics or brand often need a person involved.
We design the split between people and AI around quality, confidence, workload and explainability — not just the automation rate.
We assume it will. We design the supporting evidence, the ways to check it, how confidence is communicated, and the paths to regenerate, correct, undo or hand over to a person.
For work that really matters, human review is built into the process.
AI can make parts of research, prototyping, production and engineering more efficient. We don't simply pocket that time — we also use it to explore more options and test more often.
Pricing is based on the problem, the scope, the quality required, the timeline and the team — not on how much AI is used.
Yes. For work you repeat — framing problems, research, UI review, writing requirements, validation, generating documentation — we can design AI workflows, Skills, templates and guidelines.
The goal is something the whole team can reuse, not knowledge locked in one person's head.
We never put confidential or personal information into an external AI service without the client's permission.
We confirm in advance where AI is used, on what data, with what settings, and how retention and training are handled, and align it with your security policy.