Trust mapping
We observe where users hesitate, check an answer or stop, then investigate why.
- → First-session observation
- → Conversation and log review
- → Drop-off and hesitation mapping
- → A ranked list of trust gaps
01
AI Experience Design
AI products people can check, correct and control.
Trust depends on what people can check.
An AI answer can look convincing without giving someone enough information to act on it. Users need to know where it came from, what is uncertain and what the system will do next.
We work on these decisions around the model: its role, the evidence it presents, permissions and recovery when something goes wrong.
( What changes )
We design how your AI responds and acts, with clear limits, checkable sources and control over important actions.
( Design principles )
Clear about its limits
People should know what the AI can do, what it can't, and when to ask a person.
Checkable answers
An answer earns trust when people can check where it came from.
People stay in charge
The more an action matters, the more control stays with the user.
A way to recover
When an answer or action is wrong, people need a way to correct it or get help.
( Scope of work )
AI behaviour & voice
We define the AI's role, scope and voice, including how it communicates uncertainty.
- → Role and scope definition
- → Conversation design
- → Tone, including uncertainty
- → Rules for handing over to people
Confidence & control
We design source references, approval steps and correction tools around the consequences of each action.
- → Sources and decision explanations
- → Levels of automation
- → Confirmation before key actions
- → Undo and correction flows
Failure states
We plan for incorrect answers, unavailable tools and tasks that need a person.
- → Map of what can go wrong
- → Recovery flows
- → Escalation to a human
- → Prompts and evaluation test cases
( How it works )
- Step 1
Observe
We watch real people use the product, or the task it replaces, and note where trust breaks.
- Step 2
Define
We agree what the AI should do, should never do, and when a person takes over.
- Step 3
Prototype
We build working conversations and flows early, with your real data where possible.
- Step 4
Test & hand over
We test with users, then hand over behaviour rules, flows and test cases your team can keep using.
Selected work
From the founders' own practices, before and alongside MOMENT.
Automotive manufacturer · Italy
Relay
An AI assistant that collects supplier sustainability data through the inbox suppliers already use.
An interactive demo of the full loop, from request to verified score.
Manufacturing · Training platform
Fabri
A virtual clone of the plant, on desktop or phone, where new hires practise real jams, cleanups and emergency stops by talking to an AI guide.
Designed for practising procedures without interrupting production, and exploring changes in a virtual plant.
PropTech · 2026
Voice front desk
A voice agent for home sellers, plus a console that briefs the human agent before the visit.
A working end-to-end prototype, and 1st Prize at Hack The Peak 2026.
( Project deliverables )
- → AI behaviour and conversation guidelines
- → Permission, correction and recovery flows
- → Prototypes and evaluation test cases
( Built for )
AI-native startups and teams adding AI to an existing product.
( Who you'll work with )
Iuliia Vorobiova
Co-founder · Service Designer & Builder
AI & ML engineers and full-stack developers join when a project needs them. Meet the team
Next service
Digital Products