B2B customer portal, regulated industry · 2026
An AI assistant that shows its sources
UX consulting for a customer companion: when to answer, how to cite, when to hand over to a human.
- AI UX
- Trust patterns
- Conversation design
- Glossary
- Prototyping
Toggle the patterns to compare a generic answer with one that cites its sources, states its certainty and offers a human.
Why is my invoice for September higher than in August?
Your September invoice is 12% higher mainly because the monthly price index in your contract rose. Consumption stayed nearly the same.
Certainty: high – based on your contract and invoice
Sources
- [1]Contract §4 Price index
- [2]Invoice 09/2026, line 3
- [3]Consumption Aug–Sep
Why this answer?
I compared both invoices line by line. The difference comes from line 3 (index price). Line 1 (consumption) changed by less than 1%.
Context
A customer portal for business clients wanted an AI companion that answers questions about contracts, invoices and market data.
Challenge
In a regulated domain a wrong answer is worse than no answer. Users needed to see where an answer comes from and how certain it is – without drowning in disclaimers.
Approach
- Mapped the questions customers really ask and sorted them into answer, guide or hand over.
- Defined trust patterns: inline sources, a plain-language certainty label, a visible path to a human.
- Wrote a domain glossary so the assistant and the UI use the same words.
- Prototyped the conversation states and tested them with internal users.
Outcome
- A shared target picture for product, legal and engineering.
- Reusable interaction patterns for answers with sources and for hand-over.
- A glossary that also feeds the assistant's prompts.
My role
UX consultant for the assistant concept and interaction patterns