UX Synthesis Agent
Turns research findings into design recommendations.
- Industry
- Procure AI
- Department
- Design
- Team
- Product Experience
- Kind
- Investigator
- Approver
- Product Experience Manager
- Code
- DSN-02
It synthesises interviews, usability tests and surveys with requisitioners, approvers, buyers, AP clerks and suppliers into ranked insights that explain why they struggle, for example why requisitioners go off-catalogue or suppliers abandon onboarding, separating patterns from anecdotes. It turns each insight into a recommendation tied to its evidence and files them in the product backlog. **The problem it solves:** Why requisitioners, approvers, buyers, AP clerks or suppliers struggle: the research question, each behaviour or pain point observed, how many participants and clients showed it, where in the journey it happened and what they said.
How it works
6 steps, and a person signs off before step 6.
- 01Looks it up
Finds the record the request is about in the research repository: the raw notes, tagged observations, session transcripts and past insights for the study and journey named..
- 02Reads the request
Works out what is being asked and what a complete answer needs.
- 03Checks the facts
Checks figures, dates and rules against the source records, not from memory.
- 04Works it through
Weighs competing explanations against the evidence and shows which one holds up.
- 05Asks a person
Hands the work to Product Experience Manager with its reasoning before anything goes out.
- 06Acts
Once approved, it acts in the product backlog: it creates one recommendation item per insight, linked to its evidence in the research repository. and records what was done.
What it works with
Your knowledge, your systems, your people.
Knows
The research repository's past insights, the role personas and journey maps (request to pay, source to contract, supplier onboarding), the severity scale and the research ethics and anonymisation policy for client and supplier participants.
Reads from
the research repository: the raw notes, tagged observations, session transcripts and past insights for the study and journey named.
Acts in
the product backlog: it creates one recommendation item per insight, linked to its evidence in the research repository.
Hands back
An object with insights (a ranked list, each with insight, role, cause, cause_type (platform_design, client_configuration or policy), evidence (participant and client counts, method and anonymised quotes), severity, confidence and how_to_confirm), recommendations (a list, each with recommendation, insight, effort (S, M or L) and expected_outcome), anecdotes (a list) and open_questions (a list).
Asks a person when
A recommendation rests on fewer than 3 participants, an insight is rated critical, a recommendation touches a compliance control, or a quote may still identify a participant or client.
Settings
Your limits, set by your managers.
| Setting | Default | What it controls |
|---|---|---|
| Participants needed to call it a pattern | 3 | An observation seen in fewer participants is reported as an anecdote. |
| Client organisations needed for a pattern | 2 | A pattern seen at only one client may be that client's configuration, not the platform. |
| Most recommendations per study | 5 | Only the strongest are filed; the rest are listed as further ideas. |
| Strip identifying details from quotes | On | When on, names, client and supplier organisations, prices and contact details are removed from every quote. |
| Needs a second method before it is an insight | critical only | Insights at this severity backed by a single method are filed as hypotheses. |
Limits are settings, not code: a manager changes them without a developer.
Rules it always follows
- An observation is a pattern only when at least 3 participants from at least 2 client organisations showed it; otherwise it is an anecdote or a client configuration issue.
- Explain each pattern by its likely cause, separating platform design, client workflow configuration and procurement policy, rank the causes by evidence, and say what would confirm each.
- Rate each insight critical, major, minor or cosmetic by how often it occurs and how much it drives off-contract buying, match exceptions or onboarding drop-off.
- An insight rated critical only that rests on one method only is filed as a hypothesis with the study that would confirm it.
- File at most 5 recommendations, each tied to one insight and its evidence.
- When on is true, remove names, client and supplier organisations, prices and contact details from every quote.
- Never recommend removing an approval, screening or segregation-of-duties step to reduce friction; recommend making it clearer or faster instead.
See the UX Synthesis Agent agent on your own work.
We’ll run it on a sample of your work and show you what your approver would see.