Scenario plan · Jeton Express
An AI assistant must distinguish valve variants
A procurement-software team wants its assistant to answer questions about valve sizes, materials and connections reliably. The proposed data exercise connects each variant with its supporting document and marks questions that the evidence cannot yet answer.
These scenarios are plans for discussing a future purchase. They do not report completed customer orders.
Data for an AI procurement assistant
The buying situation
In this scenario, a valve family has similar product names across supplier sheets and catalogues, while dimensions, materials and connection types vary. The software team needs a sample dataset that keeps those variants distinct and makes missing or conflicting source evidence visible during testing.
Proposed approach
- Agree sample products and real buyer questions with the software team, including which variant fields and source permissions are necessary to support each expected answer.
- Organise product and variant records with their document references, retaining the source version and distinguishing a confirmed field from a supplier claim or unresolved conflict.
- Prepare test questions with expected evidence and explicit unanswered fields, then ask the team to review the dataset against its acceptance rules before any wider use.
Decision pack for approval
A proposed structured sample dataset, source map and answer-testing set. The software team would approve field accuracy and evidence coverage; building or deploying the assistant would require a separate scope.
Still to confirm
Source permissions, exact variant identity, current document versions and whether the agreed questions can be answered from the supplied evidence.