Analyze supplier performance with AI

Fix the contract question and review period

Start with the decision, measure and time boundary rather than a generic request to summarize supplier performance.

The fictional contract covers document-processing support for one business unit. The quarterly review packet contains a contract schedule, a supplier scorecard, 48 delivery records, three incident records and two meeting notes. The analyst is asked to explain whether the reported service measures reconcile and what evidence needs owner review. The exercise does not authorize a renewal, payment deduction or finding of breach.

Write the review period as 1 April through 30 June and list the measures exactly as the contract schedule defines them. In this case, on-time delivery means accepted by the agreed date, not merely submitted. Incident response means the first recorded response within four business hours. These definitions are synthetic features of the exercise, not recommended terms for another contract.

Ask AI to create a source index and identify missing fields, then compare its extraction with the original schedule. Keep the contract owner, data owner and decision owner separate. If the schedule is ambiguous, record the ambiguity instead of letting the model select whichever interpretation produces a cleaner score.

Normalize the scorecard definitions

A percentage is comparable only when its numerator, denominator, status rule and period match the controlled definition.

The supplier scorecard reports 46 of 48 deliveries on time, or 95.8 percent. The underlying export labels two late records as on time because they were submitted by the due date, although acceptance occurred one day later. Recalculate the contract-defined measure as 44 accepted on time out of 48, or 91.7 percent. Show both the reported and recalculated figures with their definitions.

Do not describe the difference as misconduct. The packet establishes a definition mismatch, not why it occurred. Create a normalization table with measure, contractual definition, supplier calculation, analyst calculation, variance, evidence reference and question for the owner. Round only after the numerator and denominator have been preserved.

Use AI to flag rows whose statuses depend on multiple timestamps, but reperform the count from the delivery records. A fluent explanation cannot replace the calculation. If one record lacks an acceptance timestamp, move it to an unresolved row and show how including or excluding it changes the percentage.

Sources: [2]

Reconcile incidents to response evidence

An incident measure needs one event identity and a traceable clock before response performance can be calculated.

The scorecard lists two incidents and reports both within target. The incident register contains three identifiers, but meeting notes say the third was a duplicate notification of the second event. Match the records using incident time, affected service and reference history. Do not delete the third row merely because the note says duplicate.

Create an incident lineage table. Two rows share the same underlying event and remain linked, while the third is a separate event. One first-response timestamp is missing from the register but appears in an email reference quoted in the meeting notes. Treat that reference as evidence to retrieve, not the response timestamp itself, because the underlying email is absent from the packet.

The resulting measure has one confirmed pass, one confirmed late response and one unresolved duplicate lineage. Report the confirmed records and the unresolved item separately. Do not convert an unknown timestamp into a failure or a pass. The owner can request the original message and confirm the duplication rule.

Keep service context beside the numbers

A supplier review should connect quantitative measures with documented operating context without allowing commentary to overwrite records.

One meeting note attributes the late deliveries to a buyer-approved input change. The delivery export records the revised specification date but not whether the due dates were formally changed. Preserve both facts: an input change occurred, and the packet does not contain approved due-date amendments. Do not present the explanation as a contractual excuse or reject it without review.

Add a context column to the normalized table with the supporting record and its authority. A meeting note can establish that a discussion occurred. It may not establish a contractual change if the required approval record is elsewhere. Ask AI to list plausible missing documents, then turn that list into focused evidence requests rather than assumptions.

The review can state that the raw timing result is sensitive to whether two revised dates were authorized. It cannot decide that question. Keeping the sensitivity visible lets the owner see which evidence could change the conclusion and prevents a broad narrative from hiding a specific documentation gap.

Sources: [1]

Test the review with a corrected record

A controlled supplier analysis should update calculations and narrative without erasing the earlier version.

The owner later supplies an approved change record for one of the two late deliveries. Keep version one of the calculation, record the new source and recompute the accepted-on-time count as 45 of 48, or 93.8 percent. The other delivery remains late under the supplied evidence. The update changes the measure but does not validate every supplier statement.

Use a change log with version, timestamp, source added, rows affected, old result, new result and reviewer. Ask AI to draft the revised narrative from the table, then check that it does not retain the earlier 91.7 percent headline or silently upgrade the unresolved incident.

If the contractual target is present, compare the corrected result with it as arithmetic only. Do not recommend a remedy, sanction or renewal decision. Those actions depend on the actual agreement, authority and wider evidence. The analyst's deliverable is an inspectable input to that process.

Deliver the supplier review note

The final note leads with reconciled measures, evidence limits and owner decisions rather than a single supplier rating.

Deliver the source index, definition table, normalized delivery calculation, incident lineage, context log, evidence requests and version history. Summarize confirmed performance, disputed rows and missing records separately. Name the owner for each open question and the condition that would close it.

The business-operations parent triages an order-exception queue and assigns next actions. This page reviews a supplier across controlled contract measures and linked evidence. The procurement-analyst guide compares bid scope before award, while this page examines performance records after delivery. The supply-chain guide checks stock timing rather than supplier scorecard integrity.

Review the participant on definition control, record reconciliation, transparent calculation, treatment of missing evidence and authority boundaries. A low percentage is not automatically a complete supplier judgment, and a balanced scorecard is not useful when its components use incompatible definitions.

Sources and scope

Cabinet Office guidance: contract measures should be defined, monitored and interpreted within the contract-management context.

The Cabinet Office KPI guidance defines a key performance indicator as a measure used to assess supplier performance during a contract and explains that applicable KPIs are set in the contract. The contract-management principles recommend using KPI data alongside wider relationship and service information, with controlled change mechanisms. These sources support the general measurement and management background only. They do not validate this fictional exercise or determine a legal outcome.

The contract, records, counts, percentages, incidents and change event in this exercise are original synthetic examples. They are not Two Prune customer evidence, supplier benchmarks or procurement advice. Before adapting the method, confirm the actual agreement, reporting definitions, data permissions, professional obligations and decision authority.

Sources: [2] [1]