All services · AI for one work task

One workflow · verifiable output

Price-list matching with a disputed-item queue

For purchasing staff, supply managers, or catalog managers who regularly compare supplier offers. The assistant suggests candidate matches, explains the comparison, and puts ambiguous items in a separate review queue.

Denis Skripnik — a blind specialist in working with AI agents. I examine the task, configure the workflow, and work with you to verify the result against agreed rules.

AI agent setup is already selected in the brief. Name the scenario: “Price-list matching with a disputed-item queue”. Describe the files or systems you use, the result you need, and any exceptions. Start with a description of an anonymized sample.

Input: catalog, price lists, and comparison rules

• An internal catalog export and selected supplier price lists in agreed formats, with source dates or versions.
• Rules for part numbers, manufacturers, dimensions, units, and packaging; agreed treatment of price terms, currencies, and VAT.
• Examples of correct, incorrect, and ambiguous matches. We identify which attributes must never be ignored.

Output: a table for reviewing matches

• For each catalog item: its source row, price-list candidates, file and row references, compared attributes, and the reason for each suggestion.
• Ambiguous candidates, unit conflicts, and missing attributes enter a review queue with reasons. Items with no candidate match are explicitly marked.
• A field for an employee’s decision and an agreed export. No automatic CRM or 1C writes and no record merging.

Example: similar names do not mean identical products

This is an illustrative example, not a client case study. The catalog requires an M8×20 bolt sold individually. A price list contains M8×30 bolts and M8×20 bolts in packs of 100. The first conflicts on length; the second requires checking packaging and whether conversion is allowed. Neither can be declared equivalent just because its name looks similar.

The output shows source rows and conflict reasons. A pack price may be converted only under an agreed rule; conversion does not make the pack an identical catalog item. The purchasing employee approves the final match.

What we configure for this workflow

• Loading agreed files and normalizing only fields with defined rules.
• Exact checks for part numbers, dimensions, and units, followed by candidate search. AI may help with inconsistent names, but cannot override hard constraints.
• A conflict queue, export, and test set covering an exact match, different length, pack versus piece, multiple candidates, and no match.

Limits and quality checks

This is not an all-supplier catalog, a best-price guarantee, or automated purchasing. Unreadable files, a new supplier, changed layouts, and missing attributes may require separate configuration.

Before implementation, we agree which matches are permitted, which always require review, and how errors and queue completeness are assessed. At acceptance, an employee checks source rows, candidates, conflicts, and exports in the customer’s environment.

Sources, access, and your responsibilities

• An anonymized export is enough to start: remove information that could identify a person, business secrets, and unnecessary fields. Do not put passwords or keys in the brief.
• For implementation, we agree the minimum necessary permissions, preferably a dedicated account with read access only to selected sources. An export instead of production-system access is an acceptable alternative.
• You provide lawfully available samples, rules, an authorized environment, and a designated reviewer who resolves disputed data. We agree storage, deletion, and permitted model-provider data transfers in advance.

Acceptance, instructions, and training

At acceptance, we run the chosen workflow in the customer’s environment. We agree criteria, a test set with ordinary and disputed examples, output format, and error reporting before implementation. Handover provides operating and review instructions and training for the responsible employee; their format and scope are specified in the proposal.

Iteration count, correction of nonconforming results, requirement changes, and support after acceptance are agreed separately before starting. Unlimited revisions and ongoing support are not implied.

Timing, price, and external expenses

The price and exact duration are confirmed after reviewing a sample and before work starts. They depend on data quality, formats, volume, environment, and review requirements. Existing service price guidance is not a fixed quote for this scenario.

If a sample is insufficient to define the scope reliably, a paid discovery or bounded pilot can be agreed separately, with its own deliverable and criteria. It does not automatically commit either party to full implementation.

Server costs, paid services, and model usage are separate from implementation fees. Before starting, we agree what is necessary, who pays for accounts, and spending and processing limits. Payment for the work follows scope and quote confirmation only.