Source pump-repair workflows for AI A worked buyer brief showing exactly what a sourcing assignment can ask for. Illustrative buyer brief. All quantities and acceptance thresholds are proposed example requirements, not buyer commitments or market standards. Reviewed: October 8, 2026 OBJECTIVE Find businesses that can license connected examples of industrial pump diagnosis and repair. The buyer wants to test whether an AI assistant can find relevant precedents and produce a grounded repair summary for a human technician. The buyer is not asking for an autonomous repair system. COMPANY SEARCH Requirement | Included | Excluded Business type | Independent repair shops, field-service providers and equipment service divisions | Companies that only resell parts with no service workflow Geography | Start with Michigan; expand nationally if fit is better | No claim that Michigan alone can supply the volume Data history | Prefer at least two years of digitized closed jobs | Paper-only archives requiring a large conversion project Decision maker | Owner, operations leader or person authorized to approve a license | An employee who cannot obtain company approval Commercial interest | Willing to discuss a defined paid license | Permission assumed from a public website DATA REQUIREMENTS The unit of data is a completed repair job, not a document. Each job should connect the incoming problem, diagnosis, action and available result. The initial search should assess whether these links exist before requesting any raw files. Required | Preferred | Out of scope Problem description and equipment family | Operating conditions and prior repair history | Customer identities or private contact details Technician diagnosis and actions | Parts used and labor stages | Plant-security or access information Recorded result or explicit “not recorded” marker | Test method, values and units | Unlicensed manuals, drawings or images Stable links between related files | Timestamps and change history | A large unstructured folder with no job mapping SAMPLE DESIGN Ask one willing company for a scoped, approved 30-job sample. Use 10 routine cases, 10 diagnostic cases and 10 repeat-visit or uncertain cases. A buyer should inspect both successful outcomes and cases where the record is incomplete. - 30/30 job identifiers map to the provided files. - At least 24/30 jobs contain a readable symptom and action; this is an example threshold to negotiate. - Every job has either an outcome or an explicit missing-outcome flag. - No obvious personal contact information appears in the buyer sample. - A named reviewer assesses whether the sample is useful for the stated AI task. DELIVERABLES FROM THE SOURCING ASSIGNMENT A sourcing fee pays for agreed research and coordination, not a guarantee that a company will sell data. The scope should define what counts as a qualified result. - A public-source research report with company names, source URLs and the reason for each inclusion. - A contact plan naming the roles to approach; private contact information stays in the agreed working file. - A company profile for each business that participates, separating its statements from checked facts. - A status record: researched, contacted, conversation held, interested, sample approved, or declined. - A summary of objections, missing requirements and recommended changes to the search. COMMERCIAL AND DELIVERY BOUNDARIES The buyer and SnowRock agree the sourcing fee, research scope, included outreach and payment milestones before work begins. The company’s license payment is negotiated separately. Any extraction, cleanup or labeling work needs a named provider and a separate cost. The buyer sets a budget range for the search and a separate range for a potential license. A useful first milestone is a small sample review, followed by a decision to expand, change the requirements or stop. This example intentionally contains no invented market price. DECISION RULES Continue when a company has relevant connected jobs, an authorized decision maker, a plausible permission path and willingness to discuss a sample. Pause when the files cannot be connected, the useful content is mostly third-party material, or a rights question cannot be resolved. Stop when the buyer cannot define a task or acceptance decision. Next action: the buyer fills in its budget, sample reviewer, intended uses and required decision date. SnowRock then proposes the scope and fee for the search.