Data Inventory Checklist for a First AI Partnership Conversation

Short answer

A first data inventory should make one repeatable operational process understandable without exposing sensitive material. Record what the process is, how many completed cases exist, how stages connect, where quality is checked, what is missing, and what rights or restrictions apply. You do not need a perfect catalog or every document type to begin. A small, truthful inventory is more useful than a broad but vague claim.

Copy-and-adapt inventory template

This worksheet is optional preparation, not an additional application requirement. Start with what you know; leave unmeasured items marked “unknown.”

Process or use case

______________________________

Plain-language description

What work repeats, and what decision or outcome is recorded?

Systems or record families

_________________________

Completed cases

Approximate count __________ Period: __________

Linking key

case / order / ticket / job ID / other: __________

Stages connected

start ______ steps ______ exceptions ______ outcome ______

Outcome check

Who or what confirms completion or quality? __________

Known gaps

missing fields, duplicates, corrections, selection limits: __________

Restrictions

customer, vendor, privacy, confidentiality, retention, geography: __________

Export path

format, owner, access controls, approvals: __________

Unknowns to investigate

____________________________

The six-signal checklist

Use the following prompts for each candidate record family:

• The records come from repeated work or a recurring decision.
• Completed volume and the covered period are stated.
• A stable identifier links the beginning, steps, and outcome where applicable.
• Exceptions, corrections, failures, or overrides are represented or their absence is known.
• The way completion or quality is checked is documented.
• Sharing restrictions and the responsible owner are identified.

These checks describe readiness for a conversation; they do not predict acceptance or establish licensing rights.

Fully synthetic completed example

Process: appliance repair dispatch. Cases marked closed: 1,240 jobs from January–June 2026. Rows: 6,980 event rows across dispatch, diagnosis, parts, repair, and final test. One case can have many rows, so 6,980 rows does not mean 6,980 completed cases. Linking key: job_id. Outcome check: technicians record a final test result; 86 closed cases lack that result. Administrative closure therefore does not establish a verified repair outcome for every case. Known gap: the export excludes two subcontractor teams. Restriction: customer service agreements require owner review before external sharing. Unknown: whether historical warranty notes may be exported.

How to describe quality honestly

Separate observations from estimates. “The export contains 42,000 rows” is an observation if you have counted it. “The rows represent 42,000 completed cases” requires a definition and a check. If you have not measured missingness, write “unknown” rather than implying a clean dataset. If a sample was selected from one location, team, or month, say so.

Include the unglamorous records where permitted: rework, rejected outcomes, escalation, and corrections. They can explain how the operation behaves in real conditions. Do not overstate what a sample proves. A small coherent sample can demonstrate field meaning and link structure; it cannot establish that the full archive has identical coverage or quality.

Rights and safe preparation

List who created the records, who controls them, and which agreements govern reuse. A company may possess an export while a client or vendor retains contractual control. BPO and client-service data deserve particular care. Do not make legal conclusions, promise anonymization, or include personal names, credentials, client identifiers, or confidential attachments in an initial description. Ask the responsible owner what approvals and redactions are required.

The first handoff can usually be a metadata-level description: process, volume range, date range, field categories, link keys, gaps, restrictions, and export owner. If a final data user later requests a sample, confirm authorization, selection method, and handling expectations before sharing.

DataToAsset’s current referral pilot is free to the company. The company voluntarily applies through Check eligibility; the final data user evaluates opportunities, contracts directly, and decides payments. DataToAsset may receive a referral fee from that final data user. The referral route does not guarantee acceptance, timing, earnings, or payment, and DataToAsset does not take possession of data for this pilot.