Skip to content

Manufacturing

Manufacturing data readiness checklist for AI, including data rights

By SourceX Editorial · Updated

Short answer

A manufacturing data readiness checklist for AI should test twelve points: where records live, how they link, history depth, free text, completeness, consistency, owners, export path, personal data, exclusions, contract rights and approvals. Most vendor-written checklists stop at data quality. When records are going to an outside AI developer, the rights checks decide readiness as much as clean data does.

Key takeaways

  • Readiness for licensing adds rights, exclusions and approvals to the data quality checks used for internal AI projects.
  • Linkage between quotes, jobs, quality and maintenance records matters more than any single clean table.
  • Every record family needs a named owner and a tested export path before it can be scoped.
  • A failed rights check narrows the scope rather than ending the review.

How licensing readiness differs from internal AI readiness#

Licensing readiness differs from internal AI readiness because an outside party will use the records. Internal projects ask whether data is clean and connected enough for your own model; licensing asks the same questions and then whether you control the records, may permit outside use and can document what was shared.

That is why most vendor-written checklists, which assume data stays inside your own systems, stop short. They cover sensors, data quality and integration well, and leave ownership, customer designs and contract limits to someone else.

How licensing readiness differs from internal AI readiness
QuestionInternal AI projectLicensing to an AI developer
Is the data clean and linked?CentralCentral
Who owns each record family?Rarely askedRequired before scoping
Do contracts allow outside use?Not relevantRequired for each customer and vendor
Is personal data removed?Depends on the useRequired, with a privacy record
Is the release approved and documented?Internal sign-offWritten approval and evidence of what left

The 12-point checklist#

The 12-point checklist below tests usability first and transferability second. Points one to seven ask whether the records can teach anything; points eight to twelve ask whether they can leave the company. Score each point ready, partial or not ready for every record family, not for the company as a whole.

The 12-point checklist
CheckReady looks likeCommon red flag
1. LocationEach record family mapped to a named systemRecords spread across personal drives and inboxes
2. LinkageQuotes, jobs, NCRs and work orders share referencesOrders keyed in fresh with no quote number
3. History depthSeveral years still accessible in one formatHistory lost in a past ERP or MES migration
4. Free textNotes explain causes, decisions and fixesComment fields blank or pasted from templates
5. CompletenessKey fields such as disposition and root cause filled inDispositions or causes routinely missing
6. ConsistencyCodes and units stable over timeReason codes renamed after each reorganization
7. OwnerA named person answers for each record familyNobody knows who maintains the quality database
8. Export pathA tested route to extract records and attachmentsOnly screen views or PDF reports available
9. Personal dataEmployee and contact fields identified for removalNames embedded throughout free text
10. ExclusionsCustomer designs and export-controlled jobs flaggedDrawings attached to quotes and NCRs
11. Contract rightsCustomer, supplier and software terms reviewedQuality agreements never read for outside-use limits
12. ApprovalsSigner and required consents identifiedUnclear which entity owns legacy records

How to read the scores#

Reading the scores is about deciding scope, not passing or failing the company. A record family that is ready on linkage, depth and rights can move forward even while other families are still not ready.

Maintenance work orders and nonconformance records often score well early, because they are company-authored and describe the company's own equipment and processes. Quote and job records tend to score high on linkage but need more contract review, since they reference customer parts.

  • Ready on location, linkage, depth, export path and contract rights: a strong candidate for a first package.
  • Partial on free text or completeness: still useful, but describe the gaps in the data dictionary.
  • Not ready on export path: resolve with IT or the software vendor before any further scoping.
  • Not ready on contract rights or exclusions: narrow the scope to the record families that pass.
  • Not ready on approvals: settle who signs before spending effort on preparation.

Which rights checks do vendor checklists skip?#

Vendor-written checklists usually skip the rights checks because they treat AI as an internal project. Once records go to an outside developer, the questions below decide what is in scope.

The list is general information, not legal advice; counsel weighs the contracts and laws that may apply to each deal.

  • Customer purchase terms, NDAs and quality agreements that restrict outside use of anything referencing their parts.
  • Customer-owned drawings, models and specifications attached to quotes, travelers and NCRs.
  • Export-controlled programs, which are removed in full.
  • Software vendor terms that limit bulk export or third-party access to data in the system.
  • Employee notices and policies covering names, IDs and free-text mentions in operational records.
  • Earlier licenses, data-sharing agreements or exclusivity granted to a software vendor or partner.
  • Lender, investor or parent-company consents that apply to licensing company assets.

Who should run the checklist?#

The COO or VP of operations should run the checklist, because it cuts across systems, functions and contracts. Their job is to coordinate the answers, not to produce them all.

IT or the OT lead answers location, export path and history depth. The quality manager answers linkage and completeness for NCRs and corrective actions. The controller or CFO knows which contracts and lender documents matter, and counsel reviews rights and exclusions. The CEO confirms the signer and whether the company wants to proceed at all.

Keep the answers in one shared inventory, with one row per record family, so scores, owners and open questions sit together. That inventory becomes the input for any later fit check and saves asking the same questions twice.

Illustrative: a packaging converter runs the checklist#

Illustrative: a fictional flexible packaging converter prints and laminates film for food brands. It runs an ERP for orders and jobs, an MES on its presses and laminators, and tracks nonconformances and maintenance in separate databases. The COO runs the checklist with IT, quality and the controller.

Location, linkage and history depth come back ready for press downtime, maintenance and nonconformance records. Free text is partial: maintenance notes are detailed, but nonconformance descriptions are short. Contract rights fail for printed artwork and customer specifications, which belong to the brands.

The company scopes maintenance, downtime and nonconformance records for a first package, excludes artwork and customer specifications entirely, and asks quality to write fuller nonconformance descriptions from now on. Job records are parked until the controller confirms which customer agreements allow outside use.

How SourceX uses a readiness checklist#

SourceX uses the same twelve points during Supply, the first stage of the SourceX five-step transaction, to decide which record families deserve a deeper look. The answers are metadata only, so a COO can complete them without sending a single file.

Families that pass are weighed with the SourceX Enterprise Data Value Framework and then move through Rights, Preparation, Approval and Delivery. Points nine to twelve carry straight into the SourceX Evidence Packet, which records provenance, licensing rights, permitted use, the privacy record and release authorization for each package, so the company can show later what was released and on whose authority.

Frequently asked questions

How long does the checklist take to complete?

Most points can be answered from memory and short conversations with IT, quality and finance. The contract rights point takes longest because it means reading customer, supplier and software agreements, so start that review early and run it alongside the rest.

Do we need a data lake or warehouse first?

No. Records can be assessed where they live today, in the ERP, MES, quality system or spreadsheets. Building new infrastructure before knowing which record families are licensable risks spending money on the wrong data.

Should we clean records before running the checklist?

No. Score records as they are and keep source data unchanged. Cleanup decisions come later, during preparation, and every change is documented so a buyer can see what was altered and why.

Is the checklist useful if we only want internal AI?

Yes. The first seven points apply to any AI project, and the rights points still help if you plan to share data with a software vendor or use outside tools that process your records.

What if a record family fails several checks?

Park it rather than discard it. A system migration, a contract renewal or better note-taking habits can change the result, and the checklist leaves you with a clear list of what to fix.

Does the checklist cover OT data such as historians and PLC logs?

Yes. Treat each historian, SCADA archive and machine log as its own record family. The usability points apply unchanged, while the export path and exclusion points need extra attention, since recipes, setpoints and network details are often sensitive.

Related resources

See if your company qualifies

A short company assessment. No data uploads are needed.

See if you qualify