Tacit knowledge and AI training: how agents learn unwritten company know-how

How does tacit knowledge get into AI training?

It gets in sideways, through records. Nobody writes down why a senior planner splits a large order across two carriers, but the email where she explains it to a new hire, the ticket note that records the workaround and the approval comment that cites last winter's missed delivery all survive in company systems. Read together and in sequence, those traces are how AI agent builders get at know-how that was never documented.

This matters for consultants because the documents a client hands over first, such as process maps, SOPs and policy manuals, describe the official process. The records show what experienced people actually did when the official process did not fit, and that is the material agent builders find hardest to obtain. It is also why company documents are valuable for AI mainly when they come with the working history around them.

Where does unwritten know-how leave traces?

Look for places where a person applied judgment and wrote something down while doing it. All examples in the table are illustrative.

System history adds the sequence: the order in which steps happened and how long each one took. That side is covered in ERP and system event logs as agent training data.

How do SOPs compare with records of practice?

An SOP is a map; records are the journeys people took on it. Both are useful, but they teach different things.

Paired, they are stronger than either alone: the SOP names the steps, and the records show how skilled people handle the cases the SOP glosses over.

The 3E test for spotting know-how-rich clients

Use three questions, plus depth and the baseline, on any client you know well. A clear no on Explanations or Endings usually means the know-how stayed in people's heads.

• Exceptions: do people log overrides, holds, escalations and workarounds in a system, rather than settling them by phone?
• Explanations: do notes, comments and replies say why, not only what?
• Endings: can you see how things turned out: ticket closed or reopened, quote won or lost, claim paid or denied?
• Depth: do the records go back several years, including systems that have since been replaced?
• Baseline: is it a US company with 50+ full-time employees at peak (contractors excluded), rights to its own records and an owner or executive who can sponsor a license?

For a quick, non-binding first pass on the baseline, use the company fit checker.

When do consultants see the signs?

Engagements that map how work really happens tend to surface the clues first.

Why does this matter to your clients?

Know-how leaves with people. McKinsey estimates that about six million US small and medium-size businesses will face ownership transitions by 2035 as baby boomers retire. When a founder or long-tenured manager steps back, the record of their decisions is what stays behind.

AI developers want that record because agents are moving from answering questions to doing work, and doing work well means handling the cases nobody planned for. A company can license the records it already holds, keep ownership and receive a one-time payment, without changing how it operates. For how those private deals are structured, see how enterprise AI data licensing deals work.

Who owns the know-how in the records?

Usually the company, when employees created the records as part of their jobs. The Copyright Office's circular on works made for hire explains that for a work prepared by an employee within the scope of employment, the employer is the author and owner. Material from contractors or outside consultants may not belong to the company unless a signed written agreement says so.

Two limits apply. Records a company holds on behalf of its own clients, as at many agencies and outsourcers, are a red flag unless those clients consent. And your own engagement files are never part of a referral: you do not export, upload or describe the client's records.

This is general information, not legal, tax or financial advice. Ownership questions belong with the company's own counsel.

How does a consultant make the introduction?

1. Raise licensing in a normal conversation with the owner or an executive, and ask whether they want to explore it.
2. If they do, send your referral link so the company applies at sourcex.si/apply with your code attached, or submit it through the referral form.
3. SourceX screens size, operating history, breadth of systems and rights directly with the company's sponsor.
4. The company inventories its systems and agrees price and terms with SourceX before buyers review anything.
5. The deal closes, the company delivers under the agreed redaction rules and is paid; your reward follows SourceX's receipt of its fee.

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company. It is paid only after the buyer pays and SourceX receives its fee, it is never deducted from what the client receives, and rewards are not guaranteed. Check your engagement letter and firm policy on referral compensation before you register; the referral program for management consultants covers the rest.

What records cannot capture

• Judgment exercised in hallway conversations or phone calls, and never typed anywhere, leaves no trace.
• Terse records show what happened but not why, and carry little know-how.
• Bad habits are recorded too; outcomes are what let buyers tell good practice from bad.
• Companies below the baseline, or whose records are mostly client-owned or consumer personal data, are not a fit.
• Records generated with AI in order to sell them are a red flag; buyers want evidence of human work.

Next step

Run your next process-mapping client through the 3E test. If it passes, register as a partner and make the introduction; the full sequence from referral to payment is on how it works.