Skip to content
← All articles

Angel investors

How can an angel distinguish references from independent evidence for an AI application startup?

an AI application startup: Distinguish a source's direct observation from their opinion, inference, or relationship with the company.

Distinguish a source's direct observation from their opinion, inference, or relationship with the company. References can be informative and biased at the same time. Important claims deserve a provenance trail and, where feasible, corroboration.

Start with the decision

For each claim, record the source, relationship, date, actual observation, and available supporting record. Seek permission before sharing private statements. Preserve contradictions instead of selecting only the reference that supports the preferred decision.

The illustrative scenario

This is an educational workbook. The company or fund and all numerical inputs below are fictional. They are not ADIN results, customer claims, market benchmarks, or a forecast.

The fictional team sells an AI assistant to small support departments. A demonstration works on familiar examples, but the product is still testing performance on new customers' queues. Founder time currently covers many exceptions.

Separate model output from completed customer work. A software revenue claim needs a matching view of model usage, human delivery, customer support, and rework.

Turn the scenario into an evidence decision

Consider a review of an AI application startup in which a confident explanation arrives before the supporting records. Start with the first question in the table. Write down what an answer would change: the next interview, the experiment's scope, a request for specialist review, or a decision to wait. A question without a consequence can produce a longer document while adding little understanding.

Next, inspect the record named in the table and record its date, owner, permission boundary, and limitations. Request a permissioned evaluation sample, customer usage cohorts, an exception log, and a cost model that includes human work. Keep actual results distinct from roadmap features. If the record contradicts the explanation, preserve both accounts and ask which definition, period, or workflow differs. Do not silently select the version that makes the opportunity look strongest.

Finish with a conditional statement: if the specified evidence supports the claim under these conditions, take the named next step; otherwise investigate the unresolved question. Source quality depends on relevance and independence as well as reputation. A detailed firsthand account can answer a narrow question while remaining insufficient to support a broader claim about the company. This worked exercise is about the decision process. It does not assert that a fictional company has met a real diligence standard or that a template can authorize an investment.

Worked example: an AI application startup
ItemIllustration
Decision questionWhat was directly observed?
Case-specific constraintSeparate model output from completed customer work. A software revenue claim needs a matching view of model usage, human delivery, customer support, and rework.
Record to inspectRequest a permissioned evaluation sample, customer usage cohorts, an exception log, and a cost model that includes human work. Keep actual results distinct from roadmap features.
Comparison to preserveCompare the promised workflow with the delivered workflow. The relevant question is whether adoption and delivery economics improve together.
Unresolved case questionWho reviews failures?

Use this worksheet

Answer the questions with a source, date, owner, and next action. Mark an unknown as unknown. Use public or permissioned information and involve a qualified specialist when a question requires expertise beyond the team’s competence.

  1. What was directly observed?
  2. When was it observed?
  3. What relationship exists?
  4. Can a record corroborate it?
  5. Which source disagrees?

Questions specific to this scenario

  • Who reviews failures?
  • How much founder work supports each account?
  • Which results hold on unseen examples?

Request a permissioned evaluation sample, customer usage cohorts, an exception log, and a cost model that includes human work. Keep actual results distinct from roadmap features.

A concrete next test

Observe a permissioned workflow where the assistant fails or needs founder intervention. Record the human time and what the customer sees. Compare that case with the demo so the diligence discussion addresses delivered work rather than only the best familiar example.

Interpret the result

Source quality depends on relevance and independence as well as reputation. A detailed firsthand account can answer a narrow question while remaining insufficient to support a broader claim about the company.

Compare the promised workflow with the delivered workflow. The relevant question is whether adoption and delivery economics improve together.

Sources and method

These primary resources provide background for the concepts. The fictional scenario, calculations, and worksheet are ADIN educational examples; the sources do not report or validate those invented results. Reviewed October 11, 2026.