RIGHT ENGINES
Due diligence repeats but learning does not compound
A practical guide to diagnosing due diligence repeats but learning does not compound and deciding whether an AI engine is justified.
The problem
Review teams repeatedly gather, normalize, and test evidence under time pressure.
Symptoms
Look for recurring delay, correction, escalation, or paid attention around this workflow. Confirm the pattern with scope, materiality thresholds, evidence set, source authority, deadlines, and reviewer roles. rather than relying on anecdotes.
Why it happens
Checklists exist, but evidence provenance, exceptions, and prior learning are fragmented.
What it costs
Measure the burden created by due diligence repeats but learning does not compound: responsible-person hours, elapsed delay, rework, error exposure, outside spend, and decisions deferred. Use a representative baseline; do not invent ROI.
What people usually try
Uploading a data room to a generic chatbot without access controls or a review model.
Where AI helps
For this problem, AI may help by turning scope, materiality thresholds, evidence set, source authority, deadlines, and reviewer roles. into evidence inventory, gap analysis, exception log, source-linked findings, and review workflow. It should preserve evidence and expose exceptions for review.
Where AI does not help
Do not automate legal, investment, or regulated conclusions that require qualified accountable judgment.
What an engine could do
Evidence inventory, gap analysis, exception log, source-linked findings, and review workflow.
Inputs required
Scope, materiality thresholds, evidence set, source authority, deadlines, and reviewer roles.
Success measures
Coverage, material gaps found, false-positive rate, and reviewer acceptance.
Risks
The specific failure boundary is clear: Do not automate legal, investment, or regulated conclusions that require qualified accountable judgment. Also test provenance, access, false confidence, and the effect of unreliable inputs.
Small / medium / large solution
Start with the smallest validated route. Use an Outcome Kit only where its stated inputs and tests fit due diligence repeats but learning does not compound; adapt a proven engine where workflow context differs; commission private work only when the evidence, integration, governance, or rights justify it.
Estimate this problem’s burden · Decide whether it is worth building for · Review reusable Outcome Kits
Published 2026-08-31 · Updated 2026-08-31 · Version 1.0