— Sample deliverable · AI DD

The AI Due-Diligence Memo.

The artifact an investor or acquirer reads before signing a term sheet. An independent, code-level verdict on an AI company or asset — separating real technical advantage from narrative. It answers one question without hedging: is the AI here a defensible asset, or a story wrapped around someone else's API?

— Orientation

What it is, when to use it, who it's for

What it is

An independent, code-level technical verdict on an AI asset, with a clear proceed / conditions / pass call and a deal-implications section.

When to use it

Before writing a check into an AI company or acquiring one — when the founder says the moat is the model and you need truth before terms.

Who it's for

VC, PE, and M&A teams, corp-dev, and boards who need code-level technical truth that a generalist diligence provider can't deliver.

— Structure

Sample table of contents

AI Due-Diligence Memo · structure10 sections

Investment verdict

A single, unhedged call — proceed, proceed with conditions, or pass — stated up front.

Product reality

What the product actually does today vs. what the deck and demo imply.

AI depth

Where real ML/AI sits in the stack vs. wrappers, prompts, or off-the-shelf APIs dressed as a moat.

Data moat

Whether proprietary data exists, is defensible, is legally clean, and actually improves the model.

Engineering quality

Architecture, test coverage, deployment maturity, and the team's ability to ship at scale.

Security & compliance

Data handling, access controls, model/supply-chain risk, and regulatory exposure.

Inference-cost & margin

Unit economics of running the AI — token/compute cost per transaction and the path to gross margin.

Roadmap realism

Whether the stated roadmap is achievable with the current team, data, and capital.

Red flags

An explicit list of the issues that should give an investor pause — severity-rated.

Deal implications

How findings should change valuation, terms, diligence conditions, or integration plan.

— Illustrative example

A redacted verdict & scoring summary

Illustrative / redacted sample — not a real client artifact

Investment verdict

Proceed with conditions

AI depthReal fine-tuning in one module; remainder is API orchestration.Mixed

Data moatDefensible first-party dataset; licensing clean; compounds with usage.Strong

Engineering qualitySolid architecture; thin test coverage; deployment below stated stage.Adequate

Inference-cost marginCost per transaction high; gross margin below plan until optimization lands.Weak

Roadmap realismNext two quarters credible; agent roadmap needs hires not yet budgeted.Partial

Verdict card mock — signal colors (red / amber / green) and ratings are illustrative placeholders showing the memo's scale. Not a diligence on any real company.

Dimension	Assessment	Rating
Investment verdict	Proceed with conditions — re-price on margin findings; require [redacted] before close.	Conditional
AI depth	Genuine fine-tuning on proprietary data in [module]; remainder is API orchestration.	Mixed
Data moat	Defensible first-party dataset; licensing clean; compounding with usage.	Strong
Engineering quality	Solid architecture; thin test coverage; deployment maturity below stated stage.	Adequate
Inference-cost margin	Cost per transaction high; gross margin below plan until [optimization] lands.	Weak
Roadmap realism	Next two quarters credible; agent roadmap requires hires not yet budgeted.	Partial

Assessments and ratings are illustrative placeholders demonstrating the memo's format and rating scale. This is not a real diligence on any company — GenovateAI does not publish client or target findings.

— Red-flag examples

What earns a flag before the term sheet

Representative red flags the memo is built to catch. These are illustrative patterns we look for — not findings about any specific company.

▲“The moat is the model” — but the model is a thin wrapper over a third-party API with no proprietary training data or fine-tuning.
▲Inference cost per transaction exceeds revenue per transaction, with no credible path to positive gross margin at scale.
▲Headline accuracy metrics are measured on a curated internal set; performance on real, messy production data is materially lower or untested.
▲The “proprietary dataset” is scraped, licensed on revocable terms, or carries unresolved IP / privacy provenance.
▲A single engineer holds all model and infra knowledge; no tests, no documentation, no reproducible training pipeline.
▲Demos run on hard-coded or cherry-picked inputs; the system has never been load-tested or run unattended in production.
▲Roadmap assumes capabilities (agents, multimodal, on-device) that the current team, data, and runway cannot realistically deliver.
▲Customer logos imply production usage that is actually a paid pilot, a design partner, or a lapsed trial.

— The decision it enables

An investment committee can decide to proceed, re-price, set diligence conditions, or pass — on a code-level technical verdict, not the founder's narrative.

Book an AI Decision Call

The AI DD Memo is the artifact most investors ask about first. Book a call to scope a diligence on a live deal — or to see how the memo would read for an asset you're evaluating.

Book an AI Decision Call