Work product

Sample AI diligence report

An abridged, composite version of the deliverable, showing the structure of the analysis and reporting.

Composite case study

Snapshot

AI investment diligence snapshot

24
Claims tested: 7
Unsupported: 5
Material findings

MODERATEModel performance

HIGHClaims vs. evidence

HIGHModel dependency

MODERATEData moat

CRITICALInference economics

MODERATEScalability

HIGHAI governance

LOWProduct adoption

LOWTeam & execution

Section 1

Claims tested against evidence

Each marketed claim is restated, matched to the evidence actually reviewed, and resolved into an investment impact.

Claim vs. evidence

Claim: “The product is fully autonomous — agents complete workflows without human review.”
Evidence reviewed: 60 days of production traces, escalation queue exports, two operations interviews
Finding: 31% of agent runs route to a human reviewer, concentrated in the two highest-value workflows.
Confidence: High
Investment impact: Labour savings in the model overstate realised automation. Adjust the efficiency case and treat review-rate reduction as a value creation milestone rather than a delivered capability.

Claim vs. evidence

Claim: “Our models are trained on proprietary customer data no competitor can access.”
Evidence reviewed: Data processing agreements, training pipeline review, dataset lineage sampling
Finding: Fine-tuning uses roughly 4% proprietary data; the remainder is public corpora. Two of five enterprise contracts prohibit training use.
Confidence: Moderate
Investment impact: The data moat is thinner and legally narrower than presented. Discount defensibility, and treat contractual training rights as a diligence condition and a post-close remediation item.

Section 2

Material findings

Findings are ranked by materiality to the thesis, not by technical interest.

Risk: HIGH

Finding: No held-out evaluation set exists; regression testing is manual and run at release by one engineer.
Evidence: Repository review, CI configuration, release checklist, engineering interviews
Investment impact: Quality regressions reach customers before detection, surfacing as churn and support cost rather than an engineering metric, and blocking safe model upgrades.
Recommendation: Fund an automated eval suite with a frozen held-out set in the first 90 days and gate provider migration on it.

Risk: MODERATE

Finding: Retrieval quality degrades on tenants above roughly 50k documents.
Evidence: Latency and relevance telemetry by tenant size, index configuration review
Investment impact: The largest prospects experience the worst product quality, capping the upmarket motion the growth case depends on.
Recommendation: Scope a retrieval re-architecture; hold enterprise ARR assumptions flat until it lands.

AI risk heatmap

Diligence areas scored across evidence quality, severity, likelihood and remediation difficulty. Colour carries the pattern; labels confirm it.

AreaEvidenceSeverityLikelihoodRemediation

Model & evaluation

HIGH

MOD

HIGH

LOW

Data & retrieval

MOD

LOW

MOD

Economics

HIGH

CRIT

MOD

HIGH

Scalability & reliability

MOD

HIGH

LOW

MOD

Governance & security

CRIT

MOD

LOW

Product & adoption

LOW

MOD

HIGH

MOD

LOWMODHIGHCRIT

Section 3

Dependency, defensibility and economics

What the company controls, what it rents, and what happens to margin as usage scales.

AI dependency map

Solid cyan = proprietary and owned. Dashed amber = an external dependency the target does not control.

Proprietary External

Inference economics

Revenue indexed to 100. Inference cost and residual gross margin as monthly token volume scales.

Monthly tokens processed (x). Margin compression at scale is a pricing-model question, not an infrastructure question.

AI moat matrix

Defensibility by stack layer. The filled block marks where the layer actually sits; block height rises with strength.

LayerLowModerateStrong

Model
Third-party foundation models, no fine-tune ownership
Low
Data
Labelled outcome data accumulated per customer
Strong
Workflow
Embedded in approval chain, replicable in ~2 quarters
Moderate
Integrations
Six systems of record, each a switching cost
Moderate
Customer Data
Tenant-scoped history improves retrieval quality
Strong
Brand
Category awareness undifferentiated in buyer interviews
Low
Distribution
Two channel partners, concentration in one
Moderate

Worked example

Inference economics

Technical observation: Inference cost per active account rises roughly linearly with usage because context windows are not trimmed and retrieval returns a fixed large payload per call.
Business consequence: Gross margin compresses precisely as the largest customers expand, so the accounts that drive the growth case are the least profitable to serve.
Investment implication: Underwrite margin as a band tied to usage, not a point estimate, and treat prompt and retrieval optimisation as a funded first-year workstream with a measurable cost-per-account target.

Section 4

Investment thesis matrix

Each thesis assumption is tested and assessed, so the committee can see which parts of the case the evidence supports.

Investment thesis matrix

Each thesis assumption tested against evidence rather than against the management narrative.

Assumption	Evidence	Finding	Assessment	Investment impact
AI accuracy is materially ahead of incumbents	Internal eval set, 3 customer benchmarks, no held-out test set	Advantage holds on two of six task types; eval set overlaps training data	Partially supported	Narrow the win-rate assumption; fund an independent evaluation before close
Gross margin expands with scale	12 months of provider invoices mapped to usage	Inference cost grows faster than seat revenue on agentic workloads	Not supported	Assume margin compression until pricing moves to consumption
AI features drive expansion revenue	Product telemetry, 20-account cohort, expansion ledger	Expansion concentrated in accounts using two of five AI features	Supported, narrowly	Prioritise activation of the two proven features in the value creation plan

Diligence reduces uncertainty and documents what the evidence supports. It does not guarantee that every risk will be identified, and we make no claims about investment outcomes.

Want this run on a live deal?

Bring the thesis, the data room and the timeline. We will tell you what evidence exists, what is missing, and what it means for the deal.

Discuss an AI Investment See a sample report