Composite case studies

How the work reads in practice

See how AI diligence findings connect technical evidence to business consequences and investment implications.

Type

Transaction

Sector

6 of 6 case studies

AI Due DiligenceGrowth EquityFintech

Accuracy leadership that did not survive a held-out test

Composite case study

A growth investor was underwriting a document-processing company on a claimed accuracy advantage over incumbents, with win rates in the model built on that advantage persisting.

Headline findingHIGH

Technical observation: The published accuracy figure was produced on an internal evaluation set that overlapped the fine-tuning corpus; no frozen held-out set existed, and production traces showed a materially smaller margin on four of six task types.
Business consequence: The differentiation the sales motion is priced on is narrower than represented, and there is no internal mechanism that would have detected the erosion, so it can widen without anyone noticing.
Investment implication: Win-rate and pricing-premium assumptions require re-underwriting on the two task types where the advantage holds. An independent evaluation with a frozen test set should be a pre-close condition and a recurring covenant.

How the client used it

The investor renegotiated the entry multiple and made an independent evaluation programme a funded first-year initiative.

CombinedBuyoutEnterprise SaaS

Margin compression hidden inside a flat-fee pricing model

Composite case study

A sponsor evaluating a buyout modelled gross margin expansion as the AI product scaled across the existing customer base.

Headline findingCRITICAL

Technical observation: Mapping twelve months of model-provider invoices to product surfaces showed inference cost per account growing roughly twice as fast as seat revenue, driven by agentic workflows priced under a flat per-seat fee.
Business consequence: Every incremental power user reduces gross margin, so the company's most successful accounts are its least profitable and growth actively erodes the margin case.
Investment implication: The margin bridge should assume compression until pricing moves to a consumption or hybrid model. Repricing is the central value creation lever and carries churn risk that belongs in the downside case.

How the client used it

The sponsor rebuilt the margin bridge, sized a repricing programme, and structured part of the consideration against retention through the transition.

AI Product DDSeries ALegal & Compliance

An AI flagship almost nobody used

Composite case study

A venture investor was evaluating a company whose narrative and roadmap centred on an autonomous review agent.

Headline findingMODERATE

Technical observation: Trailing telemetry showed the agent used by under 20% of eligible accounts, with 31% of its runs escalated to human review; retention correlated instead with two unglamorous classification features.
Business consequence: The product the company sells is not the product customers operate, so roadmap investment and sales positioning point away from the features that actually drive renewal.
Investment implication: Expansion assumptions tied to the agent are unsupported. Underwrite on the classification features, and treat agent activation as a milestone with a measurable adoption target rather than as an existing strength.

How the client used it

The investor proceeded at a revised valuation with adoption milestones written into the board reporting pack.

AI Technical DDStrategic M&AHealthcare

Tenant data commingled in the observability layer

Composite case study

A strategic acquirer intended to fold a clinical documentation product into a regulated enterprise platform.

Headline findingCRITICAL

Technical observation: Prompt and response payloads containing patient identifiers were written to a shared observability index with no tenant partitioning and a 24-month retention setting inconsistent with the customer agreements in place.
Business consequence: The target cannot pass the acquirer's own vendor review without remediation, and the exposure attaches to the acquirer on day one alongside a disclosure obligation.
Investment implication: Remediation and a documented retention change should be conditions to closing, with indemnity scoped to the historical retention window rather than to future operation.

How the client used it

Closing conditions and a specific indemnity were added; remediation was completed pre-close.

AI Due DiligenceGrowth EquityIndustrial

A data moat that was storage, not advantage

Composite case study

The thesis rested on a compounding proprietary data advantage accumulating across an installed sensor base.

Headline findingHIGH

Technical observation: Data volume grew steadily, but measured task performance was flat across three retraining cycles because collected data was never linked to outcome labels.
Business consequence: The asset described as a moat is an operating cost. Competitors with less data but outcome-linked labelling can close the quality gap faster than the volume argument suggests.
Investment implication: Value the business on installed-base distribution and switching costs rather than on data compounding, and fund an outcome-labelling programme as the condition for the moat thesis to become true.

How the client used it

The investor rebased the terminal growth assumption and funded a labelling and evaluation workstream.

Vendor AssessmentEnterprise PurchaseFintech

A vendor assurance that was narrower than stated

Composite case study

An enterprise buyer was days from signing a multi-year AI vendor agreement on the strength of a no-training-on-customer-data commitment.

Headline findingMODERATE

Technical observation: The vendor's own commitment was sound, but a subset of traffic routed to a consumer-tier provider endpoint whose terms did not carry the same restriction.
Business consequence: The assurance the buyer's risk committee approved does not cover all traffic, which creates a control gap the buyer would own rather than the vendor.
Investment implication: Require the enterprise provider tier and a named sub-processor list contractually before execution, with audit rights attached, rather than accepting the summary assurance.

How the client used it

The buyer had the terms amended pre-signature and added an annual sub-processor attestation.

A note on what is not here

No client names, logos, transaction values or testimonials appear on this site, and none will be supplied on request. Confidentiality is a condition of the work.

Considering an AI investment?

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