Skip to content
Probatio

AI + Technical + Product Due Diligence

AI Due Diligence for Investors & Acquirers

We evaluate AI companies. Not software in general — AI specifically: the models, the data, the evaluation discipline, the inference economics and whether any of it produces value a customer will keep paying for.

Know what's really behind the AI.

Discuss an AI InvestmentHow we work
Primary practice
AI Due Diligence
Supporting
AI Technical Due Diligence
Supporting
AI Product Due Diligence

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

The diligence frame

Eight questions every AI investment has to answer

01

Does it work?

Measured performance against the task the buyer is actually paying for.

02

Is it real?

Claims traced to evidence: evals, logs, code, not demo footage.

03

Is it proprietary?

What is owned versus rented from a foundation model provider.

04

Is it scalable?

Behaviour under load, latency budgets, failure modes.

05

Is it economical?

Inference economics and gross margin at realistic usage.

06

Is it valuable?

Whether AI features drive adoption, retention and price.

07

Is it durable?

Data moat, switching costs, model-provider commoditisation.

08

Can the org execute?

Team depth, evaluation discipline, AI governance maturity.

How we write

A technical fact is not a finding

Bare technical statements are useless to an investment committee. Every observation we record is carried through to what it costs, constrains or changes about the deal.

  • We do not report that a system 'uses a frontier model via API'. We report what that dependency does to margin, roadmap control and negotiating position.
  • Confidence is stated explicitly. Where evidence is absent, we say the evidence is absent rather than inferring.
  • Findings are ranked by materiality to the thesis, not by technical interest.

Worked example

Model dependency

Technical observation
The core capability is produced by two external frontier model providers, invoked directly from the orchestration layer with no abstraction and no tested fallback.
Business consequence
Unit cost, latency and output quality are set by a supplier the company does not control, and any provider deprecation forces unplanned re-evaluation work across the product.
Investment implication
Supplier concentration and pricing risk sit upstream of gross margin. Price the deal with a margin band rather than a point estimate, and make a provider-abstraction plan a condition of the value creation plan.

Evidence, not narrative

What the work product looks like

Purpose-built artefacts: claims tested against evidence, dependency maps, moat matrices and inference economics.

Claim vs. evidence

Claim
“Our model is 40% more accurate than anything else in the category.”
Evidence reviewed
Internal eval harness, 3 customer benchmark files, 90 days of production traces, no held-out test set
Finding
The 40% figure comes from a curated set that overlaps training data. On production traces the advantage is 6–11% and only on two of six task types.
Confidence
Moderate
Investment impact
Win-rate assumptions built on category-wide accuracy leadership are not supported. Re-underwrite the differentiation premium and budget for an independent evaluation.
Risk: HIGH
Finding
No held-out evaluation set exists; regression testing is manual and run at release time by one engineer.
Evidence
Repository review, CI configuration, release checklist, two engineering interviews
Investment impact
Quality regressions reach customers before they are detected, which shows up as churn and support cost rather than as an engineering metric. It also blocks safe model upgrades.
Recommendation
Treat an automated eval suite with a frozen held-out set as a first-90-days deliverable and a gate on any model provider migration.

AI dependency map

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

Proprietary External
ProductUI · permissions · audit trailOWNEDAI Orchestratorrouting · tools · guardrails · evalsOWNEDModel Providerstwo frontier vendors · no tested fallbackEXTERNALRetrievalown chunking · tenant vector indexOWNEDCloudsingle-region managed infrastructureEXTERNALDatacustomer corpora · labelled outcomesOWNED

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

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
MOD
Economics
HIGH
CRIT
MOD
HIGH
Scalability & reliability
MOD
HIGH
LOW
MOD
Governance & security
CRIT
MOD
MOD
LOW
Product & adoption
LOW
MOD
HIGH
MOD
LOWMODHIGHCRIT

Scope

What we assess

Sixteen dimensions across the AI, technical and product tracks. Every engagement covers the AI track; the supporting tracks are scoped to the thesis.

  • AI

    Model architecture

    What produces the output, and how much of it the company controls.

  • AI

    Evaluation discipline

    Held-out sets, regression suites, and whether results are reproducible.

  • AI

    Training & fine-tuning

    What was trained, on what, and whether it measurably improved the task.

  • Technical

    Data pipeline

    Ingestion, chunking, freshness, tenancy isolation and lineage.

  • AI

    Data rights

    Licensing, customer terms and training-use permissions behind the corpus.

  • Technical

    Retrieval quality

    Relevance and degradation as tenant corpora grow.

  • AI

    Agent reliability

    Tool-call errors, escalation rates and observed autonomy.

  • AI

    Inference economics

    Cost per unit of value delivered, and margin at realistic usage.

  • Technical

    Latency & scalability

    Behaviour under load, latency budgets and failure modes.

  • AI

    Model dependency

    Supplier concentration, exit paths and tested fallbacks.

  • Technical

    Security posture

    Customer data in prompts, logs, embeddings and traces.

  • AI

    AI governance

    Policy, model change management, human review and audit trail.

  • Technical

    Observability

    Whether quality regressions are detected before customers find them.

  • Product

    Product adoption

    Marketed AI features versus features actually used.

  • Product

    Workflow fit

    Where the AI sits in the customer's process and what it displaces.

  • Technical

    Team & execution

    Depth of people who can actually change model behaviour.

Engagements

One primary practice, supported by two disciplines

AI due diligence leads. Technical diligence evaluates the systems beneath the AI. Product diligence evaluates whether the AI creates commercial value.

Who we work with

Anyone whose capital or operations depend on an AI claim being true.

We work alongside deal teams and enterprise buyers, and scope the work to the decision in front of them.

  • Venture & growth funds

    Pre-term-sheet AI validation on a deal clock.

  • Private equity

    Margin, dependency and value creation planning.

  • Family offices

    Independent read on an AI thesis before committing.

  • Strategic acquirers

    What is genuinely owned versus rented.

  • Corporate development

    Integration risk and model dependency at close.

  • Enterprise AI buyers

    Vendor claims, governance and lock-in before signing.

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 InvestmentSee a sample report