Skip to content
Probatio

Supporting discipline under AI Due Diligence

AI Technical Due Diligence

Evaluates the technology under the AI. Scoped to the systems that determine whether the AI can be operated, improved and scaled — not a generic software audit.

See the primary AI Due Diligence engagement

Scope

The systems that carry the AI

Architecture matters here only insofar as it constrains model quality, iteration speed, unit cost and reliability.

Build and run

  • AI architecture: orchestration, agent design, tool invocation, guardrails and fallback behaviour.
  • Data pipelines: ingestion, chunking, indexing, freshness, tenancy isolation and lineage.
  • Evaluation infrastructure: held-out sets, regression suites, CI integration, release gating.
  • Scalability: latency budgets, concurrency, queueing, degradation under load.

Control and risk

  • Security posture and customer data handling in prompts, logs, embeddings and traces.
  • Third-party model and infrastructure dependency, contracts and exit paths.
  • Engineering org depth: who can actually change model behaviour and how many of them there are.
  • Technical debt priced as remediation cost and roadmap delay, not as an aesthetic judgement.

Findings

How technical facts become deal facts

Risk: CRITICAL
Finding
Customer prompt and response payloads are written to a shared observability index with no tenant partitioning and a 24-month retention policy.
Evidence
Logging configuration, index schema, retention policy, two engineering interviews
Investment impact
Regulated buyers will fail this on their own vendor review, which caps the enterprise segment the growth case depends on and creates disclosure exposure at close.
Recommendation
Partition and shorten retention before signing; make written confirmation a closing condition.
Risk: HIGH
Finding
Model calls are issued inline from request handlers with no queue, no timeout budget and no circuit breaker.
Evidence
Code review of the orchestration layer, incident log, provider status correlation
Investment impact
Provider latency spikes surface as full product outages rather than degraded responses, so an upstream supplier's bad day becomes the company's churn event.
Recommendation
Fund resilience work in the first two quarters and treat historical uptime figures as provider-dependent.

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

Scope

What we assess

The engineering dimensions we assess underneath the AI — architecture, pipelines, evaluation infrastructure, scalability and security posture.

  • 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.

  • Technical

    Team & execution

    Depth of people who can actually change model behaviour.

Need the technical layer assessed alongside the AI?

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