Why AI is not the full answer for assurance reviews

In this article we will discuss why AI is not the full answer for assurance reviews, why experts are still needed and where AI can play a role in the assurance process.

What this looks like in practice

Assurance reviews are important since they check intent versus outcome. A delivery can meet every written acceptance criteria and still fail its actual purpose because the requirements were wrong.

Where AI checks against what was documented; a specialist checks against what was meant and what the business actually needs.

Materiality is a judgement call. Which gaps matter? A missing edge case might be trivial and have serious consequences. Weighing that requires understanding of the context and consequences something AI doesn’t hold.

For example Roadmap reviews done by Dquadrant for Cardano are political and organizational, not just technical. Real assurance means probing whether teams are quietly cutting corners, whether “done” is genuinely done or optimistically reported, whether cross-team dependencies were truly resolved or just deferred. That takes reading between the lines, asking questions and healthy skepticism about self-reported status, none of which AI does realiable.

A strategic risk needs judgement of a human, an independent third party judging several risk scenarios.

Accountability and sign-off carries weight. In case of a roadmap item assurance culminates in someone attesting “this is fit for release / this roadmap item is on track and the risks are acceptable or not” That attestation has organizational and sometimes regulatory consequenses a model can’t own.

AI can be gamed by inputs. If status reports, documentation and code deliveries are optimistic or incomplete (which is often the case). AI faithfully reflects distortion. A specialist knows to distrucs the artifact and verifies independently.

Where AI does really help: it can cross check deliveries against stated requirements and acceptance criteria, flag gaps between what was promised and shipped, track dependencies and slippage across the roadmap, summarize status scattered documents, spot inconsistencies in documentation and surface metrics trends (velocity, defect rates, coverage) much faster than manual collation. All needs to line up with what is promised.

Where AI makes a real difference is by doing the heavy lifting part: evidence-gathering, traceability, gap detection and status synthesis so the review starts from a complete organized picture. Specialists spend their time where it really counts, interrogating whether the delivery genuinely achieves its purpose, whether the reported status is real, what the material risks are and whether to sign-off. AI builds the case and the assurance specialist renders the verdict.

At Dquadrant we make use of AI in our assurance processes exactly there where it matters most reducing time being spend on the wrong tasks, this time can be spend elsewhere where it matters most.