Examiners consistently ask four things, and all four are capture questions
Who approved this, on what information, when, and did the control operate throughout the period. Each is instant if captured in workflow and expensive if reconstructed.
The decisions were made properly. The evidence just was not captured at the time, so it has to be reconstructed.
Ask a compliance officer what examination preparation involves and you will hear about a project: several people, several weeks, pulled off their day jobs to assemble evidence that decisions were made properly.
This paper argues that the preparation effort is a symptom of when evidence is captured rather than how hard the team works, identifies the four control points where capture determines everything downstream, and examines what AI adoption adds to the same problem.
If you read nothing else, read these. The analysis that follows sets out the evidence for each.
Who approved this, on what information, when, and did the control operate throughout the period. Each is instant if captured in workflow and expensive if reconstructed.
Capturing evidence in the flow of work requires the workflow to exist in a system. Where it does not, no amount of documentation discipline compensates.
An exception granted without an owner and an expiry is indistinguishable from a policy after eighteen months.
What an assistant reached and what was sent in a prompt are now examination-relevant questions.
Approvals recorded in a system with a timestamp and the information the approver saw, rather than in an email thread. Access reviews performed on a schedule with the outcome recorded, including where access was left unchanged and why. Exceptions logged when granted, with an owner and an expiry. Screening and disposition results attached to the relationship record automatically rather than filed separately.
Each of these is answerable instantly if it happened in a workflow and expensive if it happened in a mailbox.
Compliance teams already know all of the above. What prevents it is that significant processes still run on email and shared drives, and capturing evidence in the flow of work requires the workflow to exist in a system first.
The fix is therefore to move those processes into a workflow with evidence as a by-product — not to ask people to document more carefully. Documentation discipline degrades under pressure, which is exactly when the evidence matters most.
The second-order benefit is the one that rarely appears in a business case: the compliance function stops spending its expertise on assembly and starts spending it on the questions that require judgement. Nobody counts what a compliance officer could have been doing instead.
Institutions deploying assistants and agents face a new version of the same question: what did the assistant reach, what was sent in a prompt, and can that be evidenced eighteen months later.
Microsoft's data security posture management for AI covers this specifically, which is a signal about how common the gap is. Institutions with mature evidence capture will find the extension straightforward. Institutions without it are adding a second unevidenced surface to an existing problem, and the examination question will be the same one they already cannot answer.
Every paper in this series ends with a framework you can run internally. We would rather you used it and reached your own conclusion than took ours on trust.
Five stages. The first is a mapping exercise and the last is what proves it.
Identify which regulated processes currently run on email and shared drives.
Put those processes into a workflow where the evidence is a by-product.
Give every exception an owner and an expiry, and review before renewal.
Cover AI interactions with the same discipline as the rest of the estate.
Simulate an examination request and time the responses. The findings are the plan.
The same argument lands differently across an executive team. These are the three versions worth separating.
Microsoft's own documentation for the product behaviour described above. We would rather you verified the basis than accepted our summary of it.
On these references: each entry names a Microsoft Learn article or documentation area by title, because deep links change while titles are stable. Searching the title on learn.microsoft.com will reach the current version. Where we have cited a figure or a product behaviour, it is Microsoft's statement rather than ours; where we have given a number of our own it is labelled as such in the text.
We will review one control process end to end and show you what capturing the evidence in the flow of work would look like against your examination expectations.
Consolidation projects are usually justified on efficiency and delivered on hope. This paper examines where the value actually is and what the honest cost looks like.
Your core system knows about a checking account, a mortgage and a commercial loan. It does not know they belong to the same household.
Describe the situation in your own words.