Mapping AI use

Mapping AI use: what a discovery audit can establish

A client uses an assistant, an automation tool and several model services. Before deciding what needs governance, your team needs to explain what they are used for, who looks after each use and what evidence supports that account.

A useful inventory goes beyond a list of tools. It separates observed facts, reported uses, open questions and areas that have not been examined. It gives the integrator and client a basis for deciding what to do next, without implying that actions are already controlled.

In brief

The question and the useful result.

Question
What can we actually establish about the client’s AI use from the sources available?
Useful result
An inventory record that links findings to sources, distinguishes reported ownership from confirmed responsibility and makes unknowns visible.

Mapping AI use

Agree what may be examined

Start with a specific purpose, such as understanding AI use in the handling of supplier enquiries. Then agree which sources the organisation authorises you to consult and the period covered.

Interviews with responsible teams, workflow documentation, accessible configurations and authorised logs can provide complementary evidence. Their availability needs to be checked. Access to a directory does not, by itself, provide access to conversations or to actions taken by the tools listed in it.

The inventory concerns the process and its AI use. It does not require a profile of individual employees’ working habits.

Mapping AI use

Create a record for each use and state what the evidence establishes

The same application might draft a response and trigger its sending. Record those uses separately when they involve different purposes or actions.

FieldWhat to record
Use and purposeThe work involved, its intended outcome and the systems concerned.
Authorised sourceThe document, interview or log consulted, its date and the period it covers.
Established factWhat the source actually shows, with a reference that lets someone find it again.
ResponsibilitiesBusiness and technical contacts; state whether their roles are reported, confirmed with them or still unassigned.
Data and actionsThe information used, its known recipients and the actions planned or observed.
Unknowns and limitsMissing information, conflicting accounts and systems outside scope.
Follow-upThe question to resolve, who will check it and what evidence is needed.

An invoice can establish that a service was purchased without proving that it is currently used. A configuration describes an intended process; it does not establish that an action was executed. Preserve these distinctions in the record.

Mapping AI use

Example: an assistant that replies to suppliers

In this illustrative example, an interview describes an assistant that drafts replies. An authorised configuration also contains a sending step. The available log records receipt of a sending request, but provides no confirmation that sending completed.

The inventory can therefore state: reply drafting reported by the team; a sending step present in the configuration examined; a request received for the operation observed. Whether sending completed, and which authorisation applied, remain to be checked.

This does not establish that all messages are sent automatically or that the reported human approval applies to every send. A request, an authorisation, an attempt, an acknowledgement and a confirmed effect are different records. The inventory needs to preserve the meaning of the evidence available.

This scenario illustrates a method. It does not assume an available messaging integration in DUBSAR.

Mapping AI use

Use the inventory to choose the next question

At the end of the review, ask the relevant teams to confirm responsibilities and open questions. Date the resulting view: a change of workflow, provider or role can make some findings out of date.

The next step might be to obtain another source, clarify a responsibility or examine a particular action. It might also be to maintain an inventory that is sufficient for the identified need. The presence of AI does not automatically justify an additional control.

The report does not enforce permissions in the systems concerned. To assess that transition, read the guide to the conditions for ongoing governance. To clarify recipients and copies of data, examine where an AI mission’s data goes.

Mapping AI use

Frequently asked questions

Does absence from the inventory mean that a use does not exist?

No. It may reflect the sources, period or scope examined. State those limits before drawing conclusions about coverage.

Is a log enough to establish that an action was authorised?

It may show an attempt or an outcome, depending on its contents. Establishing authorisation requires the rule or decision that applied to the operation. An observed event does not replace that evidence.

Does DUBSAR offer this audit?

The Flash Audit is a previous DUBSAR offer. This page retains its useful methods and their limits; it does not present an audit service currently on offer.

Mapping AI use

Current DUBSAR status

DUBSAR is developing a foundation to connect durable mission state, authorisations, human decisions and evidence alongside existing tools. The inventory method described here assumes no automatic discovery capability in the product.

Selected workflows have been tested; full B2B integration and validation are still underway. Pilots are not currently open.