Identify and collect
Build a risk register from evidence people can trace
Design the form, data checks and approval route before you ask AI to populate the register.
If a branch outage arrives as three separate messages, the incident log needs one checked event with three source references. Use that event as evidence for the forward-looking risk-register entry. Use AI to help build the intake app and flag likely duplicates. Keep the original submissions so a reviewer can correct a poor match.
Decide what people need to submit
Start with the decision the data will support. For a risk register entry, collect the event, affected service, owner and evidence link. For a monitoring reading, capture the measure, value, unit, denominator where relevant, collection time and source system. Use a common event ID so the analyst can connect related records. Keep access specific to each role and use only data your organisation has approved for the tool. HM Treasury, The Orange Book NIST, Generative Artificial Intelligence Profile
Use AI to produce the field dictionary, a prototype form and test cases with practice data. Ask staff to try a missing field, an invalid number and a duplicate report. The risk owner approves which fields are mandatory and where uncertain records go.
Work a duplicate without losing the evidence
| Submitted record | Relevant data | Reviewer action |
|---|---|---|
| Ticket 274 | Branch A unavailable at 09:10 | Keep the source timestamp |
| Service desk note 81 | Branch A calls the desk at 09:22 | Keep the report time separately |
| Register row 42 | Branch A outage attributed to power | Ask for the evidence behind the cause |
The first two records may describe one incident. The third makes a claim about its cause. Connect the records, keep both times, and mark the cause as unverified until there is evidence. The NIST generative AI profile warns about “confidently stated but erroneous or false content”. Decide with the data owner whether the register should count an outage, a branch affected by that outage, or both in separate fields. NIST, Generative Artificial Intelligence Profile
Copy this prompt
Act as a risk data designer. From the approved sample records and data definitions below, propose a short intake form and a field dictionary for the risk register and monitoring log. Specify required fields, units, source IDs and ownership. Design tests for missing values, apparent duplicates and conflicting sources. Suggest matches for reviewer approval. Do not merge or modify the source records.
Check the first export
Reconcile the number of rows received, held and accepted. Compare a sample back to the source, including every high-impact entry in the first test batch. Ask the reviewer to retrieve an original submission using its ID and correct a bad match. If that fails, fix the audit trail before relying on the dashboard.