Can you identify every material AI-assisted workflow in production?

To identify material AI-assisted work, map where AI is used in live processes and record what each use can influence. Start with workflows where an error could affect a customer, money, coverage, a claim or regulated work. Do not assume that every use of AI is equally consequential, or that an inventory by itself proves a workflow is safe.

Build the inventory around work, not software names

A list of software licences will miss AI features embedded in other systems, staff use of general-purpose assistants and automations that pass AI outputs into later steps. Trace actual work from the point where information enters to the point where someone or something acts on it.

Ask process owners, operational teams and technology suppliers where AI is used. Check live systems, integrations, team procedures and recurring workarounds. Include pilots or limited releases if their outputs can reach a customer, case file or business decision.

Record what changes the decision

For each workflow, capture enough information to decide how closely it needs to be managed:

• Purpose and process: What work does the AI support, and where does it sit in the process?
• Decision or action affected: Does it draft, summarise, classify, recommend, route, update a record or trigger an action?
• People affected: Could the output affect customers, claimants, brokers, staff or another party?
• Owner: Who is accountable for the process and for reviewing changes to the AI use?
• Inputs and outputs: What information or documents does it use, and where does its output go?
• Human review: Who checks the output, at what point, and what can they change or stop?
• Failure consequences: What could happen if the output is wrong, incomplete, delayed or acted on without review?
• Controls and evidence: What checks, approvals, restrictions and records are in place?

Use the answers to prioritise review. A drafting aid whose text is checked before use may warrant a different level of attention from a workflow that can influence a claim, coverage, payment or customer communication. Consider the actual process and possible consequences, rather than relying on the tool’s label or a supplier’s risk rating.

Check whether the record reflects production

For higher-priority workflows, compare the inventory with a sample of real cases or transactions. Check whether the recorded purpose and owner are accurate, whether the output goes where expected, and whether review and escalation happen in practice.

Keep evidence proportionate to the workflow. Useful records may include the workflow owner and scope, relevant configuration or supplier information, review steps, approval or exception records, and examples showing where an output was corrected or stopped. Decide what is necessary to understand how the process operated, and avoid collecting sensitive information without a clear reason.

Assign an owner to update the inventory when a process, model, prompt, data source, integration or level of human review changes. If ownership or evidence is unclear, record the gap and decide who will resolve it before relying on the workflow for more consequential work.

Scope of the source

The insurance-sector article informing this question was published by Insurance Thought Leadership on 20 August 2026. It is industry commentary, not UK legal or regulatory guidance, and does not establish a legal test for materiality. The inventory and checks above are practical recommendations, not a finding that a particular use is safe, harmful or compliant.

Discuss your AI decision

If you are unsure which workflows to prioritise, bring a specific AI-assisted process, the work it affects, possible consequences and the controls already in place to a Pattrn Data discovery call. The discussion can focus on a proportionate next step; it does not promise a particular outcome.