How to Audit a Company for AI: Map the Decisions Before the Tools

The wrong audit starts with a list of AI use cases

A generic list can produce dozens of plausible ideas and still miss the work that matters. The audit should begin with the business model: how demand arrives, how value is created, where margins are lost, what customers wait for, and which decisions consume scarce human attention.

From there, follow real workflows rather than documented ideals. Observe the copy-paste work, waiting, repeated questions, informal approvals, manual reports, exceptions, and places where one person is the only integration between systems.

Audit the company that actually operates, not the SOP that says it operates.

Map each workflow as a chain of decisions

For every important workflow, capture the trigger, required context, decision, owner, action, system of record, waiting state, exception, and completion evidence. This turns a vague process into an operating graph.

AI opportunities appear where the graph is slow, repetitive, context-heavy, or inconsistently owned. Some opportunities need an agent. Others need deterministic automation, better integration, a redesigned human process, or simply cleaner data.

Score the opportunity by operating reality

Economic value is necessary but not sufficient. A workflow with huge theoretical upside may be a poor first deployment if the source data is untrusted, the system cannot be reached reliably, or the consequence of a wrong action is high.

7SVN scores opportunities across value, frequency, data readiness, system access, decision clarity, execution risk, adoption burden, and time to measurable proof.

Design the smallest complete loop

A useful pilot should reach a business outcome rather than stop at an AI output. A lead-response pilot, for example, should not end after drafting a message. It should capture the lead, load relevant context, qualify, send or escalate, update CRM, schedule the next action, and measure whether the opportunity advanced.

The smallest complete loop creates evidence the business can trust and exposes the integration and governance work that a shallow prototype hides.

A complete small loop is more valuable than a broad incomplete transformation.

Audit human authority at the same time

Every recommended use case should identify what the AI may observe, draft, execute, and never decide alone. This is especially important when workflows involve money, safety, customer commitments, regulated information, or professional judgment.

The human operating model is part of the design: who approves, who takes over, what context they receive, and how quickly they must respond all affect whether the automation succeeds.

What the audit should hand leadership

The final artifact should be a decision portfolio, not a brainstorming deck. Each candidate should include the current baseline, business case, target operating loop, systems affected, data sources, authority model, implementation dependencies, measures, and accountable sponsor.

That portfolio becomes a sequence: what to instrument now, what to automate next, what to defer, and what not to automate at all.

Frequently asked questions