Why we audit an operation before we build an agent
The fastest way to waste an agent project is to automate the wrong step. The audit finds where agents should operate, where they should not, and what controls they need before anything is built.
Every operation has a version of itself that lives in a process document, and a version that actually runs. The second one is where the time goes: the copying between systems, the chasing of approvals, the exceptions that come back every week and the knowledge that sits with two people. An agent built against the first version automates a process that does not exist.
That is why we start with an audit.
What we look at
We follow the work as it happens, through the systems and the teams, across one business function. We look at eight things: the workflows, the systems each one touches, where people decide and what they look at, the exceptions and how they are resolved, the manual interventions that keep things moving, the data each step depends on, the controls that exist today, and where the effort goes.
Every step gets a kind
The audit does not assume that everything should become an agent. Each step is classified as one of three kinds:
- Rule-based. A fixed rule can decide it every time, so conventional automation should run it. It is cheaper and easier to trust than an agent.
- Agentic. It needs context gathered from several places, or a judgment that varies with the case, inside limits we can define. This is where agents create leverage.
- Human. Accountability, risk or judgment that is not yet bounded means a person decides. Agents may gather the evidence.
Knowing which steps should not be agentic is half the value.
What you get
The result is not another AI strategy deck. It is a plan you can start on: a map of how the function works today, what that map shows about where effort and risk sit, each workflow classified and ranked, the integrations and controls each one needs, the level of autonomy each should start at, and a sequenced roadmap.
After the audit
If you decide to deploy, the plan becomes the first stages of the deployment: a sandbox, then shadow mode beside your team, then supervised production. If the audit finds a workflow should stay with fixed rules or with people, it says so.