The five questions every AI audit answers

Key takeaways

• Every audit answers five questions, in order: where is the time going, where is the money going, what is fragile, what is repeatable, what is the next dollar worth automating.
• The order matters. Most operators want to start at question five; doing so wastes the engagement.
• The audit deliverables — operations map, opportunity matrix, 12-month roadmap with hard numbers — belong to the client whether they hire us for the build or not.
• The hidden value of the audit is not the answers. It is that the operator can now see their own operation with hard numbers attached, often for the first time.

The frame

The two-week audit we run with every new client — the first phase of [our audit-build-operate model](/blog/audit-build-operate-model) — is structured around five questions. They look obvious on a list. They are not obvious in practice — partly because most operators have never been asked them in this order, and partly because the answers to each one require evidence the operator usually doesn't have at hand.

We use the same five every time. We have run this loop enough now to know that the order matters. Doing them out of order — particularly skipping ahead to question five — is how most "AI strategy" engagements waste their first month.

Here they are, in the order we ask them.

1. Where is the time going?

Not "what does your team do." That answer is always wrong. We mean: in a given week, what are the actual hours spent on what?

Almost no operator has this answer at the resolution we need it. The usual data is "Maria handles tenant comms, Ali handles maintenance, I handle owners." That's a role description. What we want is closer to: in the last week, Maria spent 11 hours on tenant comms, of which 6 were lease-renewal coordination, 3 were inquiry triage, and 2 were the same recurring escalation about a parking dispute that the system has not resolved in four months.

We don't get this by asking. We get it by sitting next to the team for two days and watching. Counting. Timing things with a stopwatch. It looks slow. It is genuinely the cheapest hour you'll spend on this whole engagement. The output of question one is a calendar, not a category list, and the calendar is what every subsequent question gets evaluated against.

2. Where is the money going?

This sounds like a finance question. It is mostly a question about leverage.

Time spent and money spent are not the same thing. A small property management firm might have its highest-paid person — the owner — spending eight hours a week on owner reporting that any decent operator could do in three. The opportunity cost of those five hours, valued at the marginal hour of the constrained person, is enormous. The cash cost is invisible because she doesn't pay herself by the hour.

So the money question is really: what is each hour of each role costing the business in opportunity, not just in payroll? And specifically: which roles have the highest opportunity cost per hour, and what work are they doing that they shouldn't be?

For most of the operators we work with, the answer is some version of "the constrained person is spending 30–50% of their week on workflows that would not actually require them if those workflows existed in a different shape." The audit is, in part, a tool for surfacing how much of the founder's or operator's time is being eaten by work they wouldn't have hired themselves to do.

3. What is fragile?

Every operation has a small number of workflows that work only because one specific person remembers how to do them. Not because there's a documented process. Not because there's a system. Because Maria has been doing it for three years and Maria knows.

The audit identifies these. Bluntly. They go on a list.

The reason they matter is that fragile workflows are where deployment goes badly when you try to automate. If the original process exists only as institutional memory in one person's head, you cannot build a reliable system around it until you've actually extracted that memory into a written process. The model can't infer it. The integrations can't compensate for it. You will ship something that looks correct in week four and breaks in week eleven when the workflow encounters an edge case that exists only inside Maria's head.

So the fragile list has two uses. First, it tells us what to write down before we automate. Second, it tells us what to not automate first — because automating fragile workflows without first documenting them is how you ship systems that look successful and quietly fail.

4. What is repeatable?

This is the opposite of fragile. We're looking for workflows where the input shape is consistent, the decision rules are stable, the output goes to a predictable place, and the failure modes are bounded.

Repeatable doesn't mean "the model can handle it perfectly." It means "the workflow is structured enough that you can put a system around it and measure whether the system is working." The maintenance ticket flow is usually repeatable. The lease renewal cadence is usually repeatable. The "occasionally an owner emails the operator at 11pm with a request that requires judgment" workflow is not.

The repeatable list is where the build is going to live. The fragile list is where the system has to be careful. Mixing these up is a common failure mode for vendors who pitch "let's automate everything" — most of the everything isn't ready to be automated yet.

5. What is the next dollar worth automating?

This is the question every operator wants to start with, and it is the question that can only be answered correctly after the first four.

The shape of a good answer here is not a list of cool things you could build. It is a ranked roadmap with hard numbers attached to each item — what it would cost, what it would save, what it depends on, and how long it would take. The ranking is by ROI on the constrained resource, not by technical sexiness. Sometimes the highest-ROI build is the most boring one. (Picking the right tool for each build — Make, n8n, or custom Claude — is a different decision, covered in [our tooling decision rule](/blog/make-vs-n8n-vs-custom-claude).)

The mistake we see most often is operators starting at question five — usually because a vendor has shown them a flashy demo. The vendor's demo is always at the technological frontier, because that's what demos well. The actual highest-ROI build, for almost every operator, is at the technological back — a Postgres database, a clear schema, a structured Slack notification, three carefully written prompts, and an operator dashboard. The exciting agentic system the vendor pitched is often a year and a half early for the operation it's being pitched into.

Answering question five well, in practice, requires the discipline to recommend the boring thing. The audit's job is to give the operator the evidence base to know which thing is actually highest-ROI, regardless of how it demos.

What the audit deliverable looks like

At the end of the two weeks, the client has three documents:

1. An operations map. The calendar from question one, drawn out. Where each role's hours go in a typical week, by sub-category.
2. An opportunity matrix. Workflows on one axis, scored on time-cost, fragility, repeatability, and a model-fit estimate. The fragile list and the repeatable list, side by side, with the math.
3. A 12-month roadmap. The answer to question five, ranked. Each item with an estimated build cost, an estimated time-savings, a dependency map, and a recommended sequence. Quarter-by-quarter, so the operator can see what gets shipped first, what comes next, and what is deliberately deferred to a v2 backlog.

These three documents belong to the client. If they hire us for the build, the roadmap becomes our scope. If they don't, the roadmap is still theirs to take to another firm — or to use as a hiring plan, or to refer back to in a year when the operation has changed enough to warrant another look. We have written audits that clients took elsewhere. That is fine. The discipline of writing the document as if the client might walk away is what keeps it honest.

The hidden value

The thing operators tell us most often, three or four weeks after we deliver the audit, is some version of: "I've never seen my business written down like this before." That is the real deliverable. Not the roadmap. Not the ROI numbers. The fact that the operator can now look at their own operation with hard numbers attached and make decisions from evidence rather than from instinct.

The roadmap is the artifact. The clarity is the product. Five questions, in order, asked seriously over two weeks, are usually enough to produce it.