Most people trying to land their first AI client are selling the wrong thing.<\/p>\n
They pitch automations. Chatbots. Agents. Solutions to problems that don’t exist yet \u2014 or at least, problems the business owner can’t see.<\/p>\n
So the pitch dies. Not because the business doesn’t need AI. Because nobody buys a build for a problem they can’t see.<\/p>\n
The fix is old. Consultancies have used it to get a foot in the door for decades. You don’t sell the build. You sell the diagnosis.<\/p>\n
It’s called an AI audit. These start around two grand for a three-day sprint and scale to four-week engagements for bigger companies. This post is the full playbook: the questions to ask, how to map processes, how to score opportunities so the client signs off \u2014 and how the audit converts into the much larger build engagement that usually follows.<\/p>\n
It is not walking into a company and asking “have you tried Claude or ChatGPT?” That’s not an audit. That’s a coffee chat.<\/p>\n
A real audit is three things:<\/p>\n
The company gets clarity. You get paid to write the roadmap. And the person who wrote the roadmap is the obvious person to build what’s on it.<\/p>\n
The engine of an audit is interviews. But before you interview anyone, do some preliminary research. Two reasons: you arrive prepared, and you catch dealbreakers early.<\/p>\n
Collect this before day one \u2014 most of it comes from one email to the owner:<\/p>\n
Walking into your first interview already knowing the tool stack and the org chart separates you from every other “AI consultant” they’ve talked to. Preparation is the pitch.<\/p>\n
Fifteen to thirty minutes each. At a minimum you want two people per business function: the person who actually does the work, and the manager responsible for it.<\/p>\n
Managers tell you how the process is supposed to work. Operators tell you how it actually works. The gap between those two answers is where the money is.<\/p>\n
Here’s the question bank:<\/p>\n
You’re hunting four signals: repetition, copy and paste, waiting on approvals, and dread. That’s what automation solves.<\/p>\n
Record every interview \u2014 with permission. I use Tactiq<\/a> as the meeting note taker, so every interview produces a transcript automatically. Feed those transcripts into Claude and pull out the process steps, timings, and bottlenecks. Your notes become process maps without you re-typing a word.<\/p>\n One flow diagram per process. Order-to-invoice. Lead-to-proposal. Ticket-to-resolution.<\/p>\n I use Mermaid.ai because I can generate the flowchart straight from the interview transcripts with a prompt, then clean it up in the editor.<\/p>\n For every step in the flow, tag three things:<\/p>\n Then mark the ugly stuff in red. Manual data entry. Copy-paste between systems. Slow approvals. Low response rates. When the owner watches their lead process cross six tools and four people before a proposal goes out, you don’t have to convince them anything is broken. The picture does it.<\/p>\n <\/p>\n Now attach a fix to every red step. And be specific. Name the tool and describe how it improves the process. Vague recommendations don’t get budget.<\/p>\n You’ll come out of the last step with a list of automations. You can’t recommend all of them. So you score, prioritise, and estimate.<\/p>\n Two axes. Horizontal: cost to implement \u2014 build time, tooling, complexity. Vertical: business impact \u2014 hours saved, errors reduced, revenue unlocked. Four quadrants:<\/p>\n Then back every dot on the chart with real numbers, because a dot doesn’t get budget approved. Time and cost do.<\/p>\n The math is simple. Hours saved per month, times loaded hourly rate, versus build cost plus monthly running cost. If a task eats 20 hours a month at $50 an hour, that’s $12k a year. If it costs $3k to build and $100 a month to run, payback lands in about three months.<\/p>\n One more gate before anything makes the roadmap: feasibility. Any compliance issues \u2014 personal data, financial data? Does it need a human in the loop? Some high-impact items fail here. And showing the client what NOT to automate builds more trust than any pitch.<\/p>\n Everything so far becomes one document. This is what they’re paying for.<\/p>\n <\/p>\n And here’s one of the best kept secrets to keeping a client happy. It isn’t the build. It’s accurate estimation.<\/p>\n Every horror story about agencies and freelancers is the same story: “they said two weeks, it took two months.” That’s not a build problem. That’s an estimation and communication problem.<\/p>\n So don’t use single numbers. Single numbers are promises you can’t keep. Use PERT estimation instead. For every task, give three estimates \u2014 best case, most likely, worst case \u2014 and blend them into one weighted number: (best + 4\u00d7likely + worst) \u00f7 6.<\/p>\n This does two things at once. It protects you: when your quote already includes the worst case, an overrun isn’t a broken promise \u2014 it’s inside the range you gave them. And it means no surprises for the client: they saw the worst case before they signed. Happy clients aren’t the ones where everything went perfectly. They’re the ones who were never surprised.<\/p>\n Never just email the report. Present it live. Walk them through it. Land on the quick wins.<\/p>\n They now hold the full roadmap. They could hand it to anyone \u2014 but you’re the obvious choice. No pressure, no pitch deck. The audit does the selling.<\/p>\n Free download<\/p>\n The Example AI Audit Report<\/p>\n A complete 12-page audit deliverable for a marketing agency \u2014 see the standard before you write your own.<\/p>\nStep 3: Map the processes<\/h2>\n
\n
Step 4: Attach solutions \u2014 with specific tools<\/h2>\n
\n
Step 5: Score it \u2014 the cost vs impact quadrant<\/h2>\n
\n
Step 6: The deliverable<\/h2>\n
\n
Step 7: Present live and convert<\/h2>\n
Three mistakes that kill audits<\/h2>\n
\n
Get the full toolkit<\/h2>\n