A 30-minute AI Audit, scripted.
Most first-call AI conversations get wasted by twenty minutes of mutual throat-clearing about “AI strategy” before anybody says anything specific. The COO doesn’t know what they want. The vendor doesn’t know what the COO has. Both sides leave the call with vague enthusiasm and a follow-up scheduled.
This is the call I run instead. Thirty minutes, scripted, with a working hypothesis on the table by minute 22 that the COO can react to in real time. If the hypothesis is wrong, I find out before I leave the call. If it’s right, the next call is a scoped proposal instead of another discovery session. Either way, nobody loses the afternoon.
I’m publishing the script because I keep getting asked for it, and because the bar for first-call quality in the mid-market AI conversation is currently low enough that a vendor running this script will close above their weight. Steal it.
Minutes 0–4: Frame the call and skip the bio
I don’t introduce myself or the firm in any depth. The COO has my LinkedIn. They booked the call. The thirty seconds of “let me tell you about Truvisory®” is thirty seconds I’d rather spend on their business.
What I say, roughly: “I’ve got thirty minutes. The way I run these is, I ask you three questions about your P&L, two questions about your data, and by minute twenty-two I’ll have a working hypothesis on the table for you to push back on. If that sounds useful, let’s start. If you’d rather a different shape, tell me now.”
Nobody has ever asked for a different shape. The framing does two things: it sets expectations, and it pre-commits the call to producing an artifact. The COO is no longer evaluating whether the vendor seems smart — they’re holding the vendor to a clock and a deliverable.
Minutes 4–13: Three questions about the P&L
Not their AI strategy. Their P&L. The questions are the same every time, in the same order, because the order matters.
Question one (3 minutes): “Walk me through your top three operating costs as percentages of revenue, in rough terms.”
I am listening for two things. First, where the actual money goes — labor, COGS, customer acquisition, fulfillment, support, whatever the shape is. Second, whether the COO can answer this fluently. If they hesitate or hedge, they don’t run the business at the level of granularity that AI work requires, and the engagement is going to be slower than it should be. That’s not a deal-breaker, but it tells me to budget more time for the discovery phase.
Question two (3 minutes): “Of those, which one has grown fastest in the last 18 months, and why?”
This is the question that surfaces the operator pain, not the strategic ambition. Strategic ambition is “we want to be an AI-forward company.” Operator pain is “our customer support cost per ticket has gone up 40% because we’re handling more complex cases and we can’t hire fast enough.” One of these is a problem you can solve in a quarter. The other is a slide.
Question three (3 minutes): “If you could remove one specific cost or compress one specific workflow in your operation, what would it be — and what’s the dollar value of doing it?”
This is the question that gives me the scope candidate for the engagement. If the COO answers in dollars — “we spend $80k a month on the third-party document review service” or “we lose roughly 200 hours a week across the ops team to invoice reconciliation” — I have a real number to size against. If the COO can’t answer in dollars, the engagement isn’t ready to scope yet, and I should know that before I leave the call.
The three P&L questions take roughly nine minutes if the COO is fluent, twelve if they’re not. I keep a soft clock and steer back toward specifics if the conversation drifts.
Minutes 13–20: Two questions about the data
The P&L questions identify what’s worth fixing. The data questions identify whether it’s actually fixable inside the engagement window I sell against. Most AI engagements fail not because the AI doesn’t work, but because the data underlying the workflow isn’t structured in a way that lets the AI work. I’d rather find that out at minute fifteen than at week eight.
Question one (4 minutes): “For the workflow you’d most want to compress, where does the data live today? Walk me through the systems.”
I am listening for system names. Salesforce, NetSuite, custom internal tools, shared Google Drives, email threads, SharePoint. The number of systems matters less than the coherence of how they connect. A workflow that pulls data from three named systems with clean APIs is dramatically more tractable than a workflow that pulls data from one named system plus a lot of unstructured email and PDFs. I want to know which one I’m dealing with before I size anything.
Question two (3 minutes): “Who owns the data in those systems, and are there any access patterns I need to know about — PII, contractual restrictions on data movement, separate environments for sensitive customers?”
This is the question that surfaces the invisible scope — the data governance, security, or contractual constraints that turn a three-month engagement into a six-month engagement if you don’t catch them up front. Federal workloads have one version of this (controlled data, ATO requirements). Healthcare has another (HIPAA, business associate agreements). Financial services has another. Even commercial mid-market companies often have one or two named customers whose data has special handling requirements written into the contract, and the COO either knows or doesn’t.
The two data questions take roughly seven minutes if the COO has direct operational knowledge, ten if I have to ping-pong them toward more specific answers. By minute twenty, I have a P&L-grounded scope candidate and a data-grounded feasibility check.
Minutes 20–22: Quietly do the math
I shut up for about ninety seconds, sometimes more, and let the COO refill their water or take a glance at their phone. What I’m doing in the quiet is mental arithmetic: does the dollar value they named in P&L question three (the cost-of-the-pain number) justify an engagement of the size the data picture they described (data questions one and two) implies? Most of the time the answer is yes, but the shape of the engagement varies a lot based on the data picture.
The framework I’m running, fast, in my head: if the pain is north of $500k/year in directly attributable cost and the data lives in two or three named systems with clean APIs, the engagement is a fixed-scope six-to-ten-week build at $60k–$120k, working AI in production. If the pain is north of $500k/year but the data is in one named system plus a lot of unstructured mess, the engagement starts with a data-extraction and structuring phase that adds three to four weeks and roughly $30k. If the pain is below $300k/year, the engagement is hard to justify against the cost of doing it right, and I should say so on the call rather than pretend the math works.
The arithmetic isn’t novel. The discipline of doing it on the call, with the COO watching, is what most vendors skip — and the reason most first-call conversations don’t produce actionable scope.
Minutes 22–28: The working hypothesis
This is the part of the call most vendors don’t run, and it is the part that separates a useful first call from a forgettable one.
I say something close to this: “Based on what you’ve told me, here’s the working hypothesis. The fastest-growing operator cost you named is X. The specific workflow you’d most want to compress is Y, worth roughly $Z to you per year. The data lives in System A and System B, and the constraint that matters is W. I think there’s a fixed-scope engagement here that does [specific deliverable] in [specific timeframe] at [specific dollar range], and the way it would work is [two-sentence architectural sketch]. I am probably wrong about one or two of those, but I’d rather be wrong out loud right now than spend two weeks writing a proposal that misses what’s actually important. Where am I off?”
The pause that follows is the whole point of the call. The COO is now being asked to react to a specific proposition, not to evaluate a vendor in the abstract. Reactions are vastly more informative than evaluations. The COO will correct me on something — the dollar value is bigger or smaller than I named, the workflow is different in some specific way, the data lives somewhere I missed, there’s a constraint I didn’t catch. Every correction is a piece of scope I now have that I wouldn’t get from any other call structure.
Sometimes the hypothesis is mostly right and the call ends with the COO asking what the next step is. Sometimes it’s mostly wrong and the call ends with a much clearer picture of what the real engagement looks like, which is still a useful outcome. Either way, the working hypothesis has done the work of compressing a three-call discovery sequence into one call.
Minutes 28–30: Confirm the next step
If the hypothesis lands, the next step is a scoped proposal — typically one page, fixed scope, fixed price, fixed timeline — delivered within 48 hours. I commit to the timeline on the call so the COO knows what to expect.
If the hypothesis doesn’t land cleanly but the operator pain is real, the next step is a 60-minute working session with whoever in the COO’s org actually owns the data, to get the picture I missed. I commit to scheduling it on the call rather than handing it off to a coordinator afterward, because coordinator handoffs are where momentum dies in mid-market sales cycles.
If the hypothesis doesn’t land and the operator pain isn’t real — or the dollar value doesn’t support the engagement — I say so on the call. “I don’t think there’s a fit here that would be worth your money. Here’s what I’d watch for, here’s when to call me back.” This happens roughly one call in five, and I close more business with the four than I would by spending another six weeks chasing the one that wasn’t going to close anyway.
Why this works
Three reasons, briefly, because the meta-explanation is shorter than the script.
The COO’s time is the scarce resource. A thirty-minute call that produces an artifact is more valuable to them than a sixty-minute call that produces “let’s set up a follow-up.” Respecting the clock is itself a differentiation signal.
The working hypothesis is the trust move. Most vendors won’t commit to a specific shape on a first call because they’re afraid of being wrong in front of the buyer. Being wrong in front of the buyer is precisely what generates the corrections that turn a guess into a scope. The vendor that’s willing to be wrong out loud closes faster than the vendor that’s waiting to be sure.
The script forces operator-level specificity. P&L questions don’t let the COO retreat to strategic-ambition language. Data questions don’t let the conversation stay at the architecture-poster level. By minute twenty, both sides know whether there’s a real engagement here, which is the only thing the first call needed to determine.
Use it. Modify it. Send me what you learn.