The $25K Question That Decided the Deal

A buyer asked me the other week how he was supposed to sell my proposal internally.

The number was $25,000. The work was a referral-network activation and target-prioritization engagement for a PE-backed provider platform: take their existing provider network, referral history, and CRM context, weigh which relationships actually drive deals, rank the acquirable practices in the markets that connectivity points to, and trace the warm path from their own people back to each target. The objection was not the price or the scope.

It was that I was using AI to build it.

"How do I sell this internally if you're going to be using AI?"

I told him: you are not paying for AI. AI is the tool. You are paying for the judgment around it. I have spent most of my career on the buy side of healthcare M&A, in-house at healthcare platforms. I know how to stitch a provider roster, a CRM export, licensing data, and alumni records into one validated picture. I know how to tell a relationship that drives deals from one that just looks good on an org chart. I know which targets are ownable and which are noise, and what I would do with all of it if I were sitting in your seat. The scoring logic and the source-stitching are mine, and they stay mine. The AI makes it fast. The judgment is the product.

That conversation decided the engagement.

To see why that answer mattered, you have to understand what this platform was actually trying to fix.

It runs a partnership model. It buys into practices, the owner keeps real equity, and they co-invest in future deals together. And by a wide margin, its best deals had always come from one channel: its own providers referring peers they already respected. Those introductions came in warm, pre-qualified, and far more likely to close than anything sourced cold.

The problem was that the channel ran entirely on a couple of people's personal relationships. It had never been mapped, scored, or made repeatable. So to fill the gap, the platform had been leaning on a buy-side advisor and a brokered-list vendor, and most of what came back did not fit the model. The targets looked fine in a database and were wrong in practice: owners who wanted a retirement exit instead of a partnership, practices that matched on size and missed on everything that mattered. A lot of motion, a low hit rate, and a team spending its weeks screening deals that were never going to convert.

So the engagement was not "go find me targets." It was "systematize the channel that already works." Take the referral network that lives in two people's heads, make it explicit, score which relationships actually carry weight, and turn it into a ranked and mapped asset the team can run without those two people in the room. That is judgment work. It is exactly what a database query cannot do, and exactly why the AI question missed the point.

Start with the market context. Buyers are not skeptical of AI in M&A. Bain's 2026 M&A report, surveying 300+ executives, found that AI adoption in M&A more than doubled in 2025 and 45% of practitioners now use AI tools. The skepticism has shifted one layer up. Buyers know AI works. What they cannot defend internally is paying a premium for something a $20 ChatGPT seat appears to do.

So "are you using AI?" is the wrong question. It presumes AI is the deliverable, and if AI is the deliverable, the right price is whatever an API call costs. Nobody pays a $25,000 premium for an API call.

The right question is: what is the judgment loop wrapped around the AI?

That is what determines whether the output is a usable, ranked target universe with the warm paths traced, or two hundred names that need to be re-verified before anyone can pick up a phone.

I have been on both sides of that gap. I have seen the model return confident-looking data that turns out to be three years out of date, mid-specialty wrong, or attached to a practice that closed in 2024. The model never flagged any of it. The judgment did.

The work that goes into a $25,000 engagement is not the work the buyer sees. The buyer sees a databook and a summary deck. The work that produced it is a hundred decisions: which source to trust when two disagree, which referral relationship actually carries weight, which name to drop because the founder is not really approaching retirement, which name to add because the signal says he is. Knowing how to direct the AI is part of that judgment. Knowing the market and the relationships is the rest of it.

There are four places at the bottom of the sourcing funnel where the judgment loop is the entire engagement.

Physician retention risk at the human level. AI can flag key-person dependency from provider count and revenue concentration. It cannot tell you whether the 58-year-old founder is going to stay for the earn-out or is mentally checked out. I covered this in detail in an earlier piece on the timing signals that actually matter. The human signal is often what closes or kills the deal, and it never shows up in a database.

Deal-model fit. AI can match a target to the buy box on paper: size, age, specialty, geography. It cannot tell you whether the owner wants the kind of deal you actually do. A platform that partners with owners who keep equity is looking for something completely different from one that buys founders out at retirement. Two practices can look identical in a database and be opposite answers. Knowing which is which is the difference between a pipeline and a list of names that will never convert. It is also the exact gap that sinks most brokered lists.

Referral relationship durability. CMS shared-patient data shows billing patterns that imply referral networks. It cannot tell you that Dr. Smith sends patients to the target because they trained together at Cleveland Clinic, and that the relationship will not transfer to a new owner. In provider M&A, the referral pathway is often the business. Lose it and the multiple was wrong.

IC politics. Your investment committee does not just want to know if a target is attractive. They want to know if it fits the exit narrative the GP has been telling LPs for the last eighteen months. AI cannot answer that question. The person who can answer it is somebody who has sat in that room before.

This is not a knock on AI tools. PwC's Deals team has been public about using Harvey AI across the deal lifecycle for exactly the reason buyers should be paying for it: their Deals AI Lead described the shift as moving from "spending all of the time working through a small sub-set of contracts" to focusing diligence on the real risks and upsides. In a follow-up case study, PwC framed it as "where a team member might previously have spent a day pulling key facts and context on a target, they can now get to an informed perspective in an hour."

Same principle. The tool handles assembly. The practitioner handles judgment. The premium is on the second part.

If you are weighing how to scope an AI-enabled engagement internally, the move is not to ignore the question. It is to reframe it. Two practical steps.

First, ask your prospective partner what their judgment loop looks like before they show you the output. The answer should be specific. "How do you know a name on the list is still accurate, that the founder hasn't already sold or the practice hasn't closed?" "How do you tell a referral relationship that drives deals from one that just looks good on paper?" "What is your read on the difference between a founder testing the market and a founder ready to transact?" If the answer is vague, the deliverable will be vague.

Second, ask to see a sample before you scope anything. Not a custom build done on spec, a redacted example of work already delivered for a comparable platform, with names and identifying detail masked. A provider who does this work has one ready. You see the judgment loop in the artifact rather than the proposal. If the sample answers a question you could not answer yesterday, the paid engagement is a different conversation. If it does not, you have spent ten minutes and learned what the work is not.

The alternative most platforms reach for is a full-time BD hire to build this in-house. I argued in an earlier piece on the real cost of that hire that it is usually the wrong first move: one seat, one domain, a six-month ramp, and the pipeline does not move while you pay full freight. A scoped outside engagement lets you see the judgment before you commit to the headcount.

The buyer who asked me how to justify the proposal eventually justified it. Not by hiding the AI. By naming it, framing it as the tool, and putting the judgment loop in front of his IC.

The market is going to keep asking the wrong version of that question for another year or two. The practitioners who win in that window are the ones who have a sharper answer ready, and the artifact to prove it.

If you are running corp dev or BD inside a healthcare platform, or you sit on the buy-side of PE-backed healthcare deals, and you are wrestling with how to package AI-enabled sourcing internally, reply to this email. I am happy to talk through the framing, no pitch attached.

-Shawn

Book a call | LinkedIn | Healthcare M&AI

This newsletter is for informational purposes only and does not constitute investment, legal, or financial advice.