AI doesn't sell businesses

AI doesn't sell businesses.
The model behind it does.

Why the AI multiple gap is real, why most B2B businesses won't qualify for it, and what to do in the 24 months before a transaction.

TL;DR · 12-minute read

For founders, CEOs and PE sponsors 18 to 36 months from a liquidity event.

Same sector. Same year. Same investor pool. 9fin closed a Series C at 16x revenue while the B2B software index sold off 25%. The gap between the two — what John Mecke first named the "AI valuation gap" — is real, persistent, and most businesses won't qualify for it.

  • The premium is real. 83% of buyers paid higher multiples for AI-native or AI-integrated targets in 2026. 86% expect those premiums to persist.
  • The discount is real too. 63% of evaluated targets show only limited AI use. Most AI claims do not survive the diligence room.
  • The diagnosis is structural. Most AI investment funds operational efficiency — tool adoption, copilots, content automation — not the financial KPIs (NRR, gross margin, expansion, win rates) that actually move the multiple.
  • Operational AI changes the cost line. Structural AI changes the architecture of the revenue. Only one moves the multiple.
  • It's a 24-month build, not a 6-month polish. Cohort data, NRR attribution, governance trails, capital allocation evidence — none of it can be assembled in the run-up to a transaction.
  • The bar has moved from "do you have an AI strategy?" to "show me how AI is changing your KPIs."

If your answers to the AI Value Test — six questions, two minutes — aren't obvious, the work starts now.

The 16x question.

In April 2026, 9fin — an AI-native data platform serving the global debt markets — closed a $170m Series C at a $1.3bn valuation, on roughly $80m of projected Annual Recurring Revenue (ARR). The implied revenue multiple sits in the region of 16x.

In the same window, the B2B software index sold off close to 25% — including a single 48-hour stretch in February that wiped $285bn from SaaS market caps — while the S&P 500 held within a few percentage points of its all-time high. Customer support, marketing automation and content tooling — the categories most exposed to AI agent substitution — led the decline. The market is calling it the "Great SaaS Reset." This wasn't macroeconomic pressure. It was investors pricing in "seat compression": a single AI agent now doing the work of multiple licensed users.

Same sector. Same year. Same investor pool. Two completely different fates.

The question every founder, CEO and board should already be asking is the obvious one: what separates the businesses commanding 16x from the ones quietly being repriced?

The answer is not the one most operators are reaching for. It isn't AI investment. It isn't the size of the AI team, the number of pilots launched, or the volume of tools deployed. It's the commercial model wired around them.

The AI premium is real.
The qualification bar is higher than most realise.

The data is unusually clear for a market in transition.

83% of buyers in 2026 report paying higher multiples for AI-native or AI-integrated targets. 86% expect those premiums to persist. 95% of PE funds report that AI initiatives meet or exceed their original business case. Companies systematically deploying AI enterprise-wide are showing nearly twice the Return on Invested Capital (ROIC) of those that aren't.

So far, so encouraging. Until you read the next number.

63% of the companies those same buyers are evaluating show only limited AI use in due diligence. Most AI claims do not survive the diligence room.

Translation: there is a real, persistent, quantified AI premium. There is also, increasingly, a real AI discount — the repricing applied to businesses whose AI story doesn't hold up to scrutiny. The two together are what John Mecke first named at Development Corporate as the "AI valuation gap" (March 2026). Most B2B businesses telling an AI story will not pass the test that earns the premium. A growing number will pay for being on the wrong side of disruption their model wasn't built to absorb.

There is a clear AI premium.
Most B2B businesses won't qualify for it.

The diagnosis sits behind both statistics. Most AI investment is happening in the wrong layer of the business.

Most AI work is happening where it can't move the multiple.

Investors in 2026 are not pricing AI activity. They are pricing AI evidence in the financial KPIs that have always moved multiples — and AI is now expected to show up in them.

Four categories, in the order a sophisticated buyer reads them.

Retention. Net revenue retention (NRR) above 106% — the top quartile across venture-backed SaaS — correlates with 2.5x faster growth and meaningfully higher multiples. Gross revenue retention sits beneath it as the integrity check — proof the customer base isn't leaking quietly.

Margin. Gross margin with AI inference costs fully loaded — not the headline number that pretends they don't exist.

Expansion. Expansion revenue and seat-to-enterprise conversion. Evidence the product is pulling customers up the value curve, not just holding them flat.

Efficiency. Win rates and sales efficiency. Whether AI is making the commercial engine demonstrably faster — or just busier.
Most AI investment is doing none of this. It is funding tool adoption, content automation, internal copilots, marketing experimentation, sales enablement pilots. Useful work in places. Margin-friendly sometimes. None of it touches the metrics above.

The multiple moves when AI shows up in your retention and expansion numbers — not when it shows up in your product roadmap slides.

The implication is uncomfortable. AI investment that doesn't change those KPIs is, from a sophisticated buyer's perspective, almost indistinguishable from no AI investment at all. The slide showing how much was spent on AI tools, how many hours were saved and how many teams were trained lands as evidence of activity. It is not evidence of value.

This is the source of the multiple gap. Two businesses with similar AI investment, similar tool stacks and similar internal energy can sit on opposite sides of it — because one has wired AI into the architecture of its revenue, and the other has wired it into the cost line of an unchanged model.

AI value is structural.
AI activity is operational.

Operational AI is workflow optimisation. Content production accelerated. Internal copilots deployed. Sales enablement pilots running. Support deflection automated. It changes the cost line of the existing business model.

Structural AI is model redesign. Pricing rebuilt around the value the customer captures, not the units they buy. Customer relationships embedded so deeply that retention and expansion compound rather than being fought for quarter by quarter. Proprietary data, expertise and audience surfaced as revenue products in their own right. It changes the architecture of the revenue itself.

Both matter. Only one moves the multiple.

Operational AI changes the cost line.
Structural AI changes the architecture of the revenue.

The boards and sponsors investing heavily in operational AI without doing the structural work are buying productivity. They are not building enterprise value. The dividend they're harvesting is real. The moat they think they're building isn't.

Three places AI compounds value.
Three places it quietly leaks it.

Where AI compounds value

01 — Data assets that survive replication. Privileged access. Primary research. Longitudinal datasets that take years to assemble. Expert judgment codified as institutional knowledge before it walks out the door. The buyer's test in 2026 is: how replicable is your data by AI? The only acceptable answer in serious diligence is "not easily."

02 — Embedded customer workflows that shift NRR. AI features used daily, that drive expansion, and surface in the retention curve. The 9fin model. The metric that moves the multiple is exactly the metric this work moves. Sequence matters: AI built into the customer's daily operating rhythm beats AI bolted onto the product as a feature, every time.

03 — Pricing models that scale with value, not seats. As individual users become more productive, seat-based revenue erodes silently. The businesses getting ahead are restructuring around value delivered: identifying secondary users inside client organisations, building enterprise tiers, layering AI-premium capabilities, capturing the productivity gain rather than losing it to client headcount consolidation.

Where AI quietly leaks value

01 — Margin theatre. Efficiency gains harvested as short-term margin without reinvestment in compounding capability. Looks healthy in the next quarterly board pack. Surfaces as a question in 18-month diligence that the management team can't answer cleanly: why didn't this margin become a moat?

02 — Undifferentiated content under Generative Engine Optimisation (GEO) erosion. AI-generated content competing in a market saturated with AI-generated content. A growing share of buyer research now happens inside ChatGPT, Perplexity and Claude rather than on the open web — meaning traffic compresses even as production expands. The audiences quietly disappearing from your funnel are not coming back.

03 — Per-seat revenue in a fewer-users world. When your pricing is tied to the number of licensed users — the per-seat model that built the SaaS industry — AI that makes one user as productive as three becomes a revenue threat, not a feature. Clients need fewer seats. Renewal conversations get quietly harder. The pricing model wasn't built for the new productivity reality, and the revenue base erodes a quarter at a time without anyone naming the cause.

This brings into focus the question: when AI delivers efficiency gains, do you take them as margin improvement, or reinvest them in capability that compounds?

For any business 24 months from a transaction, the default answer is unambiguous. Reinvest aggressively. Harvested margin without compounding capability is the single easiest gap for serious diligence to expose. The buyers paying premium multiples in 2026 are not paying for what AI is being used to save. They are paying for evidence of what AI is being used to build.

The new AI diligence test.

The bar has moved from "do you have an AI strategy?" to "show me how AI is changing your KPIs."

Six questions every B2B board should now be able to answer cold. Not in a workshop. Not after a week's preparation. In the room.

I call it the AI Value Test: six questions, two minutes, brutal clarity.

  1. 1
    Is your data replicable from public sources by a well-prompted Large Language Model (LLM)? If yes, the moat is already gone, and the conversation is about what comes next.
  2. 2
    Is your revenue tied to seats AI is making redundant? If yes, your pricing model is leaking, and the rate of erosion is accelerating quarter-on-quarter.
  3. 3
    Where is the AI dividend going — margin or moat? If the answer is "margin," you're harvesting an asset you should be compounding.
  4. 4
    Can you evidence AI's contribution to a specific KPI — NRR, gross margin, win rate — over four quarters or more? If the evidence lives in slides rather than in cohort data, you don't have an AI story. You have an AI claim.
  5. 5
    Is AI owned at board level, or scattered across teams? If scattered, governance fails the test, and so does the diligence around how decisions are made.
  6. 6
    What does your AI strategy look like with zero new tools added? If you can't answer that question, you have a tool stack, not a strategy.

The businesses on the right side of the multiple gap are the ones whose answers are obvious. They've done the work. They have the data. They can walk it through with conviction. The businesses on the wrong side discover the questions in the data room — long after the answers can be built.

This is a 24-month build, not a 6-month polish.

Cohort data takes four quarters or more to be credible; NRR attribution to specific AI initiatives requires the same time horizon. Margin reinvestment needs a documented capital allocation trail with management accountability written into board minutes. AI governance needs evidence of decisions made and revisited — not intentions stated.

None of this can be assembled in the six months before a transaction. Increasingly, experienced buyers and their AI diligence specialists know it.

For founders, CEOs and PE sponsors 18 to 36 months from a liquidity event reading this: the window is now. The work done in the next four quarters is the evidence that earns the multiple in the four after that.

For those further out: the longer the runway, the more compounding the gain. AI built into the revenue model two years ago is showing up in 2026 multiples. AI built in today will show up in the multiples of 2028.

For those already in the data room: focus on what can still be evidenced cleanly. The diligence question isn't have you done everything? It's is the management team in control of its AI response? That's still rescuable up to a point. After that, it isn't.

The narrative that survives the diligence room.

What a credible AI value creation story sounds like to a sophisticated buyer in April 2026. Three components, in order. Any management team should be able to walk a buyer through them in fifteen minutes without notes.

A clear thesis on how AI is changing your market. Where is the Total Addressable Market (TAM) expanding — energy, infrastructure, manufacturing, the sectors gaining AI-driven budget — and where is it shrinking? Which of your buyer segments is gaining ground, and which is quietly losing it? What does your repositioning look like in response?

Evidence of how AI has changed your business. Which KPIs moved. By how much. Over what period. Linked to which AI investment. With what attribution logic. Numbers, not adjectives. The buyer's eye is trained to spot the gap between the two.

A defensible plan for the next compounding investment. Where does the AI dividend go next? Why is that the highest-return reinvestment available? What capability are you building that the next acquirer can't easily replicate?

Numbers, not adjectives.
The narrative is the through-line, not the deck.

The narrative is not a deck. It is the through-line that ties every operational decision back to the multiple. Sophisticated buyers can detect that line within thirty minutes of a management presentation — or its absence.

The model is the moat.

The investors paying premium multiples in April 2026 are not paying for AI capability. There is plenty of that available, and the cost is falling. They are paying for evidence that the commercial model has been rebuilt — to compound the gains AI now produces, and to defend against the threats it now creates.

The businesses on the right side of the multiple gap have done that work. The businesses on the wrong side are still treating AI as a productivity layer over an unchanged model.

There is no third option.

AI doesn't sell businesses. The model behind it does.

If the answers to the AI Value Test aren't obvious, that's the conversation worth having.

Sources & references.