What AI does to your SaaS exit

A founder I will call Andrew runs a $6 million ARR logistics software business. His GRR is 93%. His sales motion is documented. His leadership team owns its numbers. By conventional exit readiness standards, he was well prepared for his first management presentation.

The first question was not about retention, concentration, or margin. A PE partner asked: if your customers deploy AI agents across their operations teams over the next three years, how many of your seats disappear?

The next question was sharper: what percentage of your product's core outcome could a capable team reproduce with current AI tools?

Andrew began answering with his AI roadmap. That was not what the buyer was asking. The buyer was not asking whether the company used AI. He was asking whether AI made the company more valuable, or less.

This example is a composite, but the questions are real. They are appearing with increasing frequency in software transactions, and they now affect how buyers assess revenue durability, growth, margins, and strategic defensibility.

AI is not a seventh pillar

The fundamentals buyers use to evaluate a SaaS business are still there. Revenue quality, growth predictability, margins and operating leverage, the strength of the operating system underneath the business, leadership depth, and data credibility are still the six pillars of exit value.

What has changed is that AI is now repricing all six at once.

Revenue quality is repriced when a buyer asks whether your customers could achieve the same outcome with an AI tool and a fraction of the spend. Growth predictability is repriced when AI-native competitors enter your vertical growing at rates that make your motion look slow. Margins are repriced in both directions: downward when AI features carry inference costs your P&L hides, upward when AI takes real cost out of your delivery. Your operating system is repriced when a buyer models how much of your headcount an AI-enabled operator would actually need. And data credibility now has a new test inside it: whether the claims you make about AI in your business survive the same scrutiny as your revenue definitions.

Inference is the work an AI model does every time it runs. Training builds the model once. Inference is paid on every request, usually priced per token, and it scales with usage rather than with customers. Hosting is close to a fixed cost. Inference is a variable one, and a heavy user can cost twenty times a light user while paying the same subscription.

AI is not a seventh pillar. It is a new lens across the six you already have. That is why it deserves its own treatment: buyers are already reading every pillar through it, and most founders are not.

The evidence for that gap is stark. In Software Equity Group's survey of more than 200 private equity investors, strategic acquirers, and SaaS CEOs, 85% of buyers named AI-driven commoditisation as the single biggest risk to SaaS value.¹ The same buyers report that most sellers have barely moved: 63% saw only limited AI adoption in the targets they evaluated during 2025. When buyers rate a risk this highly and see it unaddressed in the businesses they assess, the gap gets settled in the price. It is not usually settled in the seller's favour.

The window

Here is the current state of play, and why timing matters more on this topic than on any other dimension of exit readiness.

Buyers are paying premiums for credible AI capability now. In SEG's AI in SaaS survey, 80% of strategic and private equity buyers report paying a premium today for companies with AI deeply integrated into core workflows, and 87% expect that premium to increase over the next year.² The same buyers expect the market to move quickly: 61% expect the software companies they look at to be AI-driven by the end of 2026.¹

Read those numbers together and the shape of the window is clear. Right now, credible AI capability is scarce enough to command a premium. Within a couple of years it will be table stakes, and the premium will not go to businesses that have it. The discount will go to businesses that do not.

This is the same dynamic that plays out with clean financials and leadership depth, compressed into a shorter timeframe. The founders who move early collect the premium. The founders who wait inherit the discount. The difference on this dimension is that the market is moving faster than any standard exit readiness timeline.

The AI Screen

Over the last two years, a consistent set of questions has emerged in buyer processes. They crop up in different workstreams, but they boil down to four key questions that together determine whether AI makes your business worth more or less.

I call it the AI Screen. Buyers are running this screen against your business. This guide gives it to you so you can run it first.

Question one: Could AI replace you? This is the substitution question, and it is the one that kills deals. The buyer is asking what happens to your product's reason to exist as AI capability improves and as your customers (and also their customers) adopt it directly.

Question two: Is your AI real? This is the substance question. The buyer is separating genuine AI capability, evidenced in retention and revenue, from AI theatre added to an information memorandum.

Question three: What does AI do to your economics? This is the margins question, and it cuts both ways. AI features carry real costs that most P&Ls hide, and AI leverage creates real operating improvements that most founders have not yet captured.

Question four: Does your revenue model survive the shift? This is the pricing question. If you charge per seat and AI reduces the number of humans your customers employ in the function you serve, your growth engine has a leak in it that compounds.

Each one deserves its own treatment.

Question one

Could AI replace you?

The most uncomfortable question in any current SaaS process is some version of this: what percentage of your users could achieve the same outcome with a general-purpose AI tool? Or its sharper cousin: could a small AI-enabled team rebuild the core of this product in a quarter?

For some businesses the honest answer is genuinely reassuring, and for others it is not. What determines the answer is which moats your business actually has, because AI has changed which moats hold.

Feature sets alone are becoming materially less defensible. AI-assisted development means a competent team can replicate a feature-based product faster than at any point in the history of software. If your differentiation is a capability rather than a position, buyers will price the erosion. Shallow, easily replicated integrations provide less protection than they once did. A directory of API connections was a switching cost when connecting systems was hard. AI agents increasingly orchestrate across tools directly. And user familiarity, the switching cost that came from your customers having learned your interface, weakens when AI can operate your tool and your competitor's tool with equal ease.

What holds is depth. Proprietary data that compounds, where usage generates data that improves the product that drives more usage, is the strongest position in an AI market, and it is exactly the asset a vertical SaaS business with years of customer interaction history is sitting on. Workflow depth holds: a product that is embedded in how a customer's business actually runs, touching their financial systems, their compliance obligations, and their audit trail, is not displaced by a general-purpose tool. Regulated domain expertise holds, because AI cannot itself assume contractual, regulatory, or professional accountability. Deep integration into mission-critical systems holds, because the cost of ripping it out is operational, not cognitive.

Vertical SaaS businesses in the $2M to $20M ARR range mostly have more of these deep moats than their founders articulate, and fewer than their founders assume. The threat is real: Bessemer tracks AI-native vertical competitors growing at around 400% year over year while already commanding around 80% of the contract value of the incumbent systems they compete with.³ But the defensible position is usually real too. The work is separating one from the other honestly, and then building the evidence for the moats you genuinely have. A buyer who asks the substitution question and receives a specific, evidenced answer about data depth, workflow embed, and switching costs has just watched you clear the highest bar in the current market. A buyer who receives a general assurance that customers love the product has not.

Question two

Is your AI real?

Buyers have seen a great deal of 'AI washing' (stories of AI adoption and innovation without substance to back it up). They ignore what the product announcement says and look for AI in the metrics.

Real AI capability shows up in the numbers you already own. It shows up in NRR, when AI features drive expansion revenue that customers pay for. It shows up in GRR, when AI capability deepens the product's role in the customer's workflow and lifts retention. It shows up in win rates and deal velocity, when AI differentiation is actually moving buying decisions. And it shows up in pricing, because capability that customers will not pay for is a cost, not an asset.

The bar is high because genuine impact is rare. More than 80% of respondents in McKinsey's 2025 State of AI survey reported no tangible enterprise-level EBIT impact from generative AI.⁴ Buyers know this number, or ones like it. When your information memorandum claims AI-driven advantage, their diligence teams go looking for evidence: adoption data, retention deltas between AI-adopting and non-adopting cohorts, revenue attributable to AI capability, and customers who talk about the AI capability unprompted in reference calls.

This has a direct implication for sequencing. An AI feature shipped eighteen months before a process, with cohort data showing its effect on retention and expansion, is an asset that supports a premium. An AI assistant added to the product in the quarter before the information memorandum goes out is worse than nothing, because it invites the washing question and buyers who find washing in one place look harder everywhere else. The fix-and-prove logic of exit readiness applies here without modification. The capability is the fix. The cohort evidence is the proof. The proof takes time, which means the work starts now.

Question three

What does AI do to your economics?

AI moves your margins in two directions at once, and buyers will examine both.

The first direction is down, and it is the one founders systematically underestimate. AI features carry inference costs, and inference costs behave differently from anything else in a SaaS cost base. They are variable, they scale with usage rather than with customers, and a small number of power users can consume a wildly disproportionate share. ICONIQ's ongoing survey of roughly three hundred executives building AI products finds companies expecting gross margins to reach around 52% on average in 2026, against the 70 to 80% that has long been the SaaS standard.⁵ Your AI features live somewhere on that spectrum, and if their costs are sitting in a general hosting line, you do not know where.

This is the hidden services problem wearing new clothes. Delivery costs buried in the wrong P&L line get found and reclassified in a quality of earnings review, and the multiple adjusts. AI costs buried in hosting follow exactly the same script. The remedy is also the same: get there first. Isolate AI costs in your P&L. Track cost per feature, per customer, and per request. Know your gross margin with AI costs correctly allocated, and price AI capability so the margin story survives scale. A founder who can show a buyer instrumented AI economics, with unit costs trending down as usage grows, has turned a diligence risk into evidence of operational discipline.

The second direction is up, and it is the most underused opportunity in businesses at this scale. AI can allow businesses at this scale to take meaningful cost out of selected workflows without degrading the customer experience. The public reference point is Klarna, which reported that its AI assistant handled 2.3 million customer service conversations in its first month, work it equated to seven hundred full-time agents.⁶ You do not need Klarna's scale to capture the same mechanics. Support deflection, AI-assisted onboarding, engineering productivity, and automated account health monitoring all reduce the cost of delivering each dollar of revenue.

Now connect that to operating leverage. Buyers pay for evidence that margins expand as revenue grows. AI leverage, captured deliberately and demonstrated across four or six quarters, is exactly that evidence. A business that grew support tickets 40% while holding support headcount flat has a slide in its management presentation that no amount of narrative can substitute for. This is also the cheapest multiple expansion available to you, because it requires no new revenue. It requires taking the tools seriously inside your own operation.

Take heart if this is not clear to you today. We are currently watching the biggest companies in the world attempt to address this, and it is not certain that any of them have nailed it.

Question four

Does your revenue model survive the shift?

The last question is the one Andrew was asked, and it deserves to be taken literally. If your pricing is per seat, your revenue is a function of your customers' headcount in the function you serve. AI is now actively reducing headcount in exactly the functions SaaS serves best: support, operations, finance, administration. When your customer automates a team of ten down to six, your product may be more valuable to them than ever, and your invoice goes down 40% anyway.

Buyers model this. A buyer evaluating a per-seat business in a function with high automation exposure will run seat-decline scenarios against the revenue base and price the result. Bloomberg estimates cited by RSM US project that subscription-based pricing could decline from around 60% of software pricing models toward 30% over the next decade, with outcome-based pricing expected to shift from 10% to 60%.⁷ Whatever the precise numbers turn out to be, the direction is not in dispute, and buyers are underwriting the direction.

This does not mean abandoning per-seat pricing before a process. Repricing a customer base is slow, delicate work, and a botched migration is a worse diligence finding than seat exposure. It means three specific things. Quantify your exposure: know what percentage of your ARR sits in seats that your customers' AI adoption could plausibly eliminate, the same way you know your customer concentration. Decouple expansion from headcount where you can: usage components, module pricing, and value-based tiers all grow with the customer's business rather than with their staff count. And where you introduce AI capability of your own, price it on the value it delivers rather than folding it into the seat, because capability given away free is margin surrendered and evidence destroyed.

A founder who walks into a management presentation with seat exposure quantified, a repricing direction established, and early evidence that expansion no longer depends on customer headcount has answered the fourth question before it was asked.

The AI Screen scorecard

Run the screen on your own business and score it. For each question, score 0 to 4: red scores 0 or 1, amber scores 2, and green scores 3 or 4, with the top of each band earned by evidence, not intention.

A business scoring AI advantaged is positioned to collect the premium the current window is paying. A business in the red zone on the substitution question will find that question dominates the entire process, because no buyer prices growth in a product they believe AI will strand.

What getting AI exit ready requires

The work follows the same fix-and-prove logic as everything else in exit readiness, on a shorter clock.

Run the AI Screen on your own business first, honestly. Answer the four questions the way a sceptical buyer would, not the way your product marketing would. The substitution question in particular needs outside perspective, because founders systematically overrate their moats and underrate their data assets.

Build depth where you have it, not features where you do not. One or two AI capabilities that compound your proprietary data and deepen your workflow position are worth more than ten assistant features. Choose the capability that will show up in retention and expansion metrics, instrument it from day one, and give it time to produce cohort evidence.

Instrument the economics before they instrument you. Separate AI costs in the P&L now. Track them per feature and per customer. Take AI leverage in your own cost base, starting with support and onboarding, and document the before and after.

Quantify your seat exposure and start the pricing evolution deliberately. Know the number. Move expansion off headcount. Price new AI capability on value.

And write all of it down. Every claim about AI in your eventual information memorandum needs a document behind it in the data room: the cost instrumentation, the cohort analysis, the adoption data, the pricing rationale. AI claims without evidence are the new aggressive revenue recognition. Buyers treat them the same way.

None of this is a separate program from broader exit readiness work. It is the same program, pointed at the newest and fastest-moving thing buyers price. Twelve to twenty-four months of deliberate work can be sufficient to build credible evidence.

Buyers now run the AI Screen alongside every other diligence workstream. The founders who arrive at a process having answered the four questions, with evidence, will find that AI is the reason their multiple went up. Everyone else will discover it is the reason their multiple went down. The difference, as with everything in exit readiness, is when you start.