Scaling B2B SaaS in the AI era: the wall everyone builds and the moat that pays

Spend time with the boards of PE-backed software companies right now and the same question keeps coming up: where is the return on the AI spend? The honest answer is uncomfortable. Only around six percent of companies turn generative AI into measurable bottom-line value, meaning more than five percent of operating profit traced to AI ^[1]. MIT finds the same picture from the other side: 95 percent of organizations see no measurable profit-and-loss effect from their AI efforts so far, even though enterprise AI spend tripled to roughly 37 billion dollars in 2025 ^[2][3]. Scaling B2B SaaS in the AI era pulls four levers at once: go-to-market efficiency, monetization, the operating model, and organization. On its own, each one moves little. Together they reset where value is created and who keeps it.

What does AI change about scaling B2B SaaS?

AI in the product has become the price of entry. Buyers assume it, and it hardly separates one offer from another. McKinsey now calls horizontal AI features like chatbots, assistants and summarizers the baseline buyers simply expect, the kind that rarely moves a P&L ^[4]. Aaron Levie of Box puts it more bluntly: a single model upgrade catches a rival up an entire year of development, and no one can take their position for granted ^[5].

Picture the business model as a castle. The castle is the value itself, the business that earns and holds revenue. The wall is the AI in the product that everyone now builds. It protects you, but everyone has the same one. The distance to competitors opens up in the moat in front of it: data your product itself generates, deeply embedded workflows, distribution others cannot rebuild overnight. The moat decides whether the value the AI creates stays with the company or leaks to the customer.

The wall reaches far beyond the product. AI runs through the entire operating structure now: software development, service, go-to-market, operations, finance. Almost every company now uses it, yet only about one percent call their use mature, and just a fifth have genuinely redesigned their workflows ^[1]. That's where the lead opens up, because the value is in rewiring the process, with only around a tenth riding on the model itself ^[20]. It's a Red Queen race, where you run just to stay in place ^[21]. Baldwin made the mechanism plain back in 2023, that whoever adopts AI later loses ground to whoever adopts it sooner ^[22]. And you can only drive fast once the guardrails are built. Good governance is what makes the speed possible ^[23].

At the same time margins are under pressure, because AI delivery cost runs alongside revenue. Classic SaaS lived on roughly 80 percent gross margin ^[6]. AI products average 52 percent, up from 41 percent in 2024, and the raw compute for the AI answers eats around 23 percent of revenue on average ^[7]. That strains the Rule of 40, the rule of thumb that growth plus margin should clear forty percent together. In public SaaS only about 20 percent of companies clear it today, with a median near 28 percent ^[8]. Bain has started floating a Rule of 30, a lower bar for a world with real variable AI cost ^[9].

Which four levers actually scale B2B SaaS today?

Four levers keep surfacing in practice. They only pay off in combination.

1. Go-to-market efficiency. Buyers self-educate through most of the decision before they talk to sales. In 2025 a buying group makes first contact only after roughly 60 percent of the journey, and in 95 percent of cases the eventual winner is already on the day-one shortlist ^[10]. Two thirds of B2B buyers want a rep-free buying experience ^[11]. Keep buying reach through paid acquisition and you pay twice: once for the AI in the product, once for a sales motion that eats the margin. The concrete move: shift selling to product and usage signals, take self-serve seriously, and move headcount from net-new selling into post-sales. At fast-growing AI companies more than half the go-to-market team already sits in post-sales ^[12].
2. Monetization. Seat-based pricing breaks once AI agents do the work that used to need more people and therefore more licenses. As a16z puts it, the seat is no longer the atomic unit of software ^[13]. Practice is following. Intercom charges 0.99 dollars per resolved case for its Fin support agent, billed only when the case is actually resolved ^[14]. Zendesk tiers outcome pricing from about a dollar per automated resolution. Salesforce moved Agentforce from two dollars per conversation to roughly ten cents per action. A chargeback specialist takes a 25 percent success fee per won case. In the AI software market the mix has already moved: 58 percent subscription, 35 percent consumption, 18 percent outcome, with consumption up from 19 percent and outcome up from 2 percent a year earlier ^[7]. Tie the price to a measurable result and the margin holds.
3. Operating model. In PE portfolios operators now drive most of the value creation, no longer cheap debt and multiple expansion. In software, 52 percent of value creation over the past decade came from revenue growth and only 6 percent from margin improvement ^[15]. This is where the AI operating partner comes in, a new role that Korn Ferry and Heidrick both describe as an answer to a gap: only around 20 percent of portfolio companies have put AI into operations with real results, the rest are stuck in pilots ^[16][17]. Where it works, it shows. In large PE portfolios, AI programs cut sales response time by as much as 65 percent and content production costs by around 40 percent ^[17]. Bain reports that companies that rebuilt operations with AI lifted operating profit by 10 to 25 percent ^[9].
4. Organization and talent. Lean beats large. AI-native vendors now reach 100 million dollars in recurring revenue with a few dozen people, and some serve thousands of customers with a handful of salespeople ^[12]. Revenue per full-time employee has climbed from 182,000 to 237,000 dollars over five years ^[12]. What matters is the right leadership at the right time, someone with the mandate and the energy to actually move the open topics. That person can join for good or fill a gap for a set period, depending on the task. One caution belongs here: Klarna replaced the work of 700 service agents with an AI assistant in 2024, then reversed course in 2025 and rehired people because quality had slipped ^[18]. Lean isn't the same as hollowed out.

The table shows why a single lever is too little.

Pull only the first lever, leave pricing and the operating model untouched, and you just move cost around. You haven't taken any out.

The moat has a number

Value is decided in the moat, and the moat is measurable. Its number is net revenue retention, how much revenue a cohort of customers still brings twelve months later, including expansion and net of churn. In private B2B SaaS the median sits near 102 percent, around 118 percent in enterprise and only 97 percent in the SMB segment ^[19]. The share of new revenue coming from existing customers has risen from roughly 25 percent in 2022 to about 40 percent in 2024 ^[19]. Companies with retention above 120 percent trade at roughly three times the revenue multiple of those below 100 percent ^[19]. When the AI in the product pulls the customer deeper into the workflow and their data lives there, it shows up in retention, long before it reaches a pitch deck. If retention doesn't rise, the moat is shallow, however good the wall looks.

How do boards start scaling in the AI era?

Four steps have worked.

1. Rewire the metrics. Outcome success rates, automation rates and net revenue retention belong in board reporting, next to recurring revenue and retention. These three numbers are still missing from many of the board decks I see.
2. Pull one lever first. Go-to-market efficiency or monetization. Four simultaneous transformations overwhelm any organization and fail.
3. Anchor operating ownership. A 100-day plan with a named person and a target in operating profit, not another AI pilot without an owner.
4. Keep the org lean. Fill a leadership gap with the person who moves the topic, for good or for a set period, depending on the task.

My Take

Pricing changes rarely fail on the product. They fail on sales incentives and on contracts that have to be reopened, and the next quarter looks uncomfortable for a while. That's what holds most teams back, and it's why pricing belongs in the board, not in the product team. The most expensive mistake is still leaving the price stuck on the old seat while AI makes the product better and the extra margin leaks to the customer.

Boards that keep demanding the old Rule of 40 in 2027 will starve the very reinvestment that protects the multiple. In the end the moat decides how much value a company keeps. And the moat gives itself away in a single number, net revenue retention, long before it appears on a strategy slide. Someone with judgment has to dig it. The model just hands over the shovel. Well, a pretty good shovel.

For how the operating levers work beyond direct sales, see the analysis on scaling B2B software beyond direct sales. For what changes for private equity at the business-model level, read the piece on MCP and PE business models.