Vijayanta Gupta

Essay

AI-native is not a product strategy. It is an operating model.

. 9 min read

The most expensive mistake a B2B software CEO can make in 2026 is to treat AI as a product strategy. The framing is convenient. It puts the work inside an organisation already designed to ship features. It gives the board a roadmap to track. It produces visible motion in quarterly updates. It also keeps AI inside the part of the company least able to change how the company makes money.

AI is an operating model. Treating it as a product strategy is the central error of this cycle, the one that explains the gap between announced AI ambition and realised economic impact. The companies that emerge from this transition with margin intact and category position improved will be the ones that re-wired how decisions get made, how capital gets allocated, how the product gets priced, and how the team gets built.

Roadmap changes will follow. They will not lead.

The old logic was a product strategy because the value chain was a product value chain

The first SaaS growth curve was a product story. Build the workflow, ship it as a service, price the seats, expand inside the account, and let services close the implementation gap. The operating model was secondary because the product carried the company’s economics on its back. Gross margins of 70 to 80 percent absorbed organisational mistakes. Net revenue retention above 120 percent forgave product weakness. The board could reasonably ask the product question first because the answer carried the company.

This logic held while three conditions held. The product owned the workflow. The marginal cost of serving a new customer was effectively zero. Expansion happened naturally as the customer hired more users. None of those three conditions hold cleanly in the AI-native cycle.

The hidden break: AI economics are operating-model economics

The break is not that AI features are hard to ship. They are not. The break is that the economics of AI features are decided in places product strategy cannot reach.

Inference is a recurring cost line. ICONIQ Capital’s January 2026 State of AI report finds inference averaging 23 percent of total revenue at scaling-stage AI B2B companies, with AI product builders expecting average gross margins of about 52 percent for 2026 SaaS Mag, 2026. Public SaaS gross margins for companies shipping meaningful AI capability have reset into a 60 to 70 percent corridor SaaS Mag, 2026. These numbers are not product metrics. They are operating metrics. They are decided by infrastructure choices, model orchestration, call-graph design, observability investment, eval discipline, and the seriousness with which the CFO and CTO co-own the inference line.

Pricing is now plural. Mixed pricing models sit at 92 percent of AI software companies in 2025 trendingtopics.eu, 2025, with seat, usage, and outcome lines inside the same contract. Salesforce already prices the AI-augmented seat at roughly 500 dollars against the 175-dollar non-AI seat Tomasz Tunguz, 2025. Designing those three lines is a finance and commercial decision before it is a product decision. The product team can build the meter; only the operating model decides what to charge for.

Talent and capital allocation has changed shape. BCG’s 10/20/70 rule, the most-cited enterprise framework for AI investment in 2025 to 2026, puts ten percent of AI budget on algorithms, twenty percent on technology and data, and seventy percent on people and processes Presenc AI, 2026. Governance is the fastest-growing line item inside that budget, rising to 8 to 12 percent in 2026 from 3 to 5 percent two years earlier Presenc AI, 2026. None of that sits inside a product strategy. All of it sits inside the operating model.

Workforce strategy is the most exposed line of all. Klarna cut headcount from 5,527 in December 2022 to 3,422 by December 2024, a roughly 40 percent reduction, with an AI agent claimed to do the work of 700 customer service staff CNBC, May 2025. By 2025 the company publicly reversed and began rehiring humans after service quality fell Time, 2025. The Klarna outcome is not a product failure. It is an operating-model failure. Capability without governance, deployment without reliability thresholds, and a workforce strategy decided outside the operating committee.

The mechanism: where strategy now sits

In a product-strategy company, the AI question gets answered inside the product review. The product leader narrates the roadmap. The CFO signs off the spend. The CRO reports adoption. The CEO summarises for the board. The operating model does not move; the roadmap does. This is roadmap theatre with extra steps.

In an operating-model company, the AI question gets answered in four different forums, each with a different decision right.

The pricing committee owns the value capture. It decides which contracts move from seat to outcome, where consumption pricing applies, and how the company guarantees floor margin. The committee includes finance, product, and the head of customer; the CEO chairs it during the transition. Its outputs become product requirements, not the other way around.

The infrastructure and inference review owns gross margin. It is run by the CFO and the CTO together, with observability, evaluation, and model orchestration treated as a single line. Its mandate is to hold the inference-cost-to-revenue ratio inside the band the operating plan can sustain, and to escalate the cases where a feature must be redesigned before it ships because the unit economics will not survive at scale.

The capital allocation review owns the operating-model bet. Twenty cents of every operating dollar that used to fund headcount-led expansion now funds model orchestration, eval infrastructure, agent reliability, and governance. The review is quarterly, the language is portfolio, and the test is whether the investments produce repeatable margin lift, not feature announcements.

The workforce design review owns the part the press is fastest to misread. It is run by the chief operating officer with the chief people officer, and it answers the question every CEO will be asked at the next board meeting: what is your Klarna-avoidance plan. The agenda is reliability thresholds before deployment, blended human-and-agent service models, and capability investment in the team that will run agents in production for the next five years.

These four forums together are the operating model. The product roadmap is a derivative of their decisions. When the four are run well, the roadmap looks tidy. When they are skipped, the roadmap is a list of features and the operating economics keep slipping.

The downstream consequences

Three consequences follow.

Boards stop asking product questions first. The first board agenda item becomes the gross margin at full AI deployment, the pricing model under stress, the inference cost ratio at the next revenue band, and the reliability threshold the company will enforce before agents touch the customer. The roadmap is asked about second, and asked to defend itself against the operating answers.

The org chart changes. The CFO and the CTO begin to operate as a pair rather than as adjacent functions, because the inference call graph and the gross margin equation are the same conversation viewed from two seats. The CRO inherits a pricing problem that was previously a financial planning problem. The COO becomes the chief reliability officer in everything but title.

Capital deployment accelerates in the unglamorous places. The market continues to reward operational efficiency and meaningful AI integration rather than speculative AI wrappers, and the spread between AI-native multiples and SaaS multiples reflects that bias. Median public SaaS sits at 3.3 times revenue in March 2026, down from 6.2 times eighteen months earlier Aventis Advisors, March 2026, with true AI-native platforms at 16 to 18 times for businesses where AI is the core product SaaSRise, 2026. The market has already concluded that operating-model conviction is worth a multiples band. The board has to decide whether to fund the conviction.

The counterargument

The strongest objection is that the operating-model framing is too abstract and slows execution. Product strategy is concrete. Roadmaps are shippable. The argument is that great companies just ship more, and that operating-model talk is consultant adjacent. There is a kernel of truth here: if the operating model becomes its own theatre, the company has gained one more committee and lost the speed advantage that made it competitive.

The answer is that the operating model question is itself a shipping question. Either the pricing committee sets the contract template by the end of the quarter or it does not. Either the inference-cost-to-revenue ratio is inside the band or it is not. Either the reliability threshold for agent deployment exists or it does not. These are not strategy artifacts. They are decisions with dates. The product roadmap is the most visible deliverable; the operating model is the set of decisions that lets the roadmap mean something economically.

The board agenda

Three changes belong on the next board agenda.

Move AI from the product update to its own standing item, with the gross margin, pricing, and reliability decisions reported together. The product report stays, but it is no longer the answer to the AI question.

Name the four forums and the executives who chair them. Most companies have at least three of these by accident. Naming them turns accident into operating model.

Fund the second growth curve operating spine before funding the next feature wave. The board that funds AI features without funding the spine behind them is buying the deck, not the company. The companies entering 2028 as recognisably second-curve businesses are deciding now which operating decisions to formalise this quarter.

The question to replace the old roadmap question is short. Not which AI features ship next. What does the company decide differently next quarter.

Sources

AI-native, operating-model, pricing, margin, boards

Back to writing