Beyond cost: AI as a revenue engine

Ask an operating partner what AI did for a portfolio company last year and the answer is almost always a cost number: hours taken out of the monthly close, tickets handled without an agent, invoices processed without a clerk. There are good reasons for that. Cost shows up within a quarter, it goes straight into an EBITDA bridge, and it doesn't depend on customers doing anything differently.

The industry's intentions have already moved on. In FTI Consulting's 2026 Private Equity AI Radar, a survey of 200 fund and operating leaders, revenue acceleration was the top AI priority, named by 41% of respondents. Most value creation plans haven't caught up. We think that gap is expensive, because over a seven-year hold, cost savings and new revenue behave very differently.

Why cost savings fade

Cost programs work. The trouble is that their gains are the easiest for competitors to copy, and in many markets the copying hands the gain to customers.

BCG's 2026 analysis of who captures AI's value lays out how this happens. Because every company can buy the same models at similar prices, the technology itself reaches parity within two to three years. After that, who keeps the savings depends on the market. In retail, BCG writes, hypercompetition pushes almost all of the AI dividend through to consumers within a handful of years, as companies use their savings to cut prices. In financial services, where a few large firms and heavy regulation limit competition, shareholders have kept most of it.

For a private equity owner, the problem is timing. A seven-year hold outlasts the parity window. A cost-cutting plan in a competitive market can add real EBITDA in years two and three and give much of it back through lower prices by years five and six, right when the exit is being prepared.

BCG also has a warning that fits many private equity portfolios closely. Midsize incumbents, big enough to carry old costs but too small to own a data advantage or network scale, get squeezed from both sides: by large incumbents pulling ahead and by AI-native companies built without the legacy. The advantages BCG expects to last are scale, data that flows through the business every day, trusted customer relationships, deep integration into customers' systems, and scarce physical assets. Those also happen to be what new revenue gets built on.

Three ways AI creates value

The first is cost: the same output for fewer hours or fewer dollars, in work like reconciliation, accounts payable, claims intake, first-line support and document review. It shows up as lower operating expense, and it's where almost every program starts.

The second is capacity: the same people producing more. A business development team that can answer three times as many requests for proposals. An estimator who turns quotes around in hours. An account manager covering twice the book. Capacity gets the least attention of the three, because it can be booked either way. Take it as smaller headcount and it's a cost saving. Point it at more bids, quotes and accounts and it's revenue without a matching rise in cost. Choosing between the two is a strategic decision, and most plans make it by default.

The third is new revenue, money the company couldn't earn before: AI features sold as paid add-ons, new products, services that turn a firm's expertise into something that doesn't need an hour of labor behind every sale, and prices that capture more of the value delivered. It has the most upside of the three and gets the least planning.

What the largest companies are showing

The clearest evidence that AI can drive revenue comes from some of the biggest companies in the world.

Amazon put an AI shopping assistant, Rufus, inside its store. In its fourth-quarter results in February, Amazon said more than 300 million customers used it in 2025 and that it helped deliver nearly $12 billion in incremental annualized sales. That's Amazon's own estimate, but it's a sales number, and it came from helping customers find and buy more.

JPMorgan used AI to give its wealth advisers more capacity and pointed it at clients. In May 2025 its asset and wealth management arm said gross sales rose 20% from 2023 to 2024, with AI tools helping advisers spend more time on high-value client work, and that it expects advisers to take on 50% more clients over the next three to five years. The second number is a target, but the direction is the one most plans skip: the bank is taking the gain as more clients per adviser.

Salesforce turned AI into a new product line. Its agent platform, Agentforce, passed $1.5 billion in annual recurring revenue in the quarter that ended July 31, up more than 240% from a year earlier (the figure now also includes some of its other AI products). Salesforce also prices much of it per action instead of per seat, which we come back to below.

Private equity portfolios are starting to show it too. The European software investor Hg said in June that it had built more than 100 AI products across its companies and that about a fifth of its portfolio companies now get over 10% of their new bookings from AI.

And firms that sell expertise are turning it into products. Thomson Reuters, whose business is legal, tax and accounting know-how, has rebuilt its products around an AI assistant called CoCounsel. In the second quarter its three largest segments grew 10% organically, and it raised its outlook for the year. For services firms this may be the biggest opportunity of all. Most AI spending so far has come out of software budgets, and services budgets are far larger. A firm that builds its methods, judgment and precedent into a system that delivers a defined result can sell that result at a fixed price, with margins that don't depend on the hours behind it. If it doesn't, an AI-native competitor will make its clients the same offer.

Pricing is changing too

AI is also breaking the per-seat price. When one agent does the work of several users, charging per seat shrinks the bill just as the product starts delivering more. Software companies are already moving. In Growth Unhinged's survey of more than 240 software and AI companies, seat-based pricing fell from 21% to 15% of companies in a year, while hybrid models that mix subscriptions with usage rose from 27% to 41%. Salesforce charges 10 cents per agent action under one of its models. In June, Deloitte published accounting guidance on recognizing revenue from outcome-based pricing for AI agents, a sign the practice has become common enough to need it.

For a portfolio software company, repricing protects revenue as seat counts shrink, and it captures value that per-seat pricing never charged for.

What makes new revenue last

The same things that let a company keep its margins make its new revenue hard to copy. An AI product built on a general-purpose model and public information can be rebuilt by any competitor with the same model within a quarter. A competitor can't easily rebuild one that runs on the company's own data, its codified expertise and its place inside the customer's workflow.

That has two consequences for how the capability gets built. The first is ownership. The objects, rules, precedent, evaluations and integrations that make the product work are the asset. If they live inside a vendor's environment, the company has built someone else's moat, and a future buyer will notice. We wrote about why ownership comes first in the sovereign frontier enterprise.

The second is model choice. Unlike seat-based software, an AI product carries an inference cost on every unit sold. Gross margin depends on sending each task to the model that does it well at the lowest cost, and on being free to switch when prices change, which they do every few months. A product tied to one model's price list has handed its margin to that model's provider. That's why model-agnostic routing is a commercial question as much as a technical one, and why training a smaller model on the company's own data sometimes pays, as we discussed in should I train my own model.

Putting revenue into the value creation plan

Revenue bets are harder to underwrite than cost programs, which is why they get crowded out. A few habits make them manageable:

• Run one revenue bet alongside the cost work from the start. It needs mostly the same foundations: clean data, rules the company owns and a governed agent.
• Decide what to do with capacity on purpose. For each gain, write down whether it's being taken as cost or pointed at growth, and why, and make it a board decision.
• Treat revenue bets as options. Start with a small release to a set of existing customers, with a price, a target attach rate and a date to decide whether to stop or fund it properly.
• Measure paid adoption and whether it holds. A shipped AI feature nobody pays for is a cost.
• Test new pricing early. Outcome and hybrid pricing change revenue recognition, forecasting and sales compensation, and those take time to sort out, more time than the quarter before an exit allows.

At exit, a buyer pays for a cost saving once, at the multiple, and only if it survives diligence, which we cover in the private equity AI playbook. A new revenue line with its own customers and its own growth rate gets valued the way buyers value growth, usually far more generously. Which sectors favor which approach is mapped in AI across a portfolio, sector by sector.