How AI Agents Could Change Business Models Built on Friction

A large number of businesses earn part of their margin from friction. Some benefit when customers leave an arrangement alone: the insurance policy that renews at a higher price, the savings balance that sits in a low-rate account for years. Others are paid to remove the friction for us, by searching a fragmented market and arranging the purchase. AI agents reduce the time and effort involved in both, and that puts pressure on any revenue line that depends on customers not checking.

The exposure does not require consumer agents to become a mass-market product. An agent with the right permissions keeps comparing rates and following up on requests long after a person would have decided they had better things to do. Spread across millions of customers, small inefficiencies that were never worth a person’s evening become material. The same logic applies inside a company’s own contract book, where terms agreed years ago are rarely read again.

Two kinds of friction revenue

The first is inertia revenue. The supplier earns more because most customers never ask for a better deal. Renewal pricing in insurance, deposit rates in consumer banking and forgotten subscriptions all work this way.

The second is search and arrangement revenue. An intermediary aggregates a fragmented market, makes the options comparable and handles the transaction. Insurance brokers, comparison sites and travel platforms earn referral fees and commissions for doing that work on the customer’s behalf.

Agents attack both from the same direction. The task they automate (checking, comparing, arranging) is the task the revenue depends on.

Inertia: insurance renewals and bank deposits

Insurers have to price risk, but customers also face the separate job of finding comparable cover and working out whether they are overpaying. That job is annoying enough that many people accept another renewal. In one widely shared example from 2026, a user uploaded an existing car insurance policy to the Muse agent, asked for equivalent cover at a better price, and reported a replacement saving $3,500 a year, with the agent buying the new policy and canceling the old one.

Bank deposits are a larger version of the same mechanism. Deposits fund lending, and keeping that funding cheap is valuable. In a September 2026 note titled “Is an Agentic Bank Run Coming?”, Apollo chief economist Torsten Slok asked what would happen if agents routinely moved household cash into higher-paying accounts. A bank that has to compete harder for balances faces higher funding costs without any crisis at all.

Suppliers will respond, for example by restricting discounts or changing contract terms. The ability to charge more simply because most customers never ask still becomes harder to rely on.

Search and arrangement: comparison and booking platforms

Comparison sites and travel platforms do much of the work people now ask agents to do. Skyscanner earns referral fees and commissions from travel providers alongside advertising. Booking.com earns commission on accommodation reservations. Booking Holdings generated $26.9 billion of revenue across its brands in 2025.

If an agent can compare suppliers and complete the booking, the customer has less reason to visit the platform, and the supplier has an incentive to sell directly and keep more of the price. Part of the platform survives. Skyscanner already supplies travel data through APIs that an agent could use without sending anyone to the website, and Booking’s supply relationships and reservation servicing still matter. What changes is the role. A business that loses the customer relationship usually accepts a different position in the chain on a different fee.

Professional intermediaries face a related shift, covered in The Company Is the Dataroom: judgment, negotiation and relationships keep their value, while the transport of information does not.

When both sides have an agent

Marketplaces are the next step. One person’s agent knows they would sell an unused camera for $400, another’s is looking for that model below $450, and the two settle the terms without either person spending an evening on messages. Anthropic’s Project Deal experiment tested this with agents representing 69 employees, which negotiated 186 real deals worth just over $4,000.

An open marketplace still needs trust, payments and dispute resolution, and those remain valuable services. Far fewer pages need to be browsed by a person, though, which matters to any business funded by listings and advertising around that browsing. Whoever organizes access between agents and establishes trust could end up running a powerful marketplace of its own.

The same mechanism inside a company’s contracts

The consumer examples are the visible ones. The larger effect for most mid-market companies sits in their own customer and supplier contracts.

At one services business we work with, every customer contract was loaded into a single structured record and an agent compared each term against the rest of the book: payment terms, damages provisions, exit clauses, price adjustment mechanisms. One customer’s fuel surcharge mechanism had stopped being tracked during the pandemic, when fuel prices collapsed, and nobody had returned to it once prices recovered. Another contract allowed open-ended back charges that no other contract in the book permitted. Finding either one took no new analysis, only someone reading every contract against every other contract, which no one had time to do.

That is the operating-level version of an agent re-shopping an insurance renewal. A company can recover margin from terms that drifted, and its customers’ procurement teams can run the same review against it. Accenture’s results in October 2026 showed the early form of this in services: pricing fell in many areas as clients sought a share of the savings from AI. The predictive operations case study describes the foundation this kind of review depends on, with operating, billing and contract data reconciled into one record.

Questions to ask of an exposed business

For a buyer, an owner or a lender, the exposure can be assessed directly:

• How much margin comes from customers who never renegotiate? Compare renewal pricing with new-customer pricing, and retained deposits with market rates.
• What does the fee pay for once searching is free? Separate the work of gathering information from the services an agent cannot provide.
• Does the business hold something an agent needs? Supply, risk capacity, payment rails, servicing and trust keep their value. Aggregation alone weakens.
• What would an agent find in the contract book? Terms that have drifted since signing are both a source of recoverable margin and an exposure once customers start checking.

Consumer adoption is still early. PNC Research data shared by a16z put US households paying for an AI subscription at 2.2 percent in April 2026, more than double a year earlier, and the interface for high-stakes decisions such as changing insurance or moving money remains unresolved. The pricing pressure does not need to wait for that. It arrives as soon as enough customers, or their procurement teams, can check at almost no cost.