Sierra Charges Only When Its AI Agents Actually Fix Your Problem

There is a small, almost boring idea buried inside Sierra, and it is the reason a company that is barely three years old is worth $15.8 billion. The idea is this: you should not pay for customer-service software. You should pay for a customer whose problem got solved. Everything else about Sierra - the AI agents, the voice, the Fortune 50 logos - grows out of that one swap.

Sierra was started in 2023 by Bret Taylor and Clay Bavor. Taylor is the more visible of the two: co-creator of Google Maps, former chief technology officer of Facebook, former co-chief executive of Salesforce, and now chairman of the board of OpenAI. Bavor spent 18 years at Google running its virtual reality group and its experimental Labs division. They knew each other from Google. When they reunited, they did not pick a flashy consumer product. They picked customer service, which is about the least glamorous corner of enterprise software, and which almost everyone hates in roughly equal measure.

That choice looks smarter every quarter. Customer service is a place where a company loses money, loses patience, and loses customers, all at once. It is also, conveniently, a place where the work is text and voice - exactly what large language models are good at. Sierra's pitch to a big company is simple: let us build an AI agent that speaks in your brand's voice, plugs into your systems, and actually does things. Not a chatbot that reads a script. An agent that processes the return, files the claim, refinances the mortgage, updates the account.

01 / THE PRICE IS THE PRODUCTPay when it works, not when you log in

Most software bills you per seat. You buy a hundred licenses, you pay for a hundred licenses, and whether those seats do anything useful is your problem, not the vendor's. Sierra threw that model out. It charges on outcomes: you pay when one of its agents resolves a customer interaction. A conversation that goes nowhere, in principle, costs nothing.

This sounds like a billing detail. It is actually the whole strategy. When a vendor only earns money on a resolved conversation, the vendor and the customer suddenly want the exact same thing. Sierra cannot pad its numbers with unused licenses. It cannot hide a mediocre bot behind an annual contract. It gets paid when it works, and only then. That is an uncomfortable promise to make to a Fortune 50 procurement team, and it is exactly why they listen.

You only pay Sierra when its agents resolve a customer's issue. The pricing is tied to outcomes, not seats. How Sierra describes its model

For any founder reading this, that is the part worth stealing. Not the AI - the incentive. Look at your own pricing page and ask whether you get paid for your customer's success or merely for their signature. Sierra chose success, on purpose, and made it the hardest thing about the company to copy.

02 / THE NUMBERSFrom four design partners to $150M ARR

Sierra started with four design partners. It now says billions of customer interactions run on its platform, spanning mortgage refinancing, insurance claims, retail returns and fundraising for non-profits. The revenue curve reads like something out of a pitch deck, except it happened. The company reported crossing $100 million in annual recurring revenue in late 2025 - roughly seven quarters after launch - and about $150 million by early 2026.

The funding kept pace. Sierra raised $350 million at a $10 billion valuation in September 2025. Eight months later it closed a $950 million round led by Tiger Global and GV, Google's venture arm, at a post-money valuation north of $15.8 billion. In total it has pulled in about $1.58 billion from a roster that includes Sequoia, Benchmark and Greenoaks. Reaching a $100 million run rate in under two years is the kind of thing most software companies take a decade to do, if they do it at all.

03 / THE ROOMA crowded field, a different question

Sierra is not alone. The market for AI customer-service agents is loud and getting louder. Intercom has folded its Fin agent into a large existing support business. Aisera works the enterprise IT and HR side. Quiq leans on reliable, controllable automation across chat and voice. DRUID sits at the intersection of employee-facing and customer-facing agents. Each has real customers and real claims.

What Sierra did was change the question being asked in the room. The usual pitch is "how good is our bot?" - a claim that is hard to verify and easy to inflate. Sierra's pitch is closer to "who pays when the bot fails?" By tying its own revenue to resolution, it quietly reframed the whole category around accountability. A buyer comparing vendors now has a sharper test than a demo: whose money is on the line when the conversation goes wrong?

It also changes what a sales cycle feels like. A seat-license vendor has to convince a buyer that usage will be high enough to justify the contract - a forecast, essentially, that the buyer is asked to trust. Sierra can skip most of that argument. Start with a slice of the volume, watch the resolution rate, and expand only if the agent earns it. That lowers the emotional cost of saying yes, which for a Fortune 50 support leader with a career on the line is often the real obstacle. It is a friendlier way to sell precisely because it is a riskier way to get paid.

More than 40% of the Fortune 50 are Sierra customers. Sierra, on its enterprise footprint

04 / THE FOUNDERSThe second act nobody expected

There is a personal thread here worth pulling. Bavor spent nearly two decades at Google building virtual reality that never quite found its audience - Cardboard, Daydream, the ambitious Labs projects. It was inventive work that mostly stayed a curiosity. Now he is co-running a company whose agents answer millions of ordinary, unglamorous questions a day: where is my order, why was I charged, can I change my flight. Sometimes the useful act is the second one.

Taylor, for his part, keeps landing at the center of whatever the industry is arguing about. He co-created a mapping product a billion people use without thinking. He was in the room for Twitter's sale to Elon Musk as board chair. He chairs the board of OpenAI, which makes some of the models that competitors to Sierra also rely on. And rather than sit on those credentials, he went back to building - on a problem most people find tedious. That instinct, to chase the boring-but-enormous problem instead of the exciting-but-small one, is most of what separates Sierra from the pack.

05 / WHAT YOU CAN ACTUALLY DO WITH ITThe Agent OS bet

Strip away the valuation and Sierra is a tool with a concrete job. A company uses it to stand up an AI agent that carries its brand's tone and connects to its back-end systems, then lets that agent handle customer conversations across chat, voice, email and messaging. The agent does not just answer - it takes action inside the business, with guardrails the company sets. Sierra frames this as an "Agent OS," an operating layer for building, deploying and supervising agents rather than a single canned bot.

If you run a support organisation, the practical appeal is blunt: deflect the routine volume, keep humans for the hard cases, and only pay for the conversations the agent genuinely closes. If you build software, the lesson is broader and cheaper to apply - Sierra rarely names which foundation model it runs on, because it is not selling the model. It is selling the resolved conversation on top of it. In a year when everyone is racing to wrap a large language model, Sierra's edge is that it sells the result and eats the risk.

There are real risks under the story, and they are worth naming. Outcome-based pricing lives or dies on the definition of "resolution," and that definition gets harder the more complex the conversation. A customer who abandons a chat halfway, an issue that reopens a week later, a problem the agent partly solves - each is a line that has to be drawn, and drawn in a contract, and defended when the invoice arrives. Sierra also builds on foundation models it does not own, which means part of its cost base and capability ceiling sits with OpenAI, Anthropic and others. And a $15.8 billion valuation on roughly $150 million of recurring revenue prices in a great deal of the future working out.

Whether the model survives contact with a hundred more customers is a fair open question. But as a statement of confidence, it is hard to beat. Sierra built a business that cannot get paid unless you would have said thank you - and then went and signed up 40% of the Fortune 50 on those terms. The rest of the category is now being asked, one procurement meeting at a time, why it will not do the same.