Cresta began inside a Stanford artificial intelligence laboratory, as the kind of project researchers build to test an idea rather than to sell one. Nine years later, Salesforce has agreed to buy the company for $2.7 billion, according to people familiar with the matter, in a deal that says as much about where corporate software budgets are heading as it does about the startup.
Salesforce announced the acquisition on Nov. 3 without disclosing terms. The company declined to comment on the price. Cresta makes software that sits beside human agents in customer service centers: it listens to calls as they happen, surfaces answers, scores how each conversation went, and feeds the results back to managers who are trying to decide which of their agents need help and which need replacing.
The founders are Sebastian Thrun, Tim Shi and Zayd Enam, who started the company in 2017. Mr. Thrun is the Stanford computer scientist who led Google’s self-driving car program and went on to co-found the online educator Udacity. He had spent years building artificial intelligence projects inside Google before turning to teaching machines to hold a sales conversation.
Cresta was valued at $1.6 billion in its 2022 Series C round. The price Salesforce has agreed to pay represents a modest step up from that mark, achieved across four years in which the market for enterprise software was repriced twice, first downward as growth slowed, then upward as buyers began paying for artificial intelligence again.
The commercial logic runs through the contact center, one of the last large pools of corporate labor that has not been reorganized around software. A mid-sized company may employ hundreds of agents answering variations of the same twenty questions. The work is measured in minutes, the outcomes are recorded, and the savings from automation can be pointed to in a spreadsheet rather than argued about in a board meeting. That combination is why contact-center software has become one of the first places chief information officers test tools that act on their own.
For Salesforce, the purchase adds a set of deployments and a customer list to Agentforce, the company’s push to sell artificial intelligence agents alongside its core sales and service products. The problem it addresses is pricing. Salesforce built a business on seats, charging each user a monthly fee, and software that does work without a human at the keyboard does not fit that model neatly. Buying a company whose product is already sold by the conversation gives Salesforce a way to charge for outcomes instead of logins.
Salesforce has used acquisitions to enter adjacent markets before, most visibly with Slack, and it has spent the past two years arguing to investors that its own artificial intelligence products are growing quickly enough to justify the multiple at which it trades. The shares have lagged the wider software index for much of that stretch, and each large deal invites the same question from shareholders: whether the company is buying growth it could have built.
The case for buying rather than building rests on time. A contact-center product takes years to harden against the call volumes of a large bank or an airline, where a dropped session is a customer lost before the agent says hello. Salesforce’s sales organization can carry an existing product into accounts faster than its engineers can ship a new one, and the difference of eighteen months is worth more to the company than the difference in price.
What $2.7 billion buys beyond the software is a founder with a public profile and a research network. Mr. Thrun has moved between academia and industry for two decades, and his name opens doors at the large enterprises that buy this category of tooling, which is a useful thing for a salesforce that is trying to sell agents to companies still deciding whether to trust them.
The risk sits in the overlap. Salesforce already sells service software to many of the same buyers, and Cresta’s models will have to work inside a platform designed for an earlier generation of the category. Integration projects of that kind have a habit of taking longer than the announcement implies, and customers who signed with Cresta for its independence will want to know what happens to the roadmap.
There is also the question of what remains scarce. Contact-center automation is crowded with vendors selling similar promises, and the difference between a good product and a mediocre one shows up in the second year, when a customer’s call volume doubles and the model has to handle accents, interruptions and the specific vocabulary of an insurance claim. Cresta’s early research pedigree is evidence that it can build the model. It is not evidence that it can run the deployment at a thousand seats.
For Salesforce, the deal is small relative to its market value and large relative to what it has paid for artificial intelligence talent in the past. It reads less like a bet on a category than a purchase of a position: a set of enterprise customers, a working product, and a founder who can explain to a skeptical chief information officer why an agent should be allowed to speak for the company.
Cresta’s investors, who marked the company at $1.6 billion four years ago, will take the difference in cash or stock depending on terms that have not been disclosed. For customers, the practical question is whether the product keeps the roadmap it had as an independent company, or becomes a feature of a larger suite sold on a three-year contract with the rest of the stack.
The announcement stands on its own for now: a nine-year-old research project, a $1.6 billion valuation in 2022, and a $2.7 billion price in a market that has become considerably more careful about what artificial intelligence is worth.
