The Shovels That Sell

A customer-service AI startup most people have never heard of is worth $15.8 billion — roughly 100× revenue. And it’s not even the frothiest bet on the board.

A company most people have never heard of is worth $15.8 billion. Sierra, the customer-service AI startup founded by former Salesforce co-chief Bret Taylor, raised $950 million in May at a valuation that would rank it among the more valuable software firms on earth. It sells roughly $150 million of software a year. That works out to about a hundred times revenue, and Sierra is not even the frothiest name on the board. Across Silicon Valley, investors are pouring billions into one intoxicating bet: that artificial intelligence is about to devour the call center, and whoever builds the replacement inherits a market worth hundreds of billions. The money has already made up its mind. The only question left is whether the technology agrees.

Follow the money

The roster is deep and the cheques are staggering. Decagon, a rival support-agent startup, tripled its valuation to $4.5 billion in under six months on perhaps $35 million of revenue. Berlin’s Parloa tripled to $3 billion in eight. On the voice side, PolyAI answers the phones for the likes of PG&E and carries a $750 million price tag despite still losing money, while Vapi hit around $500 million after Amazon picked it, over 40 rivals, to route every inbound Ring support call through its software. Even the picks-and-shovels layer is ablaze: voice-model maker ElevenLabs raised at $11 billion and is reportedly already fielding offers at double that.

This is not a handful of outliers. Roughly a billion dollars flowed into customer-service AI startups in 2025 alone, part of a wider agentic-AI funding wave several times larger. And the believers can point to results, not just slide decks. Retell, a voice startup, reached an estimated $60 million in annual revenue on barely $5 million of seed money. When the economics work, they work spectacularly.

The empire strikes back

Here is the first crack in the story. The startups’ most dangerous rivals are not the aging call-center giants they are trying to bury. They are the incumbent software titans, who are simultaneously shipping their own agents and buying the challengers wholesale. Salesforce’s Agentforce is already at roughly $800 million in annualized revenue, up 169% in a year. Then the checkbooks opened. Salesforce is buying Intercom’s Fin for $3.6 billion, NICE paid $955 million for Cognigy, and consulting group Capgemini spent $3.3 billion on the outsourcer WNS to rebuild it around AI, with ServiceNow and Zendesk snapping up challengers of their own. The unbundlers, it turns out, are being bundled right back up.

The reckoning

And then there is the awkward matter of whether the technology works at scale. MIT found that 95% of enterprise generative-AI pilots deliver no measurable return. Gartner expects more than 40% of agentic-AI projects to be scrapped by 2027, and reckons only about 130 of the thousands of firms waving the “AI agent” banner are the genuine article; the rest it calls “agent washing.” The cautionary tales keep coming. Klarna replaced 700 agents with a chatbot, then quietly rehired humans when service quality slipped. Air Canada was ordered to pay compensation after its bot invented a refund policy that did not exist.

The economics underneath the valuations are awkward, too. AI agents often run at gross margins of 25 to 60%, well short of the 80 to 90% that makes software so profitable, because every conversation burns expensive model tokens. A hundred times revenue is a bold price for a business that pays a tax on its own success.

Which shovels sell

None of this proves the bet wrong. Gartner also expects AI to resolve 80% of common service issues without a human by 2029, and companies like Sierra and Retell show the demand is real and the savings can be vast. But a gold rush and a good business are not the same thing. In 1849, the fortunes went to the few who struck the richest seams and to the many who sold the shovels; most of the rest went home broke. Some of today’s challengers will define the next era of customer service. Many, at these prices, will be remembered as the ones who mistook a mania for a moat. The survivors will be whoever still has customers when the funding music stops, and can serve them for less than it costs to run the machine.

The question for your business

If AI answered your customers tomorrow, would they thank you, or leave?

Read more thought leadership articles here:

• • Two Industries, One Falling Share Price
  • The 500-Megawatt Bet
  • McKinsey Hired Lilli