Somewhere right now, a customer is on the phone, and an agent is trying to keep them. The hold music has done its work. The script is wilting. The screen has 14 tabs open. And in the corner of that screen, a small panel quietly types out the right answer half a second before the agent needs it. That panel is Cresta.
It does not look like a revolution. That is the point. Cresta has spent the better part of a decade making AI that sits next to a human instead of replacing them - and then made the financial case for why that arrangement is more interesting than the one everyone keeps tweeting about.
01 / The ProblemThe world's hardest job, automated badly
Contact centers are the unglamorous spine of modern commerce. They handle returns, retentions, claims, complaints, debt collection, account changes, password resets, and the occasional existential outburst. They employ roughly fifteen million people. They are, by most measures, the place customer experience goes to be tested and frequently to die.
The industry's previous answer was a chatbot. You have met it. It asked you to type your account number, then asked you again, then transferred you to a human who asked you a third time. The chatbot did not solve the problem. It moved the problem one menu deeper.
Cresta's founders looked at this and concluded the opposite was true. The bottleneck was not too many humans. It was that the humans had no help.
A new agent at a major contact center is typically expected to be fully productive in ninety days. Average tenure at many centers is under twelve months. Do the math.
02 / The BetThree founders, one stubborn idea
Cresta was founded in 2017 by Zayd Enam and Tim Shi, two researchers who walked away from a Stanford PhD and a stint at early OpenAI, respectively, with Sebastian Thrun - the man who built the first DARPA-winning self-driving car, started Google X, and co-founded Udacity - signing on as a third co-founder and chairman. It is not the most modest cap table in San Francisco. It also, in fairness, may be the most credentialed.
Their bet was specific. Real-time AI - not summary-after-the-fact AI, not chatbot AI, but a model running live alongside a human agent - could compress the months it takes a new hire to become an expert into days. The pitch had a tagline that aged better than most: experts on day one.
Investors agreed, eventually with enthusiasm. Greylock wrote the seed check. Andreessen Horowitz led the Series A in 2020 with a public note arguing Cresta was building the canonical real-time AI company. Sequoia and Tiger Global piled in. By late 2024, the cap table had quietly grown to include the Qatar Investment Authority, Workday Ventures, Accenture, Qualcomm, J.P. Morgan and World Innovation Lab - a list more typically associated with infrastructure than software.
03 / The ProductFour pieces of software in a trench coat
There is not one Cresta product. There are, depending on how you count, four or five, and they are designed to interlock so completely that customers tend to describe them as a single thing.
Agent Assist
Real-time prompts, suggested answers, knowledge surfacing and auto-summarized notes while the call is still live.
Director
An AI-graded scorecard and coaching console for managers. Replaces the spreadsheet quietly killing your QA team.
Insights
Conversation analytics that surfaces emerging customer trends, churn signals and revenue gaps from every call.
AI Agent
A generative virtual agent that handles end-to-end voice and chat - and hands off cleanly when it shouldn't.
Four products, one platform, an organizational chart that finally makes sense.
The clever bit is the seam. Cresta's same underlying model watches every call, learns from the best agents on the team, and uses what it learned to coach the rest. The AI Agent and the human agent share a brain. When the AI escalates, the human picks up the conversation already half-solved. When the human handles it, the AI is taking notes and turning them into next week's training material.
The receipts, in chronological order
04 / The ProofNumbers that survive a second look
Enterprise AI is a category overrun with pilot programs that never quite become contracts. Cresta's defense against that fate is straightforward: it sells outcomes, and it gets them. Customers report double-digit lifts in conversion and retention, single-digit reductions in average handle time, and the abolition - or close to it - of after-call work. That last one matters more than it sounds. The minutes an agent spends typing notes after each call are the silent tax on every contact center on earth.
What $401M looks like when you spread it out
The customer list tells the same story from a different angle. Hilton, Marriott, United Airlines, Alaska Airlines, Cox Communications, CarMax, CVS, Intuit, Verizon, Porsche, Brinks Home, LendingClub, Oportun and Snap Finance all run some flavor of Cresta in their contact centers. These are not logos a startup acquires lightly. Procurement at a Fortune 500 telco is its own kind of qualifying exam.
05 / The MissionHuman-centric AI, said with a straight face
Cresta brands itself as human-centric AI, which in 2026 is the kind of phrase you have to defend. The defense, in their case, is the product. The Agent Assist panel does not write the agent's job out of existence. It reads the room with them. It hands them the right knowledge article ten seconds before they would have asked for it. It quietly grades the call so the manager does not have to listen to forty randomly sampled recordings a week and decide who needs coaching.
The company's stated mission is that every employee should be an expert on day one. Read the small print and a second mission appears - that contact centers should stop being treated as cost centers and start being treated as the place enterprises actually meet their customers. Both of these are easier to write than to deliver. Cresta has been delivering anyway.
The question for the next decade is whether AI replaces the agent or upgrades them. Cresta has staked $401M on the second answer - and is winning the argument quarter by quarter.
06 / TomorrowWhy this matters past the next earnings call
Every enterprise software wave has had a category that was unsexy until it wasn't. CRM was paperwork until Salesforce. HR was a filing cabinet until Workday. Contact center software was a punchline until, well, now. The data is there. The models are there. The buyers are there. The unanswered question was whether anyone could ship a real product into a regulated, voice-heavy, latency-sensitive environment and make it stick. Cresta did.
That puts the company in an unusual position. It is not chasing a wave. It is the wave - or, more accurately, it is what arrives after the wave breaks and the procurement cycles begin. The contact center is a $20B+ software market that has been waiting, with admirable patience, for someone to actually fix it.
So. Back to that customer on the phone. The call ends. The agent did not have to put them on hold. The notes wrote themselves. The manager will get a coaching summary in the morning. The customer, against their own prediction, will renew. Multiply that small, ordinary outcome by a few hundred million calls a year and you get Cresta - the AI company quietly doing the unsexiest thing in the entire industry, and quietly being right about it.
The chatbot moved the problem one menu deeper. Cresta moved it off the call entirely.