Live
Replicant raises $78M Series B led by Stripes Total funding crosses $113M Thinking Machine handles up to 80% of tier-one calls HQ: 1 Letterman Drive, Presidio, San Francisco Founded 2017 by Gadi Shamia & Benjamin Gleitzman Team: ~180 humans, many more voice agents Stack: Python, TypeScript, GCP, Kubernetes Replicant raises $78M Series B led by Stripes Total funding crosses $113M Thinking Machine handles up to 80% of tier-one calls HQ: 1 Letterman Drive, Presidio, San Francisco Founded 2017 by Gadi Shamia & Benjamin Gleitzman Team: ~180 humans, many more voice agents Stack: Python, TypeScript, GCP, Kubernetes
Replicant logo
YesPress · Company File · No. 0429

Replicant.

The voice on the other end of the line - quietly, mostly, an algorithm.

Photographed in the lobby of a Presidio office that smells faintly of espresso and TCP packets. The receptionist was, of course, also a voice agent.

Share on LinkedIn Share on X Share on Facebook Instagram
By YesPress Staff · Filed under Voice AI · Issue 0429 San Francisco, CA

It is 9:14 a.m. on a Tuesday and a woman in Phoenix wants to change her insurance address. She dials a number, listens to one ring, and starts talking. Whatever picks up sounds patient, slightly bored, and entirely human. It is not human. It is, in the parlance of the industry, a Replicant.

The call ends in ninety-one seconds. No hold. No script-reading. No "I'll transfer you to my supervisor." The customer hangs up mildly impressed, mildly unsettled, and goes back to her morning. Somewhere in a server in a Google data center, a few hundred milliseconds of compute log themselves as "resolved."

This is the unremarkable miracle Replicant has spent eight years building, one tedious call at a time.

"Press 1 for billing. Press 2 for hold music that hasn't been updated since 2003. Press 3 for a Replicant." — YesPress, on the user experience that's quietly going extinct

/ 01 — THE NOWWho they are, on a Tuesday

Replicant is a San Francisco company of roughly 180 people that sells voice AI to contact centers. That sentence is technically correct and almost entirely misleading. What Replicant actually sells is the absence of a contact center - or at least the absence of the part of one customers hate. Their flagship product, called the Thinking Machine, picks up the phone, has the conversation, and writes the ticket. The human agents who used to do those calls get to do harder ones, or, in the kinder reading, fewer of them.

The company sits at 1 Letterman Drive, in the Presidio - the former Army base that now hosts Lucasfilm, a Yoda statue, and an improbable number of AI startups. The address is the closest thing to a metaphor the company has: an old institution quietly being repurposed by something newer.

FILED UNDER: Voice AI · Conversational AI · CCaaS · Enterprise SaaS · Customer Service Automation · San Francisco · The Presidio · Series B

/ 02 — THE PROBLEMThe problem they saw, before everyone else did

The contact center is one of the largest, most thankless industries in the world. Roughly fifteen million people work in one. Most of those jobs involve answering the same forty questions in roughly the same order, then doing it again, then again. Turnover runs above 30% annually at many call centers; agent burnout is so reliable it is treated as a line item.

Customers hate it too. The phrase "we're experiencing higher than normal call volume" has, by some accounts, been true continuously since 2003. Everyone is unhappy. Almost everyone has decided to live with it.

Co-founder Benjamin Gleitzman noticed all of this firsthand. As an undergraduate at MIT, he worked the phones at a campus call center. He found the work mostly fine, except for the calls that any decent script could have answered. Those, he thought, were the kind of problem software was supposed to solve. It took roughly a decade and a wave of advances in natural language processing for him to be right.

"The phrase 'we're experiencing higher than normal call volume' has been true, by some accounts, continuously since 2003." — On the slow-motion crisis that became a market

/ 03 — THE BETThe founders' bet

In 2017, Gleitzman teamed up with Gadi Shamia, who had spent the previous four years as COO of Talkdesk, helping that company grow from a seed-stage curiosity into a multi-billion-dollar pillar of the customer service software stack. Shamia knew the contact center as a buyer; Gleitzman knew it as a victim. They started Replicant on a simple, contrarian wager: that voice AI - then mostly a punchline - was about to become good enough to actually pick up the phone.

This was not an obvious bet. Chatbots had been overpromising and underdelivering for the better part of a decade. Most enterprise buyers had a Slack channel devoted to mocking their last one. The team's pitch was that the next generation would be different not because of any single model, but because the whole stack - speech recognition, language understanding, retrieval, telephony - had quietly turned a corner.

The seed investors at Atomic agreed. Norwest Venture Partners agreed louder in 2020, with a $27M Series A. Then, in April 2022, Stripes led a $78M Series B - bringing total funding to more than $113M, with Salesforce Ventures, IronGrey, Omega Venture Partners and Atomic all joining the cap table. The category Replicant had been describing - "contact center automation" - was suddenly something other people described, too.

/ 04 — THE PRODUCTWhat the Thinking Machine actually does

Strip away the demo gloss and the Thinking Machine is, essentially, three things stitched together: a voice front end that listens and speaks, a reasoning engine that decides what to do, and a fistful of integrations into the systems where customer data lives - Salesforce, Zendesk, Five9, Genesys, NICE, whatever the customer happens to run.

It handles billing questions. Appointment scheduling. Account verification. Order status. Returns. Outbound reminders. The boring 80%. It does it in 28 languages and across voice, SMS and chat. When a call exceeds its competence, it hands off gracefully to a human, with the context already populated in the agent's screen - which, by the agent's standards, is closer to a gift than a handoff.

Customers, when they show their math, talk about resolution rates approaching 80%, handle times falling by half, and a slow but real decline in attrition among the agents who remain. Reviewers on Gartner Peer Insights tend to use the word "uncanny," which is meant, mostly, as a compliment.

"We're not replacing agents. We're replacing the part of their job that was making them quit." — a paraphrase of every Replicant pitch deck, more or less

The math of the boring 80%

// Composite figures from customer case studies + Replicant disclosures
Resolution
~80%
Handle time
-50%
Languages
28
CSAT lift
+15–35%
Funding
$113M+

A bar chart, like the company itself, has opinions. These are approximations - useful, not surgical.

A short company on a long phone call

  1. 2017
    Replicant founded in San Francisco by Gadi Shamia and Benjamin Gleitzman, incubated at Atomic.
  2. 2019
    Seed round closes; the first commercial deployments of the Thinking Machine ship.
  3. 2020
    $27M Series A led by Norwest, in the middle of a pandemic that briefly made every contact center the front line.
  4. 2022 · April
    $78M Series B led by Stripes. The category gets a name: contact center automation.
  5. 2023
    Multilingual support and outbound calling expand; deployment time shrinks from quarters to weeks.
  6. 2024
    Thinking Machine layers in LLM-based reasoning; enterprise pipeline broadens into insurance, healthcare and travel.
  7. 2026
    ~180 employees. Quietly handling millions of calls a month for customers that mostly prefer not to be named.

/ 05 — THE PROOFCustomers, money, and other receipts

Replicant's customer roster skews toward the industries with the longest hold times: insurance, healthcare, financial services, travel, retail, consumer brands. The company publishes case studies with marquee enterprises - the Canadian Automobile Association among them - and runs deployments at scale at carriers most of us have on speed dial.

The financials, to the extent a private company shares them, are tidy: roughly $24M in annual revenue at last public estimate, growing inside a contact-center software market that PwC has measured north of $20 billion globally. The customer logic is, by enterprise software standards, refreshingly simple. If Replicant resolves a million calls a year and each saved minute saves a dollar, the math basically does itself.

$113M+Total funding
2017Founded
~180Employees
28Languages
80%Tier-one resolution
SFPresidio HQ

/ 06 — THE MISSIONWhat they say they're doing

Replicant's official mission, as written on its About page, is to "use AI to resolve as many conversations as possible, while elevating human agents to do what they are uniquely good at." It is the kind of sentence that, in a less interesting company, would be wallpaper. Here, it is operative. Every product decision is a vote on which calls are routine enough to automate and which deserve a human - a question that is harder than it sounds and gets harder as the AI gets better.

The company is careful, in interviews, not to oversell. Shamia tends to talk about "agent augmentation" before "agent replacement." Whether one believes that distinction holds depends on whether one is the agent, the executive, or the shareholder. So far, attrition data from customer deployments suggests it holds enough.

"Voice AI was a punchline for a decade. The joke landed. Then somebody picked it up." — Filed under: things that happened slowly, then quickly

/ 07 — TOMORROWWhy this matters next

There are two ways to read what Replicant is building. One is mundane: a better mousetrap for the call center, a software line item that helps the CFO. The other is structural: the contact center is one of the largest remaining manual workflows in the global economy, and Replicant is one of a small handful of companies methodically dismantling it. Either reading is interesting. Both are probably true.

The harder question, which the company is too polite to answer aloud, is about labor. If the Thinking Machine keeps getting better - and it appears to - then somewhere down the line the math that currently rewards "augmentation" is going to start rewarding something simpler. Replicant's bet, to its credit, has always been that the calls remaining will be the calls that humans actually want to take. We will know in about five years whether that turns out to be true.

/ 08 — BACK TO THE BEGINNING9:14 a.m., Phoenix, redux

The woman in Phoenix did not, technically, talk to a person. She also did not, technically, mind. She got her address changed. She did not hear hold music. She did not press 1, 2, or 3. She did not get transferred. She did not call back.

The Replicant on the other end of her call did roughly three thousand of those Tuesday morning, before lunch.

It is a small thing. It will probably be most of customer service in ten years.