Bret Taylor and Clay Bavor have sterling Silicon Valley pedigrees, with leadership roles at the biggest tech companies in the world. Now they’re in the startup trenches, on a quest to make AI agents the primary way companies interact with their customers.
Bret Taylor, CEO of Sierra, a $10 billion startup that builds AI customer service agents, is grinning as he blows into a massive alphorn stretched before him, an 11-and-a-half foot horn made of California redwood. The wail is at first shaky, but he eventually produces a sustained and clear tone. “It requires a little bit of a lesson on how to blow like a trumpet,” he later tells Forbes. “It’s so awkward and goofy that it’s just utterly perfect.”
Proper alphorn technique might not be the expected topic of conversation with Taylor, one of the most celebrated executives in Silicon Valley, with noted tenures at Google (co-creator of Google Maps), Facebook (CTO), Twitter (board chairman), Salesforce (co-CEO) and now OpenAI (board chairman). But it’s on the docket today because he and cofounder Clay Bavor, an 18-year Google veteran who headed the company’s emerging tech efforts in virtual and augmented reality, decided soon after they started Sierra in 2023 that they didn’t want to commemorate signing new customers by ringing a stuffy old sales gong — a cliche at other enterprise tech companies. Instead, they opted for something in line with the startup’s mountainous corporate identity. (They even list free alphorn lessons as an official benefit on job postings, and purchased the instrument, coincidentally, through a company called Sierra Alphorns.)
Taylor says they’ve blown the horn “hundreds” of times, an indication of the health of their customer service business, though the company won’t share an exact number of clients. They include a mix of startups and large consumer brands, including retailer The North Face, electric vehicle maker Rivian, home security giant ADT and digital radio company SiriusXM. That particular slice of the market is intentional: The company is going after big business, touting that over half of its customers have revenue of more than $1 billion, and 20% of them have revenue of more than $10 billion. That focus has paid off, with Sierra on track to exceed $100 million in annualized revenue by the end of the fiscal year in January, according to a source familiar with the company’s performance.
If you’ve interacted with one of Sierra’s agents, there’s a good chance you were already annoyed. That’s because they take care of the more knotty customer service issues — like returning a pair of shoes or canceling a subscription — that typically required chatting with a bot on a little popup screen on a corporate website, or worse, needing the attention of a human representative. But Taylor and Bavor don’t think of Sierra as a business that only helps handle customer complaints. They think agents will become so powerful that they’ll eventually become the primary way that businesses interact with customers. You’ll engage with them across several channels — text, phone call, app, WhatsApp and more. And they’ll become personal concierges working on behalf of a company, remembering past conversations with customers and making suggestions based on their personal preferences. “One of the promises of the internet was personalization, but the only full manifestation of that we’ve really seen is very targeted ads,” said Bavor, with no tinge of irony for someone who spent almost two decades at Google.
Got a tip for us? Contact reporter Richard Nieva on Signal at username RNieva.26 or over email at rnieva@forbes.com.
For now, though, agents can still be pretty rudimentary. On Wednesday, the company announced a handful of new products designed to make its agents act more like multi-purpose concierges. The goal is to not only have their agents tend to reactive issues like complaints. Instead, they want agents to pop up proactively when a company thinks it makes sense. For example, Bavor explains, when you land abroad, a phone carrier’s agent could offer you a promo of free gigabytes — instead of the customary text reminding you of data roaming charges.
Blue chip investors like Sequoia, Benchmark and Thrive Capital are sold. In September, the startup raised $350 million in a round that valued the company at $10 billion. “What differentiates them is a high ceiling,” Neil Mehta, founder of the round’s lead investor Greenoaks Capital, told Forbes. “There’s really no company that could solve the sophisticated problems that they’re solving for some large companies in America.” The infusion of funding has also paid off for Taylor and Bavor, officially minting them as billionaires for the first time, with each owning a roughly one-quarter stake in the company, according to Forbes estimates.
But back to the alphorn. It’s not only a gimmick. It’s also supposed to symbolize how Sierra is trying to stand out in an already fiercely competitive market, which includes $1.5 billion-valued Decagon, a buzzy startup that’s attracted a frenzy from venture capitalists clamoring to invest; Kustomer, a smaller New Jersey-based rival; and Intercom, an established competitor with thousands of customers. The field is crowded because it’s one of the few sectors that is primed to immediately reap the benefits of AI, said Taylor, and the market is sizable: more than $12 billion in 2024, on track to hit nearly $50 billion by 2030, according to the research firm MarketsAndMarkets.
Even Taylor’s old boss Mark Zuckerberg is getting into the game. In an interview, Zuckerberg pointed out that Meta, too, builds AI business tools, but focuses more on small companies rather than the large enterprises Sierra is targeting. “Part of when [Taylor and I] connect these days, we’re just sharing notes on what we see, building what are actually complementary products at different parts of the market,” Zuckerberg told Forbes. Customer service is a unique sector because you’re selling to big companies, but your products are aimed at helping the common consumer. Taylor’s resume working at the disparate tech giants — Google, Facebook, Salesforce — makes him particularly suited for that task. Said Zuckerberg about Taylor: “The consumer taste and the enterprise taste and the technical taste — all of that, I think, go together to build something good here.”
As the AI gold rush attracts legions of younger than ever founders across the world to San Francisco in search of fortune, Taylor and Bavor are an increasingly unlikely pair: Longtime Bay area residents with more than 40 years combined working for some of Silicon Valley’s most successful tech giants. Born in Oakland, Taylor’s mom was an executive at Chevron and his dad was a mechanical engineer in the HVAC business. Bavor, the son of a cardiologist and quiltmaker, grew up in Mountain View, Google’s backyard. Both, coincidentally, first got into tech by building websites for friends and small businesses while in high school.
Taylor, 45, and Bavor, 42, met at Google, where they were both hired for an associate project manager program by Marissa Mayer, who would go on to become CEO of Yahoo. It was both of their first jobs after college — Taylor went to Stanford and Bavor to Princeton — and the two bonded during monthly poker nights. From there, their paths diverged. Bavor spent the next 18 years at Google. Early on, he led product and design teams for Gmail, Docs, Drive, and enterprise apps. He later developed Google’s virtual reality platform Daydream, and its next-gen video chat service Starline, which renders 3D versions of call participants, as if they were in the same room. But it was a quirky launch in 2014 called Google Cardboard, a DIY kit made of cardboard that turned a smartphone into a makeshift VR headset, that Google CEO Sundar Pichai remembers most fondly, he told Forbes. “It captures this sense of Clay, in that he has this wonder,” Pichai recalls. “As if you’re in the garage and tinkering.” (Starting their own company “felt like a once in a lifetime opportunity” for Bavor and Taylor, said Pichai, who worked with both separately.)
Taylor’s career, meanwhile, has made him the Forrest Gump of the tech world, involved in some of the most pivotal moments in Silicon Valley’s recent history. At Google, he famously coded an early web interface prototype of what would become Google Maps over the course of a weekend. In 2007, he left to found a social media service called FriendFeed, acquired by Facebook two years later for nearly $50 million; he became Facebook’s CTO and helped develop the “Like” button. After that, he founded Quip, a word processing app he sold to Salesforce, where he rose to the ranks of co-CEO alongside founder Marc Benioff. Meanwhile, as board chairman of Twitter, he squared off with Elon Musk as the billionaire Tesla CEO fought to take over the iconic social site. Perhaps biggest of all: he is now chairman of OpenAI, a position he landed in 2023, after a power struggle at the company engulfed the entire industry.
Zuckerberg said he and Taylor still keep in touch, grabbing dinner or meeting up at Allen & Company’s annual conference in Sun Valley, Idaho, often described as summer camp for billionaires, where they go on walk-and-talks. But as Zuckerberg went on a high-profile spending spree earlier this year to poach the industry’s creme de la creme of AI talent, he opted not to pursue Taylor, he told Forbes. Taylor had spent years at big companies like Facebook, Google and Salesforce, Zuckerberg noted, and wanted to return to creating his own companies. “I just respect that he’s trying to build his own thing,” he said. “But you know, life is long, and I’m sure we’ll intersect at some point, one way or another.”
Soon after Taylor left Salesforce in early 2023 to “return to his entrepreneurial roots,” he grabbed lunch with his old Google colleague Bavor. Over a grilled whole branzino at Evvia Estiatorio, a tony Mediterranean restaurant in Palo Alto, the two geeked out about ChatGPT, released a few weeks earlier. OpenAI’s new app caused the “proverbial deck of cards to be shuffled,” Bavor said, and the pull was so strong that it lured him away from the only place he’d ever worked in adulthood. “I almost left a few times to start a company,” he said. “But the planets never quite aligned.”
The duo left the meal knowing they would found an AI startup together, but didn’t know exactly what it would be. They began pulling from their collective rolodexes and reached out to every CEO and business leader they could think of, asking each of them to name the top three specific challenges they have in running their business. Anthony Tan, CEO of Grab, the Uber of Southeast Asia, mentioned the issues he was having with customer service, particularly because of all of the language differences in the region, and their dealings with various stakeholders like riders, drivers, merchants and other partners. “It was one of those conversations where it just crystallizes in your head,” Taylor said.
Some Sierra clients say the company is already making their offerings more capable. Rocket Mortgage CEO Varun Krishna said its Sierra-powered agent can refinance a mortgage in about 30 minutes, when previously it would have taken “several hours over multiple days.” Eric Glyman, CEO of Ramp, the corporate credit card company, said Sierra agents run support chats through the Ramp app, with 90% of requests handled without the help of a human.
It’s a good thing, too, because Sierra doesn’t get paid when its bots have to hand the reins over to human reps. The company uses an “outcomes-based” business model, which means it only charges a fee if its bots resolve a task completely on their own. The startup isn’t the only one to use that business model; rivals Intercom and Zendesk use it too. Sierra won’t disclose what it charges per resolution, only noting that the fee varies based on the complexity of the task. By comparison, Intercom charges 90 cents for most resolutions.
Now Sierra, with a growing team of more than 300 employees, is announcing new products it hopes will take its agents to the next level. One new feature allows companies to plug their agents directly into ChatGPT, so they can answer customer queries directly through OpenAI’s popular chatbot. Another feature called LiveAssist works as an AI copilot for human call center workers, assisting representatives. The feature can pull up answers for them from FAQs, check internal databases for policies or recommend next steps in real time. Perhaps most significant, a new service called Agent Data Platform gives agents the memory of past conversations with customers. That means that it combines data from what a customer has said in chats, emails, calls and texts with a company’s internal data from its billing systems, warehouses and transactions. “It’s starting a conversation on second or third base, as opposed to from scratch,” said Bavor.
Outside the conference room window of Sierra’s headquarters in San Francisco’s SoMa neighborhood, Salesforce Tower looms large. Taylor downplays any competition with the company he formerly co-ran, saying he foremost thinks of the tech giant as an integration partner. He won’t comment on Agentforce, his former employer’s rival platform for customer service agents, which has struggled to gain momentum. (Meanwhile, Sierra is reportedly on the move, getting ready to lease a new 300,000 office space in the city’s South Beach neighborhood.)
As advanced as agents are getting, they still make simple mistakes. In a demo, Sierra Head of Product Zack Reneau-Weeden called Sirius XM’s customer service agent, powered by Sierra, using the 1-800 number live on the company’s website. Reneau-Weeden role played as a customer interested in changing their subscription plan. At Forbes’ prompting in an effort to stress test the agent, he yelled “Don’t touch that!” mid-sentence, as if he were a parent scolding a child, to emulate real world chaos. The agent got tripped up and began to transfer him to a human agent. Afterward, he reviewed the call log in Sierra’s backend. “Right now, that was misclassified as an interruption, but should have been ignored,” Reneau-Weeden said.
When told about the demo, SiriusXM COO Wayne Thorsen cuts the agent some slack. “People make mistakes too,” he said. Voice interactions between customer and machine can still be tricky, but he said the agents have cut down on most errors like misinformation or policy mistakes. There’s also the issue of people who flat out refuse to engage with AI bots, and will do anything in their power to get their call transferred to a human. “That’s going to be a problem for a while,” he said.
The founders are so relentless about errors that they hold board meetings around every six weeks instead of quarterly, and begin each session with a list of what they need to fix, said Sequoia partner Ravi Gupta, a Sierra board member. Even competitors give them their due: One CEO of a rival customer service company conceded that Sierra has a sound strategy with its white-glove approach to courting large enterprises. But as agents become a more crucial part of a company’s identity, the CEO doubted why they would entrust that responsibility to a vendor like Sierra, instead of getting a more off-the-shelf product that they could manage more easily in house. “The big question is durability,” they said. “I foresee a great philosophical war between us and Sierra, and it’s going to be really fun to see who the market chooses.”
A few months after Taylor and Bavor started Sierra in 2023, OpenAI stunned the tech world. In a bombshell move a few days before Thanksgiving, its nonprofit board ousted CEO Sam Altman. A dramatic tug of war ensued. When the dust settled five days later, Altman was reinstated and Taylor was named the company’s new chairman of the board. But when he was offered the job, it wasn’t a foregone conclusion.
“My first call was to my wife, and my second call was to Clay,” Taylor recalled. He pored over the decision, fearing it would take him away from Sierra at a formative time in its infancy. Bavor gave him his emphatic support, and he accepted the offer. “To some degree, we probably wouldn’t have started Sierra if not for OpenAI,” said Taylor. “It’s just such an important organization, and we felt indebted to it — for creating the market that we operate in with the GPT models and ChatGPT.”
He acknowledges it takes some of his time away from Sierra. “But what’s been really great about it is not only being able to be close to the world’s greatest research lab, but also the most important consumer service in AI.”
He considers that all a positive for Sierra, and is careful to draw a distinction between his roles at both. “They’re pretty different jobs,” he said. While he’s primarily focused on building products at Sierra, at OpenAI, whose overarching goal is achieving AGI — or artificial general intelligence, tech-speak for when AI will match or surpass human capabilities — he’s more focused on “inventing that future” and learning from Sam Altman and his “degree of ambition.”
Still, inventing the future isn’t limited to AGI. If agents are going to become as important as corporate websites, like Sierra believes, then the business implications are lofty. The company is investing heavily in voice technologies, working with model makers including Google, OpenAI, and London-based startup ElevenLabs. Sierra has an engineer they’ve given the unofficial title of “voice sommelier,” who mixes and matches elements of models to create voice personas for brands, said Bavor.
Looking out even further, not only will every company have AI agents, Taylor argues, but consumers will have their own agents as well to act on their behalf. So Sierra is preparing for a future where agents are interacting with other agents, he said. There’s also the uncomfortable prospect of job displacement, and what happens to call center workers in a world dominated by AI bots. It’s a concern that extends beyond customer service, and Taylor warns that companies have to be involved in reskilling workers.
Sierra’s vision of customer service is all about balance. If every company has proactive agents, there’s a scenario where hundreds of businesses will begin to bombard consumers daily with nonstop promos and offers, a constant din of corporate alerts. Sierra said it will work with its clients to help find the right moments for its agents to pop up. The goal is for the experience to be “personal and proactive, but not pushy,” Taylor said. “The remarkable thing about these technologies is I think they can have that level of nuance,” he said. “If we’re doing our job, you’re not being spammed.” It could be harder than scaling the Sierras.
Read the full article here
