Brian Schiff

The through-line

Schiff's recurring argument is that the parts of a voice AI business people notice are the parts that matter least. The model layer is commodity, the name is an afterthought, and the word "AI" is now so loud it has become noise. What actually decides whether a customer call gets resolved is the unglamorous plumbing underneath, the workflows and the integrations into the systems a brand already runs on [1]. And what decides whether the company keeps working at all is the team, which he treats as the real product of the scale-up stage [1]. Both halves come from the same instinct: value sits one layer below the visible surface.

The other constant is timing. He has been building phone automation since the Alexa era, when the pitch was met with disbelief, and he is candid that the arrival of large language models was less strategy than fortune [1]. That shift has moved his ambition from proving the concept exists to claiming a durable position in a category he now expects to be crowded [1], and it has changed what he thinks the technology is for, from cutting support costs to running the phone as a channel that can drive growth [2].

On what actually makes voice AI work

The product he describes is a replacement for the phone tree everyone hates. "We replace those with a modern voice AI experience" [1], aimed at the visceral, universally understood frustration of calling customer service, a reaction he says you would get from all ten of the first ten strangers you stopped on the street [1]. He breaks the system into three layers: the AI layer that understands people, the workflows, and the integrations. The first is not a differentiator, because "Everybody has access to the same models" [1]. The differentiation lives in the second and third. Cancelling an order means knowing the specific logic a brand uses to decide whether cancellation is allowed, what steps complete it, and which systems have to be updated to execute it. Get that wrong and you have shipped "another robot that sort of sounds nice but has very little utility" [1].

That makes integration depth the measure of value, not conversational polish. A brand that turns Flip on with no integrations and lets it answer basic questions "you're going to get very little out of it"; add the order management system and returns and subscriptions and it compounds, so "the more integrations a customer turns on with Flip, the better the results will be" [1]. His useful observation is that the categories of reasons people call map almost one to one onto the SaaS tools a brand already uses, Shopify, Salesforce, subscription platforms, returns platforms, which means the tech stack of the next customer is usually one Flip is already plugged into rather than a bespoke build each time [1].

On focusing by vertical

Flip is deliberately narrow: retail and e-commerce as the main market, transportation, and a third vertical in progress, and only over the phone [1]. Schiff frames the constraint as what lets customers "step in, turn it on really easily and get amazing results from day one" [1], because the workflows and integrations that make the product useful are only reusable within a vertical. The choice of e-commerce was reasoned the same way. When automating support calls with AI still sounded like science fiction, financial services and healthcare were unlikely to experiment, while e-commerce brands would take new technology for a spin if it improved something they cared about [2]. Geography follows the verticals too: the UK office exists because the ground transportation industry, still a meaningful part of the business, runs out of the UK, where all the major platforms are headquartered [1].

On being early, and the LLM windfall

He is unsentimental about how much of this was luck. Asked about skill versus luck, his answer is that you need an abundance of both, and that "the biggest stroke of luck was just being in the right moment at the right time when LLMs came into existence and commercial readiness" [1]. What made them usable in production was not raw intelligence alone but "that magical combination of quality, speed, and cost effectiveness" [1]. The founding sequence bears out the earlier bet: an Uber knockoff for Cornell built while Uber was not legally cleared for upstate New York, then the realisation during the Alexa moment that the technology, imperfect as it was, was already better than the robots people got when they called customer service [1][2]. They took that back to the taxi owners whose calls they already understood, proved it out in ground transportation, and raised ten million dollars in 2022 to go after retail and e-commerce [1].

Just as valuable as the model improvements, in his telling, is what happened to market perception. Five years ago "the idea that AI was going to automate most of your support calls, that was like controversial" [1], and "it's really hard to go and and change hearts and minds by yourself" [1]. The last two years of AI enthusiasm did that education for him.

On AI as a category and the noise around it

His stated unpopular opinion cuts against his own business: "I am like so burnt out from all the AI talk" [2]. The irony is that for years the company avoided the label entirely. "We didn't want to call it AI" [2], because it read as complex and futuristic; they positioned it instead as an automation tool driving business outcomes in a channel that is hard to handle manually [2]. Only after ChatGPT made AI something people wanted to be associated with did they go back to the drawing board and make peace with the term, without bolting it onto the name [2].

On competition, he expects the category to mature the way mature categories do: a handful of companies that own the market and are synonymous with it, plus niches and bolt-ons attached to other products [1]. He wants Flip to be one of the household names that emerges, and treats that position as available rather than assured [1].

On the second wave of automated conversations

He splits the market into two phases. Roughly three quarters of current customers bought to save money and get people quick answers, and that is the whole of their thinking today [2]. The second wave, as he sees it, is brands realising they can hold high quality conversations with customers constantly, conversations the customer initiated, and orchestrate them toward outcomes: growth, retention, loyalty [2]. His concrete example is someone calling to ask when an order will arrive. Tell them, offer the tracking link, and also offer to enrol them in the SMS marketing programme, so a support call produces a genuine subscriber who actually wants to be there, and the CX team has delivered growth in a fully automated way [2]. He describes the endpoint as treating the phone the way brands already think about orchestrating their other channels [2].

Subscription brands are where this is sharpest today, because cancellations are a huge call driver. His position is careful: "you do not want to introduce friction in the cancellation process", while still looking for the adjustment that serves the customer better than cancelling, where that genuinely is in their interest [2].

On the team as the thing you're building

His pitch rests on a simple claim about post-product-market-fit companies: "the most important thing that we're building is the team" [1]. The arithmetic behind it is that as a company goes from a couple of founders to a couple of thousand employees, the share of total work done by the founders falls continuously, so leverage moves entirely into the people doing the work and the culture they operate in. Hence the northstar that "the team should be the best thing that we're building" [1]. Consistent with that, he does not think candidates should be sold by him. The signal comes from talking to the people you would work with daily and hearing how they talk about the business [1]. The second leg is the work itself: a problem everyone recognises viscerally, met by a technology that changes every day, with millions of production calls a week going through the platform, and an explicit strategic requirement that people stay at the cutting edge and keep experimenting [1]. The company is 40 people across New York, LA and the UK, having more than doubled in the previous year with more growth planned [1].

On hiring, culture fit and how he interviews

Skill has to be there as a baseline, but he says around 95 percent of what they recruit for is culture fit [1]. The profile is specific: people "sick of having four walls and a ceiling surrounding them", frustrated that nothing is happening around them, who want to build something real and fast [1]. He also wants a particular kind of breadth, "a level of generalist, maybe not in skill set, but at least in care" [1], meaning salespeople who care about the product and relay insight back from customers, and engineers focused on the outcome the technology achieves rather than the technology itself, with empathy for who the customer is [1]. The same quality shows up in how work gets done: many improvements never reach a sprint planning cycle, because people notice something wrong and fix it without needing to be celebrated for it [1].

His interview technique is deliberately plain. There is no Elon Musk riddle [1]. He asks how someone got to where they are, what they want next, and where they see themselves in a decade, then gets out of the way: "just kind of shut up for a few minutes and let the person talk" [1]. What he listens for is how deeply they understood what they did, what was magical about it and what was frustrating. He is explicit that "I want to learn, I want to grow" with no ten-year plan is a perfectly good answer [1]. The strongest positive signal comes from radical honesty about the job: "We are abundantly transparent about what working at this company at this stage of the journey is and isn't in the interview process" [1], and the candidates worth hiring get more excited as the non-glamorous parts come out. He insists "we really do view it as a matching game" [1], right people for the company and right company for the people, with no implication that one path is better than another.

On prioritisation and winning two games at once

The line he found himself repeating through a product hiring cycle: the team could write down its ten best ideas today and eight of them would be in the product three years from now. So "the job of the PM is not actually discerning good ideas from bad ideas. It's discerning what are the right ideas for right now and the sequencing of those ideas" [1], finding where the leverage is so that today's work lays the foundation for what has to come after. He is blunt that there is always too much to do and too few people, and that the hardest thing about the scale-up stage is duality: "The rules of the game are you have to win short-term and long term" [1]. If you only had to hit the quarter, or only the five-year goal, decisions and execution would both be easier. He calls it an extreme environment that suits a particular kind of person [1]. Decision-making itself is deliberately uneven, small calls made in the moment, bigger ones involving more discussion and more people, held together by people incentivised to want the right outcome for the business [1].

On names, rebrands and not overthinking them early

Schiff has renamed the company twice, from Red Route to Flip CX to Flip, and he is amused by it [2]. His view of naming at the start is dismissive in a useful way: "the last thing that you're thinking about when you're getting a business off the ground is like what the name of it is" [2], so they recycled the Cornell rideshare name and worried about it once there was something worth naming, at which point the marketing team told them it had to go [2]. The eventual rebrand was timeboxed to a week of evening sessions, and the group's favourite candidate was killed by the trademark lawyers, which he counts as a relief [2]. The criteria that survived: one word, spelled the way it sounds so it is easy to search, fun and upbeat, and suggestive of dramatic change [2]. Dropping the CX later was, in his view, long overdue, partly because it made the company hard to find [2]. He treats launch marketing with the same willingness to be undignified, running for as long as he was on hold with an airline once signups hit a threshold, and dyeing his hair neon green when registrations passed 500 [2].

Takeaways

• The model layer is not a moat; workflows and integrations into a brand's order management, returns and subscription systems are what separate a useful voice agent from "another robot that sort of sounds nice but has very little utility" [1].
• Value compounds with integration depth, so a customer who enables only basic Q&A will get very little, and each additional system connected makes the product materially more powerful [1].
• Vertical focus is what makes the product turnkey, because reasons for calling map onto a predictable SaaS stack that is largely shared across brands in the same industry [1].
• E-commerce was chosen over financial services and healthcare because those brands would experiment with unproven technology if it improved something they cared about [2].
• The next phase of CX automation is growth rather than cost saving: resolving a "where is my order" call and converting it into a genuine SMS subscriber, or turning a cancellation into a better-fitting adjustment without adding friction to cancelling [2].
• Post product-market fit, the team is the primary thing being built, because the founders' share of total work falls with every hire and all leverage moves to the people and culture [1].
• Skill is table stakes and roughly 95 percent of recruiting is culture fit, with a preference for "generalist, maybe not in skill set, but at least in care" across sales and engineering [1].
• Total transparency about the unglamorous parts of the job is the best filter: candidates who get more excited as the dirty laundry comes out are the right match [1].
• Prioritisation is sequencing, not selection, and the scale-up test is that "you have to win short-term and long term" at the same time [1].