Alex Sambvani

The through-line

Sambvani's whole argument starts from one unglamorous statistic: restaurants miss roughly half the calls that come in, and every one of those is a guest and a piece of revenue walking out the door [2][3][5]. He frames this as a hospitality failure before it is a technology problem, since "your restaurants are in the hospitality business it's not good Hospitality to to like ignore half of your customers" [2]. The fix, in his telling, is not a replacement for staff: "we're not about like replacing the human touch like that always needs to be there were about using AI to fill in the gaps in the experience today" [2].

What has shifted over the years is his confidence in how far that gap-filling can go. Early on the company avoided saying the product was AI at all, for fear of alienating buyers [6]. Now he leans into it, and he reports a threshold being crossed: the CEO of a well-known steak brand told him the AI was "better than the average host within his restaurant group," something he says he had never heard in almost seven years of running the business [1]. The ambition has widened in parallel, from a phone answering tool to what he describes as an AI staff member operating across voice, text, email and web, wired into the rest of the restaurant's stack [5].

On the missed-call problem

The numbers do most of the persuading. Around 60% of calls to a restaurant that takes reservations are reservation-related, with the remaining 40% being common questions about location, hours, dress code, parking or the menu [2][3]. On average 50% of calls go unanswered, and 50% of restaurants still use landlines, so there is no automation or sophistication in the channel at all [2][3]. Missed calls cluster exactly where they hurt most, during peak service and after hours, and he points out that it is simply impossible for staff to answer five simultaneous calls unless the restaurant has five hosts [1][3]. By the time a restaurant calls back, the guest has usually booked somewhere else [1].

He also argues the phone carries disproportionate value. It is a high-intent channel: "If they're calling you, they want to do business with you" [1]. Average party size over the phone runs between four and five covers, against two to three online, so the calls being dropped are the larger tickets [4]. Private events leads are the sharpest version of this, which is why the Tripleseat integration exists, and why he insists that "in private events speed is everything" [1]. Dietary questions matter for the same reason: a guest who cannot confirm a gluten-free option is a guest who will not drive over [1].

On transparency and the uncanny valley

He holds a firm line that guests should always know they are speaking to an AI. "I personally believe that transparency is everything" [4], and the goal is to make the AI "feel as humanlike as possible while making it still clear that it's an AI" [5]. The justification is both ethical and practical. Tricking a guest destroys trust, and he cites research that a user who knows they are talking to an AI will actually be more successful with it, because they speak to it differently and get down to business rather than telling their life story [3][5]. He predicted that voices would become indistinguishable from humans within a year or two, creating a new disclosure problem the industry would have to solve [3].

The current frontier, as he describes it, is emotional rather than acoustic: making the system understand when a guest is upset and respond with empathy rather than matching their energy, which is what makes an interaction feel natural in the gut [5].

On conversation design as a real discipline

Sambvani treats conversation design as the core craft of the company, not a garnish. "We think about conversation as an interface just like we think about a website as an interface" [4], which means someone has to think through every word, every pathway, and what happens when things go wrong and a guest needs to be brought back on track [4]. Slang employs three people focused entirely on this, led by an expert who also teaches a college course on the subject, and their job is to stop the AI from doing things a human never would, such as firing five questions in rapid succession [1][4]. He estimates fewer than 500 people in the world, possibly fewer than 200, are genuinely good at this work, and that they arrive from linguistics, writing or visual design backgrounds because the job demands command of language plus standard user experience thinking [4].

The details he cares about are granular. Naturalness comes from inflection and cadence rather than from voice timbre, and monotone or misplaced emphasis is what makes a system feel robotic [4]. Whether a user can interrupt, known technically as barge-in, is a genuine design trade-off: allowing interruption is friendlier but makes the system more likely to go off the rails, especially when background noise from a bar gets mistaken for speech [4]. Restaurants also get a portfolio of over 100 digital voices, accents, background music and fully customisable answers for things like corkage policy and parking directions [3][6].

On focus as the winning strategy

He is candid that Slang started backwards relative to standard advice: "we started with product and kind of backed into End Market" [6][11]. The first prototype answered his aunt's hair salon phone and worked, but VCs at the time disliked SMB and beauty, so the team ran pilots in healthcare, retail and e-commerce before landing on restaurants roughly three years in [6]. COVID sharpened the choice, since restaurants were understaffed and desperate for anything that kept operations running, call volumes were high, and the ROI was correspondingly high [6].

From that he drew a general lesson: "focus is everything" [6]. Two arguments support it. Natural language models perform better when they are vertically focused and use-case specific, and go-to-market becomes very difficult when you are selling into meaningfully different segments [6]. "we actually thought we move faster if we narrowed our ideal customer profile," he says, which reads as counterintuitive because you are cutting market away [6]. In practice that means full-service restaurants that take reservations, not drive-thru or QSR, with SMB and mid-market defined by location count up to about 100 and enterprise above that [6]. He still describes this discipline as a temptation resisted, since the same product could be pointed at many parts of the market [1].

On what actually differentiates in a crowded category

He acknowledges the field has filled up, with roughly 20 companies doing various things and competitors clustered around drive-thru or ordering rather than full-service hospitality [1][5]. His claimed edge is first-mover position and scale: largest by capital raised, clients, and calls answered [1], with a data set of over 30 million industry-specific calls he considers impossible to replicate [5]. That data is what lets the system automate 60 to 70% of inbound calls with no staff involvement, because the common call reasons are already known and pre-trained [6].

The team is the other half of the answer. Every team at the company has at least one former operator, and all customer-facing staff have worked in restaurants, married to AI expertise drawn from places like Amazon Alexa and Spotify [1][5]. He notes the compounding effect: good people want to work with other good people [5].

On what AI does to front-of-house jobs

His framing is task removal rather than headcount removal: the phone is a task hosts no longer need to be doing, freeing them for the people physically in front of them [1]. He describes the familiar failure mode of walking up to a host who says "excuse me one second" and answers the phone, or being asked to hold for five minutes [3]. Given that the first few minutes after a guest walks in shape the entire experience, a host stuck on the phone is a bad start [5]. He reports that frontline staff often become the biggest champions, because guests compliment the AI and because half their call load disappears [5].

The most common objection he hears from operators is not cost but fear that guests will be annoyed, often expressed as "we have an older demographic" or a less tech-savvy one; he says those clients are consistently pleasantly surprised once they try it [5][6]. Consumers arrive with baggage from bad automated phone systems, but he sees the tune changing as people encounter automation that actually works [6]. When guests are given a genuine choice, the option to say "host" and be transferred instantly, the vast majority still give the system a shot [3].

On the product roadmap and what operators actually want

The expansion runs in two directions: more channels and more of the guest journey. Text arrived first, letting a SevenRooms client book a reservation or send a private event lead straight to Tripleseat by text, with the same personality across every channel [1]. Beyond that he points to email, web conversations, inbound and outbound, plus ordering, CRM, catering and loyalty [1][5]. CRM and loyalty integrations are what unlock personalisation, so the AI knows on the call whether a caller is a VIP or regular, their birthday or anniversary, dietary restrictions, table preferences and lifetime spend, and can flag any of it to the front of house team [1][5].

Underneath that is a view about how operators buy: "operators don't really care about the the underlying tech so much," and the question is which problems can be solved with AI in a way that is easy to use and fairly priced [5]. He argues the deeper wish is for every platform an operator uses to talk to every other one, and that being a newer technology is an advantage because integration can be built in from the ground up [5]. He places the company in voice AI today but in the broader AI agent category going forward [6]. Pricing has been its own learning curve, run as an exploratory exercise that has landed at a $399 per month model, with the coming complexity of multi-product bundling and unbundling still ahead; he has argued pricing and sales deserve to be taught formally [8].

On knowing you are on the right track

His two barometers are customer behaviour and guest data. He watches whether restaurant partners stay, grow, and install the product when they open new stores, ideally immediately or even weeks before opening [5]. Separately he tracks guest satisfaction over time, on the logic that if it is not improving as the product changes, the wrong things are being changed [5]. He cites a guest satisfaction score of about 95%, meaning that share of guests who go through the product say it was helpful [6]. Marketing signals shifted after ChatGPT too: search volume for AI-related restaurant keywords rose from close to nothing, and messaging moved from hiding the AI to foregrounding it [6].

On raising money and hiring for a startup

On fundraising, across four rounds and roughly $20 million, his main advice is ruthless qualification: "you want to make sure you're spending zero time with those investors that are just going to waste your time" [2]. Qualify on portfolio commonality, published theses, and above all stage, then ask directly how many term sheets or deals they have written in the past two quarters and interrogate what those deals looked like [2]. He also recommends talking to founders one stage ahead of you, both to find who is actively investing and to get honest feedback on what those investors are like to work with [2]. His warning about the venture path is blunt: it is not for everyone, because the expectation is a specific rate of growth that you have to genuinely want and believe you can deliver [2].

On hiring, three things matter to him that a resume will not show. First, avoid candidates preoccupied with title, since "titles don't mean anything you are there to do the work and grow the company quickly," and "you become more senior and more valuable by driving results not by just like arbitrarily being given a title" [2]. Second, decide deliberately where your company sits on the collaborative-versus-independent spectrum and hire to it, because hires on the wrong side do not fit [2]. Third, back-channel references, especially for managers, and speak to people who worked for them rather than only their former bosses; he says these have taught him a great deal, both good and bad [2]. For anyone considering joining an early-stage company, he offers the test he and his co-founder applied to leaving Spotify: "assume that a year from now or two years from now the company fails will you still feel like you won," having learned enough to come out more employable [2].

Takeaways

• Roughly 50% of restaurant phone calls go unanswered, and about 60% of calls to reservation-taking restaurants are reservation-related, with the rest being common questions like hours, parking and menu items [2][3].
• Phone callers are worth more than online bookers: average party size over the phone is four to five covers versus two to three online [4].
• Guests should always be told they are speaking to an AI, both for ethical reasons and because research shows they use the system more successfully when they know [3][5].
• Naturalness in voice AI comes from inflection and cadence, not from how human the voice timbre sounds [4].
• Narrowing the ideal customer profile to full-service reservation restaurants sped the company up rather than limiting it, because vertically focused language models perform better and go-to-market gets simpler [6].
• The main operator objection is fear that guests will be annoyed, not price, and frontline staff frequently become the strongest internal champions once call load drops [5][6].
• When fundraising, disqualify investors early on stage and portfolio fit, and ask directly how many deals they have done in the last two quarters [2].
• Back-channel references on managerial hires by contacting people who reported to them, not just their former bosses [2].