Anis Bennaceur

The through-line

Two convictions run through everything Anis Bennaceur says, and they mirror each other. The first is about product: sales software should not replace the seller, it should strip out the admin around the seller. "we're not replacing the sellers per se. what we're doing is replacing the admin work that not only sellers but teams are doing right that has always been our vision since day one" [5]. That began as automatically filling Salesforce after a call [4] and has grown into AI agents that generate demo decks, win-loss analyses, pipeline reports and competitive intelligence end to end [1][7]. The second conviction is about people: a startup is won or lost on whether it hires level-five talent with founder instincts, and hiring should be run with the same discipline as product. "you have to make sure that you treat hiring just like you're treating product" [5].

Underneath both sits a strong preference for sequencing. Get product-market fit before touching growth [6]. Sell to your smallest customers first and climb, never jump [1]. Start agents with a human in the loop and remove the human only as trust accumulates [7]. The pattern is consistent from his account of the earliest days, when a huge inbound logo was blocked by InfoSec in the first three weeks of selling, which he now reads as luck: "we would have gotten destroyed if he we had work" [1].

On what the product actually does, and for whom

Attention captures sales interactions across calls, emails, messages, Slack and more than 200 integrations, then automates the work built on top of that data [7]. At the rep level, a conversation is recorded, transcribed, processed and pushed into every relevant CRM field, whether the framework is MEDDIC, SPICED or something custom [4]. At the manager level, it fixes the forecasting problem where CRM information is "incomplete uh if not empty" [4]. In aggregate, it can segment an entire call collection by stage, industry or segment and answer questions like why deals are being lost or what top reps do differently, producing a report with supporting quotes in minutes rather than the month an analyst would need [4][7]. The current roadmap is over a hundred specialised agents, released in stages, each doing one job end to end: competitive intel analyst, pipeline report analyst, and so on [7].

He frames the value in time returned: the average customer saves roughly seven years per company across the automations, from CRM autofill and follow-up emails to pre-call prep, post-call coaching and leadership pipeline risk analysis [5]. The wedge was deliberately narrow and painful. Everyone hates filling in the CRM, so that was a must-have for any sales leader, and the rest of the coaching and analysis capability was built while selling it [4]. He is explicit that the customer conversation belongs to the whole company: "the voice of your customer is this gold mine of information that should drive the work of every single function at a company" [5], and "The best companies in the world are customer obsessed" [5]. Product and marketing teams have become secondary users building agents off sales conversations, which he argues gives the sales team more internal power rather than less [1].

The customer profile has moved firmly upmarket: teams of at least ten reps, better at twenty to two hundred, with real conversation volume and recorded calls [6][7]. Companies with very low sales volume or no recordings are a poor fit [7]. The most inventive use he has seen came from an unexpected direction, consumer call centres selling things like dental implants, whose volume made them more sophisticated than most B2B buyers. One built an agent that analyses calls one and two specifically to work out what was said that produced a no-show on call three, something he had never seen a B2B customer attempt [1]. Another asked to "manage the managers coach the coaches" by scoring one-on-ones against their own scorecard [1].

On getting to product-market fit before you touch growth

The single lesson he carried from his first startup is stated flatly: "get to product market fit before you hit the accelerator on growth" [6]. The corollary is ruthless about engineering allocation. Early on you sell to teams under ten reps precisely because they will not demand admin settings, and "admin settings do not contribute to your product market fit. Billing does not contribute to your product market fit" [6]. Only as PMF tightens do you go more upmarket and build the systems larger organisations need [6].

His method for finding it is design partners and conversation volume. In the first week he spoke to 35 people, then held himself to at least three meetings a day, and the goal was "have five design partners that you build for", people who will say build this just for me [3]. They need not share an industry if the product is horizontal; Attention's first three were in legal or HR tech, sales tech and security [3]. The signal that PMF has arrived is a reversal of who is chasing whom: "you go from basically people ignoring you to people asking you or telling you what you could do better and telling you that they absolutely love your product" [3]. A blunter test: give ten beta users the product free and see whether seven pay within two months. Attention got two or three, so they dug into why and doubled down on those winning features [6]. He also cites the Superhuman question about how disappointed customers would be to lose the product [6], but rates active referral above NPS because people lie on surveys, making the K factor the better overlay on PMF [3].

He treats PMF as unstable and layered. You can lose it overnight or have it diluted, so you must keep innovating and keep watching the market half of the phrase: who else is in it and how much room they have [6][3]. Beyond product-market fit sits go-to-market fit, where pricing works for customers and for your own gross margins, and he flags anything under 80% gross margin as a problem [3]. He repeats a line he was given, that you do not really have product-market fit until $20 million ARR, while insisting the real evidence is usage, referral and what customers say [3].

On choosing what to build

The specific moment that unlocked Attention was a user saying their company would pay a lot of money to have the MEDDIC fields filled in Salesforce [3]. What made it a business was the structure underneath: the CRO reads those fields daily to forecast to the board and looks terrible if the call is wrong, while "the sales rep does not have anything to gain by filling Salesforce" [3]. Recurring pain, high cost of failure, misaligned incentives. That generalises into his framework for evaluating any idea: how much time does the task take, how often does it recur, and what does it cost the business. Daily and expensive makes you a must-have, occasional and cheap makes you a nice-to-have [3].

Asked how he would start today, he would pick a domain he knows and cares about, then look for niches with weak incumbents, such as industries where the leaders are private equity backed and therefore not innovating, or fragmented spaces with very low NPS, and stitch together an end-to-end solution [3]. He admires unglamorous verticals, citing a founder building an ERP for food distributors [3]. Vibe coding tools mean an MVP can be tested quickly in a way that was impossible three and a half years ago, but the constant remains: someone builds, someone talks to users, ideally both do both [3]. He also points to services as a route in, describing a contractor who did AI SEO for Attention, repeated the workflow for other logos, productised it and raised money on the result [3].

His own choice between growth and sales as a category was reasoned rather than sentimental. Growth is all alpha and arbitrage with diminishing returns as more people adopt the tactic, which makes for unstable product foundations. Sales is less sexy but structurally stable, and moving a client from 15% to 25% conversion while shortening cycles is real top and bottom line impact [3].

On founder-led sales and the first two AEs

Founders should personally sell the first $200,000 to $300,000 of ARR, and he sees some go to several million before hiring, which he considers acceptable in spirit even if extreme [1]. Whether that is four deals at $50k or ten at $20k matters less than reaching the point where you have enough requests and feedback to keep iterating and can articulate exactly what early customers love, because that is what you will teach your first AEs [1]. Even in the mid to late seven figures he and his CTO co-founder still talk to customers constantly: "it's extremely important that keeps you rooted into the decisions that you're making for your business" [1]. In practice he sold alone from February to June of the first selling year, ran 14 to 16 sales calls a day, then added two reps and had to start dogfooding the product as a manager as well as a user [6][2].

The most common founder mistake he names is hiring salespeople before doing growth and marketing, then expecting those reps to generate their own qualified pipeline and gather product feedback rather than close [1]. His own hardest transition was generating enough pipeline for two reps where there had been none [1]. The second rule is that the first sales hire should not be a head of sales but two founding AEs. Two, because they learn from each other, compete, have no excuse if one is closing and the other is not, and the better one earns the promotion. That is exactly what happened: one of the two founding AEs became head of sales [1][2]. Pushed on the argument for three rather than two, he was open and noted Attention was running the same experiment with three founding technical account managers [1].

The hiring process itself taught him what not to do. Posting the job on LinkedIn produced many terrible candidates [2]. His first two offers collapsed because he introduced the two hires to each other before they signed and they talked each other out of the risk, which he now reads as a fortunate escape [2]. The second time he did not disclose the other AE until both offers were signed [2]. What worked was an investor introduction to a specialist early go-to-market recruiter who sent exactly one candidate instead of thirty, and that candidate, Jacob, became the hire [2]. Jacob arrived at the first interview knowing nothing about the competitors or the space, apologised because he had been working until 11:30 every night, and Anis read that as a signal of work ethic and gave him a second round [2]. What sold him was scrappiness and storytelling: "I felt like I would buy anything from him" [2].

On the triangle of talent

He uses a five-level framework he credits to Shan Kuru [2]. Level one is useless, people who get it wrong even when told exactly what to do. Level two is the task monkey, who executes when told what, how and when. Level three is the problem solver, who is told what and works out how. Level four is the systems thinker, who is given the problem and builds the system, process and team around it. Level five is the superstar, who identifies the right problem in the first place and gets it solved [2]. Job title is irrelevant to placement: "Every employee is a problem solver", and a young person with steep slope can be a five while someone with heavy experience sits low [2]. His advice is to count the fives on your team, remove level ones immediately, and understand that level twos drag down the expectations of everyone around them [2].

He is pessimistic about promotion between levels. With enormous training you might move one or two people, "but in in my experience, it's just kind of something that people come with", formed by growing up on a quest for excellence [2]. That is why the filter matters more than the development plan, and why he refuses to trade quality for cost: "do not compromise on the quality of the employees that you're hiring because the cost of a mish hire is going to be way bigger than uh whatever pennies you're saving" [6]. He pays accordingly and ignores investor pressure to underpay, though he reads a seed-stage rep asking for $300k OTE as evidence they do not understand the stage, while $240k to $250k plus equity is a deal he will do for a genuinely top rep [6].

On founder DNA and the quest for excellence

Before hiring anyone, map the competencies you and your co-founder already cover, because the point of the first hires is to close the gaps standing between you and early PMF [2]. In Attention's case he was strong at growth and decent at sales, Matias was strong at front end and product design and weaker at backend, and the real-time coaching product they were building demanded low-latency infrastructure. Within three or four months they brought in Johan as backend engineer, now VP of engineering [2]. His summary of what a pre-PMF company actually needs is stark: "all you need early on to find early PMF is just talk to users and write code" [2].

The profile he filters for is entrepreneurial history rather than entrepreneurial ambition. "Ideally, your first employees should have traits of founders" [2], people who have tried to found something or intend to, because the unknowns are overwhelming and thriving in that obscurity is rare. He probes what candidates built growing up, how they spotted ways to make money, whether they built their own solutions when blocked, whether they are scrappy, and he applies this to engineers, sales and growth alike [2]. Johan qualified on all counts: a former founder who built something robust but could not sell it, who had written himself a mobile app that pinged him whenever an article hit the top of Hacker News, and who had read every Paul Graham article [2].

He wants more than scrappiness, though. He wants people steeped in the startup ecosystem, because knowing what the best companies did before you is what puts a person on the quest for excellence in the first place [2]. He applies the same test to co-founders: when he and Matias compared the top ten lessons from their first startups, seven overlapped, both had read all of Paul Graham and absorbed the YC ethos without going through YC [2]. Signals of world-class ambition in any domain count. One rep was once the highest-paid professional frisbee player in the world; another reads and listens constantly, thinks two positions ahead of his current role, and is, in his words, a better seller than Anis himself [6]. He tells the story of an engineer who, asked what he would do with his raise, said he would buy a bigger home office so he could keep working harder [6].

At the current stage the machine is largely self-feeding. The best people at the company conduct the interviews, which raises conversion because candidates see the calibre of the business; strong hires refer their friends and get paid referral commissions; attrition is very low, so the loop compounds [5]. That is where the parallel with product becomes explicit for him: great talent, like a great product, gets referred [5].

On climbing upmarket, deal by deal

Attention started with its smallest possible customers. Two years into selling, the average of the top 10% of clients was around $6,000 to $8,000 a year; that number later passed $100,000 [1], and by a later account $200,000 per client [5]. The method was never to make massive jumps, always to increase customer size incrementally and to learn from each cohort before attempting the next [1]. He argues you want people using your product early regardless of size or even payment, because they push, break and reinvent it in ways that prepare you for what comes next [1].

The economic argument for enterprise is a velocity ratio. Dividing deal size by days to close, enterprise selling ran at roughly $4,000 to $6,000 per day versus under $1,000 for SMB, which transforms CAC payback and means salespeople should be pointed at the largest available deals [5]. Enterprise customers also churn more slowly [5]. Referral chains do the rest: it took about a year to close the first six-figure logo, that logo introduced the next, and when a head of revops moved companies they took Attention with them, because enterprise people move to other enterprises rather than down to startups [5]. Deal cycles average two to four weeks, with six-figure deals occasionally closing in five days and low five-figure deals dragging on for months [4].

He rejects most of the standard complaints about enterprise as myth. Enterprise buyers accept realistic timelines, do not demand things tomorrow, and generally ask for capabilities that will benefit other enterprise firms, so building for them serves the roadmap. The genuine time sink was pre-PMF startups trying to distort the product into something only they needed [5]. What is real is political risk: enterprise buyers optimise for minimising risk and have heard the stories about people fired for implementing the wrong tool, so they will only bring you in if they have tested you or heard it from a trusted peer [5]. That is why a paid referral scheme with well-connected outsiders does not work; nobody puts their reputation behind a product they have not used [5]. Procurement negotiation is an art to be traded rather than resisted: give the discount, take the logo rights and a case study inside 90 days [5]. One of his reps asks for five customer introductions in exchange for every discount, and gets them [5].

Attention deliberately has no self-serve and no free trials. Deployment is heavy, needing four deployed engineers to understand how a business operates before setup, and a named account manager per client [5][7]. Free access was tried and abandoned: people who will not spend anything on a paid pilot do not take it seriously, and if a buyer cannot get $200 approved internally, they will not get $20,000 or $100,000 approved either [6]. He concedes self-serve enterprise land-and-expand can work, pointing to how Cursor spread through Fortune 500 users, and expects AI-led onboarding to eventually replace much of the deployed engineer motion, but says Attention is not there yet [5]. On PLG generally, his view is that it depends entirely on the buyer: selling to engineers, product managers or growth hackers makes PLG obvious, while selling to sales teams that need system connections, compliance approvals and IT review makes it a poor fit, and doing both motions simultaneously is very difficult [6].

On growth stacking and outbound in an AI world

His growth education began at Tinder, doing things that did not scale: campus tours, sponsored club events, creating accounts on people's phones in real time, and the pitch that made it work, asking someone whether they wanted to know who around them liked them [4][3]. The structural lesson he took was that a marketplace is won by acquiring the hardest-to-get users, which at Tinder meant investing everything in getting attractive women onto the app [4]. He carried that directly into Mixer, where the hardest-to-get users were world-class creatives, band members from The Strokes and Maroon 5, Oscar-winning filmmakers, top graphic designers and animators, and the value he offered them was access across disciplines they could never otherwise reach [4]. Mixer was bootstrapped and cash-flow positive, which forced him to automate instead of hiring and made him fluent in workflow tooling and the problem of unstructured data, the exact problem GPT-3 solved when he saw it on 19 July 2020 [3].

At Attention he replaced growth hacking with what he calls "growth stacking instead of growth hacking": run as many experiments as possible, keep the winners, automate them, then layer new channels on top, which compounds into exponential growth and sustained 30 to 60% month-over-month rates [6]. One early stack was a script that detected whether a prospect had spoken on a podcast, pulled the transcript, matched Attention's value prop against what they said, and sent a genuinely personalised email, generating 20 to 30% response rates and 10 to 15% positive ones, fully automated [6]. He also feeds sales back into marketing: after a sophisticated customer taught the team to ask why us, why now, why do anything, Attention captured what worked in conversations and reinjected it into top of funnel messaging [6].

He is candid that this advantage decays. People will respond less as they lose the ability to tell human from AI, which makes intent data, who visited your site, your G2 competitor page or a specific blog post, the thing that turns cold outreach warm because the value prop lands on a live pain [6]. His broader prediction: "in a couple years we'll have five people doing the work of 50" [6].

On agents, humans in the loop, and whose job goes

The path to autonomy is incremental. "you start with humans in the loop and then you remove the humans over time just so that um you end up automating entire segments of of functions" [7]. Customers currently want control and are not ready for fully automated action; one wants competitive intelligence surfaced weekly so a human can inspect it before it enters the enablement knowledge base, and he expects that same customer to eventually ask for the knowledge base to update itself [7]. He sees the world moving toward reinforcement learning, with human feedback continually steering the model, and locates the company's moat in knowing what an accurate representation of the world looks like for each client [7].

On jobs, he does not think functions disappear, he thinks they get multiplied: enablement, revenue operations and layers of managers whose work is to inspect, report and alert can become 100x more productive, with two people doing what twenty or eventually two hundred did [7]. Expertise still has to originate the ideation, creativity and initiative that directs an army of agents [7]. The blunt version: "AI won't get rid of your a players it might get rid it will most likely get rid of your C players and as a company maybe we'll see what B players are" [7]. He expects buying to change too, not through AI agents transacting autonomously but through procurement doing far deeper pre-diligence first, as he did himself using deep research to map a vendor's competitors and pricing before still taking the call [7].

On the competitive map, he regards Salesforce and other incumbents as competitors by virtue of adding AI, and expects newer players to copy and undercut on price without much concern [7]. Clay he views as complementary rather than competitive, since Clay works top of funnel on public data and outreach while Attention works bottom of funnel on conversations [7]. Overlap may appear with dialer companies as they add AI, and would only get interesting if they ship agents [7]. Pricing reflects the architecture: a seat fee for users recording calls plus consumption pricing for agents, or consumption only if a customer brings their own call recorder [7]. Longer term he considers CRMs on the way to obsolescence and says so directly: "I'm bullish on Attention, which will actually take over the CRM" [5], while continuing to sit on top of Salesforce and HubSpot today because that is what the largest clients run [5].

On dogfooding, and on what he got wrong

He uses the product on his own company and treats the results as evidence. A pipeline analysis flagged that a rep had not built a compelling business case on a large six-figure deal; the same diagnosis came independently from an investor's VP of sales; the rep built the case, got introduced to the CFO and closed it [1]. He credits the software with teaching every rep, himself included, how to create urgency and improve the ACV over deal velocity ratio [4]. Preparing an analyst briefing for Gartner, he had Attention build a ten-slide deck from the voice of the customer in about an hour, work he estimates would have taken him three or four days [5]. On honesty, he says the loss analysis will tell a leader plainly where the deal went wrong and where to improve, though it will not tell anyone to fire themselves [1].

The failures he names are mostly about pace. Growing too fast early meant onboarding customers faster than an understaffed engineering team could serve, and a brutal quarter of requests from two outsized customers ended in one churn [6]. Giving product away free taught him that unpaid users do not take deployment seriously [6]. And the enormous inbound logo in week three that InfoSec blocked was, in retrospect, protection from a customer he was not built to serve [1]. His personal operating mode is unapologetic. Investment banking taught him rigour, attention to detail and the ability to work insane hours [4]. Asked how he keeps his sanity while chasing a business he now describes in trillion-dollar terms, he answers "I'm not sane. I think about this business at night and day" [7], then describes the actual arrangement: three nights a week at the office until midnight, two nights home for dinner with his fiancée, Saturdays entirely off with her, Sundays back at the office [7].

Takeaways

• Automate the admin around the seller, not the seller: Attention's customers save roughly seven years per company through CRM autofill, follow-up emails, pre-call prep, post-call coaching and pipeline risk analysis [5].
• Never spend engineering time on anything that does not contribute to product-market fit, including admin settings and billing, and only go upmarket once PMF tightens [6].
• Test willingness to pay early: give ten beta users the product free and see whether seven convert within two months; Attention got two or three and doubled down on the features they paid for [6].
• Founders should sell the first $200,000 to $300,000 of ARR themselves, do growth and marketing before hiring reps, and keep talking to customers well past seven figures [1].
• Hire two founding AEs rather than a head of sales, so they compete, learn from each other and have no excuses; do not introduce them before both offers are signed [1][2].
• Use the triangle of talent to count level fives, remove level ones fast, and accept that level three and four rarely become level five no matter how much training you apply [2].
• Climb customer size incrementally rather than jumping: Attention went from $6,000 to $8,000 ACVs to over $100,000 and later $200,000 without skipping stages [1][5].
• Judge enterprise on dollars per selling day, roughly $4,000 to $6,000 versus under $1,000 for SMB, and trade procurement discounts for logo rights, case studies and customer introductions [5].
• Replace growth hacking with growth stacking: run many experiments, automate the winners, layer new channels, and expect 30 to 60% month-over-month compounding [6].
• Roll out agents with humans in the loop first and remove oversight only as customers learn to trust the output [7].