George Bonaci, VP of Growth @Ramp: How Ramp Became the Fastest Growing SaaS Company Ever |E1264

Summary

• George Bonaci’s core operating system: growth is science, and most marketers are bad at science — come in with a blank slate, form hypotheses, and run experiments at high velocity, because copy-pasting a playbook from a past company “generally doesn’t work.” Assume the majority of bets fail; “if they’re not failing, honestly they’re probably not doing their job well.”
• When a channel works, ideally take it toward saturation quickly — graph the response curve rather than blindly jumping from $10K to $200K, because “most things saturate probably more slowly than people expect” and most startups get to the asymptote too slowly. Macro saturation can take a really long time to hit CAC: new products, geographies, channels, and halo effects keep resetting the curve.
• Alpha in growth = doing what no one knows about or what everyone is convinced won’t work. Direct mail — dismissed as “junk mail to people’s homes… that absolutely won’t work” — became one of his biggest channels because nobody was doing it and it could reach 200,000 people tomorrow for parallel experiments. Today’s underappreciated pick: B2B influencer marketing, run like scaled outbound to 10,000 micro-influencers. Most polluted channel: paid search — “you’re paying a tax to Google.”
• Velocity beats rigor, but with a floor: “if you’re just doing a bunch of sloppy things… you’re not going to actually learn anything.” His cautionary tale — changing everything at once on a decaying homepage 3x’d conversion in weeks and saved the quarter, but “we never actually knew what did or didn’t work” and had to unwind changes after later A/B tests.
• On hiring: always skew junior and hire for potential — smart generalists who can think from first principles and do math (engineers, ex-finance, ex-consultants), while he has personally stayed away from profiles shaped by years at much larger companies. The most expensive irreversible founder mistake: “hiring for experience because they don’t know any better.” Test candidates with a messy real-world Salesforce dump, and know what good looks like before you send it.
• A good leader should know how to do everyone’s job “but poorly” — the poorly part is important, because if you do it better than your hires, you hired wrong and you’ll micromanage. Invest in management deliberately: Samsara shipped leaders a box of 15 business books, one per month, with structured discussion and required practice.
• AI lowers the technical bar but doesn’t automatically create alpha: he no longer thinks growth hires need to be nearly as technical as before, and thinks AI “helps uncreative people more” — but budget-allocating AI “almost by definition… just going to give you incremental gains on things you’re already doing.” Having AI interview 50 industry insiders and synthesize alpha is “really difficult… at least at the state of AI today, but maybe not in a year.”
• He changed his mind on brand in the last 12 months: Gong invested in completely unmeasurable brand work and it showed up in inbound — once at scale, “if you’re not investing in brand… you’re going to screw yourself over in the future.” And on “build it and they will come”: “absolutely not… that is antithetical to everything we stand for.”

Deep dive

1. Growth is science — and most marketers are bad at science

• Bonaci’s opening frame: the goal of growth is finding repeatable channels with “predictable outputs given some inputs,” and “the honest answer is no one really knows — every business is different.” Taking one playbook and copy-pasting it into a new company “generally doesn’t work,” yet that’s the tendency of most marketers: they think in terms of what do I know and how can I apply it, not in terms of experiments.
• The gap is partly a profile problem: the person who becomes a chemist (as he was) or engineer thinks differently from someone drawn to writing or PR — “scoping an experiment and thinking about what my hypothesis is and how I’m going to measure results is just a different part of your brain.”
• The method: blank slate, hypothesis, run a bunch of experiments — “you’ll be surprised about what works, you’ll be surprised at what doesn’t work, but if you run enough of those experiments you’ll find something.”

2. Run growth like a venture portfolio — and expect most bets to die

• Growth must be both incremental and needle-moving: allocate across time horizons, from high-risk big swings to high-confidence bets that deliver “the 2, 3, 4, 5% improvement this quarter” — and that allocation “is actually a conversation you should probably have with finance, with leadership.”
• On failure rates: “if you’re doing things right, you should assume the majority of your bets are going to fail — which is why velocity is probably more important than getting things perfect.” Long-term content bets get 12-18 months; be okay with no early results if that bucket is sized intentionally.
• Concentration doesn’t scare him: saturating a winner fast means “you will be, almost by definition, really concentrated there” — the real question is how quickly you can stack different bets and wins to diversify. “Concentration in and of itself is not a bad thing. It means that you’re doing a good job maximizing an area.”
• On org design: growth should be as independent as possible; at Ramp, the team reports to a co-founder — “one of the reasons why I love Ramp is the growth team reports to one of the co-founders.” The mandate: “growth team’s job is not to make anyone happy, it’s to make the business successful.”

3. Velocity beats rigor — his own cautionary tale

• The floor on sloppiness: “if you’re just doing a bunch of sloppy things, it doesn’t matter how many things you run — you’re not going to actually learn anything.”
• His confessed example (company unnamed): the #1 channel’s webpage conversion was trending down, and instead of controlled experiments they changed everything at once on gut feel. It worked — 3x’d webpage conversion in a couple of weeks and hit the quarter’s number — but “we never actually knew what did or didn’t work, and we had to go back and undo a lot of the changes” once properly A/B tested. Under the gun, optimize for velocity; once off it, go back and buy the rigor.
• Two prioritization dimensions most teams miss beyond impact and effort: confidence and time-to-results. “If you’re really confident something’s going to work, you should just do it”; if you’re unconfident but results come fast, do that too.

4. When something works, scale toward saturation

• Most startups’ mistake is timidity: they see a channel work and merely double spend. Instead, “take that channel to saturation as quickly as possible” — graph the response curve, watch for linear-to-decay, and scale until incrementality fades. Harry’s pushback — wouldn’t $10K → $200K overnight be wildly inefficient versus stepping up gradually? Bonaci’s answer: yes, but it depends; watch the curve, because “most things saturate probably more slowly than people expect… going from 10K to 200K might not be as insane as it sounds.”
• On whether CAC rises over time: theoretically yes — every incremental acquisition should cost more as share grows — but “the reality is that’s usually not true.” New products improve LTV, new geographies and channels open, and combined channels create halo effects you’re not capturing in CAC. “It takes a really long time to get to the point where the macro effect of saturation is hitting your CAC.”
• On early-stage LTV worship: it’s “false precision — you’re not going to know what your LTV is if you’ve been in business for a year.” Just agree on a spend threshold as a business and revise it as experiments teach you.

5. Premortems, postmortems, and the red-button lesson

• Premortems should be granular — write out the specific failure modes (“we won’t have a large enough sample size”) with probabilities. For high-confidence bets they’re 90%+ accurate; big swings always carry “some black swan… you cannot anticipate” (every growth team’s COVID excuse — “that was a tough one for us all”). A postmortem is most interesting when the failure was unanticipated; if you’d predicted it, “I probably wouldn’t even do a postmortem.”
• Mechanics: the DRI writes it in a Google Doc, sends it at least 24 hours ahead, then holds a live conversation rather than a comment thread. The key test: did we learn something generalizable, and are the right cross-functional stakeholders in the room?
• His best generalized learning: a red homepage button beat every A/B variant despite red’s “don’t hit the red button” connotation — but segmented data showed it severely underperformed for Enterprise — “a great example of Simpson’s Paradox.” Since most content, webinars, and direct mailers targeted Enterprise, the “red is best practice” takeaway was silently degrading everything else the company ran.
• Culture rule: “no one should be that attached to any experiment… they should acknowledge that the vast majority of what they work on is going to fail — and if they’re not failing, honestly they’re probably not doing their job well.”

6. Alpha is what no one knows or nobody believes

• His investing-borrowed frame: “what’s your unfair advantage?” Alpha comes from “doing things that no one else knows about” — the TikTok-early-B2B trade, where first movers had “huge returns” before saturation — “or doing things that everyone is convinced will not work.” Direct mail was mocked (“junk mail to people’s homes — that absolutely won’t work”) and “became one of our most successful, biggest channels after a few iterations.”
• What he actually saw in direct mail: no one was doing it, it was incredibly scalable, and huge sample sizes meant fast learning — “there are very few channels where you can go, hey, tomorrow let’s reach 200,000 people.”
• Three ways to find these: academically (adapt old playbooks — media mix modeling, today’s hot attribution topic, “is a really old concept from the Mad Men-style advertising days in the 1950s and 60s”), from peers (“probably not the best unfair advantage since someone else knows about it”), and — most interesting — from other niches, verticals, and geographies: WhatsApp is a massive marketing channel internationally, so “could we try WhatsApp instead of sending emails? It’ll probably fail — but if you’re doing enough of these experiments, you’ll find something no one else is doing.”
• Today’s pick: B2B influencer marketing, especially when moving upmarket toward Enterprise — treat it “almost as an outbound funnel: make a list of the 10,000 micro-influencers… and do some scaled outbound to them.” His caveat, twice: “it’s a lot of work.”

7. The channel scorecard: paid search taxed, events a trap, display and brand underpriced

• Most polluted channel: paid search — “everyone goes to paid search because they have to… it saturates quickly, you’re paying a tax to Google. Relatively uninspiring of a growth channel that will scale long term.” Biggest regret: early-stage event sponsorships — “you’re in a sea of other companies and no one’s paying attention to you”; the money is better spent on a guerrilla tactic or paid ads. Harry’s rule, which he endorsed: if you do events, go all in — billboard outside the hall, hotel key cards, speaker slot — “not just a small booth.”
• Missed opportunities: direct mail and gifting even earlier at Samsara, and display advertising — “so inexpensive right now” and mismeasured because nobody clicks, but serving the right impressions into an account “actually does have a halo effect.”
• His change of mind in the last 12 months: brand investment. Gong “were willing to invest in things that elevated the brand even if it was completely unmeasurable — but you saw that in the inbound numbers.” It doesn’t make sense for very early startups without that type of resource; past a certain scale, “if you’re not investing in brand as part of your long-term bucket of bets, you’re going to screw yourself over in the future.” The interesting version of brand isn’t awareness but true demand generation: “there are people out there that don’t realize they have a problem… this problem exists, there’s actually a better way.”
• Most impressive recent strategies are the unfashionable ones: cold calling — “a tremendous channel for a lot of companies because it is hard to do” — and door-to-door, borrowed from medical-device sales, now that return-to-office makes “a true field sales team that goes door to door and brings a cake” an unfair advantage.

8. Hire junior generalists who can do math — and test them on messy data

• For a series A at ~$1M revenue: “I would always skew more junior. Hiring for potential especially early on is far more important… it would be a mistake to hire someone more senior.” Best backgrounds: people who’ve demonstrated they can think logically and do math — engineers, ex-finance, “an ex-consultant that hated consulting and really wanted to get their hands dirty.” He has personally stayed away from profiles shaped by years at a company an order of magnitude larger — “going from playbook to first-principles thinking is generally difficult to do.”
• Process: first call sells (off a strong warm referral and back-channel — the real first interview stage), second is a take-home. Make it real and quantitative: “take a Salesforce dump and ask, what’s the best campaign here?” Real-world data is messy — duplicate leads, misaligned dates — and how they handle it, what questions they ask, and how fast they adapt are all signal before you even see output. “Understand what good looks like before you send the test out.”
• On speed of judgment: inklings come “in the first couple weeks” — structured onboarding and shipping in week one let you “assess based on facts versus vibes.” Harry’s confession that he keeps doubtful hires three months so the firing looks fair got no pushback: “I don’t actually think that’s wrong.”
• When hires fail, “it was on the manager… a mis-hire” — the trap is a laundry list of skills. The harder, better path: “get crystal clear on what is the one thing we need — one skill, one trait, one problem — then hire someone you have high confidence can fill that gap.” Quickfire version of the founder mistake: “hiring for experience because they don’t know any better.”

9. Leaders do every job — but poorly — and build learning into the system

• His management maxim, prompted by ex-colleagues praising his willingness to roll up his sleeves: “a good leader needs to know how to do everyone on their team’s job but poorly — and I think the poorly part is important. If they know how to do the job better than the people they’ve hired, then they didn’t hire the right people, and it’s going to lead to micromanagement.”
• Most companies say they invest in learning but struggle to do it: “actually doing it is really difficult… it has to be top-down.” His template is Samsara’s leadership principles: every leader got a box of 15 business books shipped home, read one per month, discussed it with peers, then had to demonstrate putting it into practice. Against Harry’s charge that 20-year-old leadership books are obsolete: “I don’t actually think business has actually changed that much” — his example is The Goal and the theory of constraints, 40 years old and still “one of the core jobs of being a manager.”
• His own bottleneck at Ramp: “the velocity of experimentation — there are so many opportunities, we don’t have enough time, we don’t have enough resources.” External vendors and legal review slow everything; at Ramp, he says they do a good job of prioritizing speed: “let’s optimize for speed… launch it and test it, and figure out the details if we’re going to scale it up.”
• Onboarding follows the same discipline: first 30 days scheduled “in excruciating detail” (his Samsara first two weeks were planned “to the minute”) to learn the business and the job; beyond 30 days, new ideas; by 90 days “the fruits of those ideas.” Early wins should exist — and if someone can’t put points on the board, find out fast whether they were set up for success.

10. AI lowers the technical bar but doesn’t automatically create alpha

• On the AI-allocates-your-budget future: “it’s hard to find alpha if you’re just having an AI go find the opportunities for you.” Harry pressed — wouldn’t connected data history make it perfectly personalized? Concession, then the counter: personalized, yes, “but almost by definition that’s just going to give you incremental gains on things you’re already doing.” True advantage would be AI interviewing 50 industry people and synthesizing what applies to your ICP — “that sounds really difficult to do at least at the state of AI today. But maybe not in a year, who knows.”
• AI upended one of his strongest convictions: “a few years ago I was such a strong advocate that everyone on the growth team needs to be technical — learn SQL, learn Python. AI upended that.” And between scientific and creative growth people, “maybe it helps uncreative people more” — he now asks ChatGPT for jokes, analogies, and visuals he’d previously have begged off product marketing.
• On competing in crowded markets like Ramp’s: “you just have to have a better product… that probably solves 80-90% of the battle,” plus an unfair advantage in distribution. But build-it-and-they-will-come? “Absolutely not… that is antithetical to everything we stand for. It leaves your fate so much to chance.”