George Bonaci on Growth Experiments, Hiring and Finding New Channels

Ramp VP of Growth George Bonaci joins Harry Stebbings to discuss growth experiments, marketing attribution, hiring and AI. His views reflect his experience at Ramp and other companies and his commercial role in growth. The interview does not substantiate the video's 'fastest-growing' superlative or independently verify its growth and conversion claims.

Bonaci calls growth a kind of science: define a hypothesis, run experiments and measure what happens instead of assuming a playbook from another business can simply be copied. Every company differs, he says, so an unexpected failure or success is useful information. He contrasts this with launching tactics purely because they worked elsewhere.

A growth portfolio needs both high-risk bets that could produce step changes and more predictable improvements that help meet near-term goals. He recommends agreeing with finance and leadership on how to allocate resources across time horizons. The appropriate risk tolerance depends on the company's stage and whether it prioritizes growth or profitability.

He expects most experiments to fail and generally favors velocity, but warns that sloppy tests can be too confounded to teach anything. An extremely rigorous experiment can take too long, while many poorly designed tests can produce no reliable learning. The aim is to balance speed with a way to understand the result.

At an unnamed company, a key web page's conversion rate was falling. The team changed many things at once and, he says, tripled conversion over a couple of weeks, helping meet a quarterly goal. They later had to test and undo changes because they could not identify what caused the improvement. This is an uncontrolled example, not proof that all those changes worked.

For slower bets such as content, Bonaci suggests agreeing in advance on the time horizon and looking for leading indicators or smaller tests. He prioritizes experiments by expected impact and effort, but also by confidence and time to results. A low-confidence experiment can still be attractive if it can produce a useful signal quickly.

Before launching, he says teams should define what can be measured, how long it will take and whether a result can be statistically significant. If quantitative measurement is not feasible, they can gather qualitative feedback, provided they acknowledge the uncertainty and possibility of being wrong. A data specialist can help with the design.

Bonaci advocates scaling a successful channel toward saturation quickly. Stebbings challenges whether jumping from $10,000 to $200,000 in spending could destroy efficiency; Bonaci agrees the response curve matters and says to watch when incremental returns begin to decline. He believes startups often reach that point too slowly, but the appropriate spending pace depends on the channel.

He says acquisition costs can rise as an audience saturates, but new products, geographies and channels can offset that pressure, and cross-channel effects may be missed in simple attribution. He considers lifetime value a useful framework but acknowledges the difficulty of estimating it early. Teams should agree on what they are willing to spend with the uncertainty in mind.

He argues teams should not become so attached to a particular experiment that they cannot stop it. If a long-term bet is highly convincing, revisit the confidence and the agreed time horizon; otherwise, recognize that failure is normal when trying new things. This is a management principle rather than a rule to cut every slow channel.

He recommends a premortem before a test: list plausible reasons it could fail, such as an insufficient sample. A postmortem becomes especially useful when failure came from an unanticipated cause or yields a lesson that applies elsewhere. His rough estimate that teams can foresee over 90% of failure modes for high-confidence bets is experience-based, not a measured general rate.

The person responsible for an experiment should write the postmortem, ask whether it was designed and executed well, and identify lessons that transfer. He prefers sharing a document at least 24 hours before a live discussion with relevant cross-functional stakeholders. The format matters less than learning and preventing avoidable future failures.

In one A/B test, Bonaci says a red button won overall but performed poorly among enterprise visitors. Applying the overall result blindly to enterprise content and webinars could therefore hurt those campaigns. He uses the example to stress segment-level analysis; the interview supplies no sample sizes or detailed results.

Bonaci prefers growth to be independent enough to work across product and marketing on whatever most helps the business. He says Ramp's growth team reports to a co-founder, while other organizations use a chief growth officer or a roaming team. The structure, he acknowledges, depends on the business.

He borrows 'alpha' from investing to mean an advantage others have not yet captured. One source is trying an emerging channel before it becomes crowded; another is testing something others dismiss. He recalls direct mail initially sounding implausible but becoming a major channel after iterations, a company experience rather than a universal return claim.

To find ideas, he suggests combining reading, conversations with peers and observation of other industries or geographies. He cites WhatsApp as a channel more established in some international markets than in US business marketing. Historical methods can be adapted rather than copied, as with applying older marketing-mix modeling ideas to digital campaigns.

He liked direct mail because it was relatively underused and could reach large samples, allowing parallel tests and faster learning. He also sees business-to-business influencer marketing as a possible opportunity, though building it is labor-intensive. Neither channel is guaranteed to work for a particular company.

On attribution, he treats brand investment as a longer-term, higher-risk bet that may be hard to measure directly. It can build awareness among people already seeking a solution or make a wider audience aware of a problem. He suggests considering brand as direct-response channels mature, while recognizing that attribution remains uncertain.

He says concentration in one growth channel can be reasonable temporarily when the team finds a strong opportunity and scales it. The challenge is to build and diversify subsequent bets rather than remain dependent on one channel indefinitely. Growth teams also need to communicate their way of thinking and align their cross-functional work with company goals.

For an early-stage company around a Series A, Bonaci favors a relatively junior, capable generalist over an expensive senior specialist hired only for a résumé. He looks for logical problem-solving, numerical ability, rapid learning and internal motivation. Engineering, finance or consulting can be useful backgrounds, but his preference is a heuristic and does not rule out other paths.

He cautions that people accustomed to large-company playbooks may find it difficult to adapt to a small startup's hands-on work. For interviews, he suggests references and a realistic quantitative case or take-home assignment. Founders should know what good looks like before testing candidates and make the opportunity attractive as well as evaluate it.

His example assignment uses messy real-world data: candidates may have to notice duplicates, inconsistent dates or missing information before deciding which campaign or leads performed best. He looks at their questions and reasoning, not just the final answer. Other team members should meet the candidate to assess whether the relationship works both ways.

A generalist hire may discover that growth work is not enjoyable, he says, and even strong hiring managers will make mistakes. He recommends transparency about the job and how performance will be assessed. Different stages suit different working styles, but he does not believe people are permanently destined for one stage.

Bonaci considers management a skill requiring explicit investment. At Samsara he recalls a structured leadership program with books, peer discussions and an expectation that participants apply the lessons. He argues that leadership must provide time, resources and accountability, rather than simply tell employees to learn on their own.

Stebbings asks whether older business books can apply to modern technology and a changed workforce. Bonaci thinks tactics change more than many management principles and cites the theory of constraints: identify the bottleneck preventing a team from achieving more and work to remove it. He recommends selecting a principle deliberately rather than copying an entire playbook.

At Ramp he identifies experiment velocity as a bottleneck because there are more opportunities than time and resources. He says a leader should understand each team member's work well enough to do it poorly, while hiring people who can do it better. The aim is to avoid micromanagement while retaining enough understanding to help remove obstacles.

He expects new growth hires to spend the first 30 days learning the business, their role and how the company makes money. He favors a detailed 30-, 60- and 90-day plan, with early work that can be assessed against facts rather than feelings. By 90 days he hopes to see results from the person's ideas, while recognizing that some problems take longer.

He says a common reason a growth hire fails is a manager's mistake: the role was poorly scoped, the business changed or the wrong profile was hired. Rather than assembling a long wish list, he recommends identifying the one problem or skill most important to the business and assessing whether the candidate can address it.

On AI, Bonaci thinks automated budget allocation based on historical data may improve existing activities but may struggle to find genuinely new opportunities. He says AI has changed his own hiring assumptions by helping people write code and analyze data without as much technical background. That view reflects tools available at the interview, not a guarantee of future capabilities.

He sees AI as a collaborator for both creative and performance work: it can suggest ideas or help analyze data. He personally finds it especially useful for brainstorming and writing prompts because creativity is not his strongest skill. Outputs still require judgment, and his account is personal experience rather than a measured productivity estimate.

In crowded markets, Bonaci argues a better product is essential but must be paired with a distribution plan. He rejects relying on 'build it and they will come,' saying unusually strong word of mouth can work but leaves too much to chance. As Ramp's growth leader, he has a direct stake in this product-and-distribution perspective.

In quick fire, he warns founders against hiring solely for experience without first defining and testing what good looks like. He proposes learning from people who have built the relevant system. He names business-to-business influencers as an underappreciated channel and paid search as one that can saturate quickly; these are judgments, not benchmarked returns.

He regrets some early event sponsorships at small companies, where a modest booth disappeared among competitors. At greater scale, he says, an event strategy can combine speaking, advertising and follow-up outreach. He also sees potential in direct mail and display advertising, but emphasizes that display's indirect effects are difficult to measure.

Bonaci says he has become more appreciative of brand investment after seeing Gong invest in it, even when individual activities were difficult to measure. He notes that inbound growth followed but does not establish causation. He closes by considering less fashionable channels such as cold calling and field sales, while treating them as opportunities to test rather than guaranteed winners.