Mike Duboe on Growth Teams, Experiments and Paid Marketing

Mike Duboe studied engineering and began at Bain, where he learned structured thinking but wanted work closer to his interests. A food startup introduced him to the practical work of a small team, including a group-payments feature that improved conversion.

He joined group-payments company Tilt after cold-emailing its founder, then moved to Stitch Fix through an introduction from a former colleague. Stitch Fix had grown through organic demand and retention under profitability constraints before building a substantial in-house acquisition function.

One lesson from Stitch Fix was to choose the right objective. Duboe led acquisition while a separate team owned retention; maximizing new users and minimizing acquisition cost could have sent low-quality customers downstream and hurt the other team's results.

The teams benefited from a more holistic, later-funnel measure. Duboe argues that growth can be harmful if a company scales the wrong metric without understanding how acquisition, engagement and retention interact.

A second lesson was incrementality. Duboe says Stitch Fix drew on data-science experience to use holdout tests rather than assume a precise attribution model could correctly divide credit for every conversion among channels.

He defines a growth leader's job broadly: accelerate the company's learning through experiments and build systems that give it more control over its north star. As an example, paid traffic can test several messages and landing pages faster than waiting for organic traffic alone.

Duboe favors cross-functional growth pods because product, engineering, marketing, analytics and design can test a hypothesis without depending on a long sequence of handoffs. He notes that successful growth practices can eventually become part of ordinary product and marketing work rather than requiring a permanent separate team.

An acquisition team organized around marketing channels can suit a straightforward e-commerce funnel, while a cross-functional pod may suit a product with more complex growth mechanics. Reporting lines can change; when growth work requires significant product and engineering changes, he generally prefers placing it under product.

He advises against hiring a dedicated growth leader before product-market fit. A generalist with growth experience can still help find it, but scaling the top of the funnel before users retain and the team knows who they are can damage a business.

His preferred early growth hire is often an analyst who understands the business and can help build the analytics foundation. He distinguishes that role from a data scientist and wants product managers eventually to treat instrumentation as part of building the product.

Analytics debt appears when basic questions, such as how a recent cohort retained after 30 days, are expensive or slow to answer. Dashboards alone cannot fix missing data: the product must first record the relevant user events.

Duboe praises Faire's early analytics for focusing regular reporting on a small number of business-critical measures rather than tracking everything. He says he knew the company early but was not an investor, and notes that marketplace transactions can be easier to measure than social behavior.

To choose a north star, he starts by writing down how the product actually grows. In a content loop such as his Pinterest example, existing content attracts users, who create more content that can bring in further users; the right measure can change as the business matures.

Stebbings sketches a loop for 20VC: strong guests bring listeners, which helps attract founders and fund investors, feeding the business. Duboe suggests that the relevant metric might change from guest reach to later retention or conversion measures as the show develops.

Before hiring for growth, define what success looks like: for example, a working growth model or analytics system by a given month. Duboe says the vague instruction to make a good product grow leaves both the founder and the hire without a shared standard.

He recommends talking to growth leaders at companies with comparable mechanics to calibrate what strong work looks like. A job description should explain the opportunity and interesting problems while specifying the experience, behavior and outcomes the particular role needs.

Because experienced growth candidates may be scarce, Duboe favors targeted outreach through trusted networks and relevant communities. He also recommends checking references early, while respecting the candidate's confidentiality, rather than treating references as a late formality.

One interview question asks a candidate to pick a favorite product and explain how it grows. He looks for a systems-level answer that combines quantitative thinking with an understanding of user psychology, rather than a single isolated anecdote.

Another question asks about a failure and what the candidate learned. Experiments will fail, he says, so a growth leader needs the willingness to acknowledge a result, learn and iterate rather than conceal it.

He also asks candidates to show or sketch a daily dashboard. A useful answer has a high signal-to-noise ratio and connects a few measures to the business; an unwieldy collection of numbers can be a warning sign.

For a deeper exercise, he asks for a prioritized experiment roadmap for a feature. The goal is to see whether a candidate can weigh impact and resources against business context instead of proposing an indiscriminate list of tests.

His example is a three-month plan to increase daily active users of Messenger. Another exercise asks a candidate to model a comparable business's growth; he gives people a few days and suggests limiting the work to a few hours rather than testing only their ability to improvise.

Duboe recalls Keith Rabois's distinction between people who execute tasks and the rarer leaders who can take ownership of an entire stream of work. His advice is to give those leaders responsibility and appropriate equity because they increase how much a company can do at once.

For a hypothetical Series B hire in a January 2023 conversation, Duboe estimates roughly $150,000 to $250,000 in annual salary, with equity often around 0.5% to 1%; he discusses possible variation above and below those figures. He stresses that compensation is dynamic and depends on seniority, stage and the balance of cash and equity.

A growth loop is a system in which an output is reinvested into an input: a Pinterest user creates content that attracts another user, who may create more content. Duboe says this can compound, while a conventional funnel describes more linear steps such as awareness, acquisition, activation and retention.

Funnels are not always wrong: he says they can be practical for e-commerce. The risk is organizing separate teams around individual steps and missing interactions between them, or optimizing one local number at the expense of the whole system.

Operationalizing loops means connecting distribution to product strategy and selecting shared measures across functions. Cross-functional teams can then work on the relationship between acquisition and retention instead of pursuing isolated tactics.

Duboe wants ideas to come from anywhere in the company and a culture where a failed experiment is acceptable if the team learns from it. Process should help teams prioritize and retain learning, not simply add approval layers.

His experiment one-pager records the objective, hypothesis, design and resources, duration, quantitative success criteria and next steps for either result. Writing these before the test reduces the temptation to redefine success after seeing the data.

Ideas enter a shared, prioritized list. Duboe describes using an approximate impact, confidence and effort score to compare proposals and avoid the instinct to test everything equally.

Weekly growth-team reviews cover completed experiments, lessons and next steps, and introduce new tests. At Tilt, the team also shared important findings at all-hands meetings, sometimes asking colleagues to guess which variation won to make the learning engaging.

A useful learning is specific enough to change the next product or strategy decision and reaches the teams that can use it. For example, advertising creative that resonates with users may also inform product messaging; analytics staff can help keep test design honest when other variables change.

Paid marketing should be an accelerant rather than a crutch, Duboe says. As a test of product-market fit, he asks what happens to retention and new users when paid acquisition stops; heavy early spending can mask weak demand and distract from more durable work.

He warns that paid performance can degrade as spending scales, making low-volume acquisition costs a misleading basis for a future business model. His rough suggestion of 30% to 50% paid acquisition for some e-commerce businesses is explicitly contextual, and he says attribution makes even that split hard to measure reliably.

A business may still need paid acquisition if no other channel works, provided customer value and payback support it. For an early company he agrees with concentrating on a working channel, while a larger business needs to consider the risk of dependence on one platform.

At Stitch Fix, he recalls a limit of no more than 50% concentration in a single channel as the company approached its IPO. When Facebook ad performance weakened, a broader mix gave the team flexibility to reallocate; the example is his account of that period.

Duboe says durable paid-marketing advantages come less from clever buying tricks than from measurement and a steady supply of effective creative. At Stitch Fix, lowering production barriers and using simpler iPhone-shot assets let the team test more ideas.

He cautions against both last-click attribution, which can credit the final interaction entirely, and expensive models that imply more precision than the data supports. Brand activity should also be measured, he argues, but often over a longer period.

For a television example, Stitch Fix paused national advertising for eight weeks and bought local spots in five regions. Duboe says the team compared lift and interactions with other channels to estimate TV's incremental contribution, including effects that a short post-ad traffic spike would miss.

For budget decisions he prefers an explicit return or payback threshold that fits the business. He suggests a first-transaction payback can be useful for e-commerce, while a subscription business may work with a longer horizon; these are heuristics, not universal rules.

He recalls Stitch Fix using about one year as a contribution-margin payback threshold, supported by its repeat-purchase data. A young company without that history should be careful about assumptions extending far into the future.

Duboe identifies two weaknesses in a blended CAC-to-LTV ratio: an early company may not know a customer's lifetime, and averaging hides differences by channel, audience or keyword. He prefers measuring payback more granularly and comparing paid acquisition costs with the returns from paid users.

He cites Bill Gurley's warning that the LTV formula is a tool, not a strategy. Better customer understanding and creative may produce a stronger advantage than manipulating a formula to justify a bigger marketing budget.

In quick-fire answers, Duboe says he no longer believes growth solves every problem: it can mask problems and create new ones. He warns founders against hiring without understanding a candidate's actual past role or treating growth as the growth team's responsibility alone.

At Tilt, he learned that repeated incentives can make organizers dependent on promotions. He says Stitch Fix subsequently avoided discounting products, and he wants more companies to see growth as a holistic system that can include large bets, not just small optimizations.

Duboe praises Faire's business-to-business referral engine, which he says brought participants across both sides of the marketplace. He reiterates that he is not an investor in Faire; the example is his assessment of a company he followed.

Stebbings thanks Duboe for the conversation, and Duboe closes the interview.