想清楚、定得快：Stripe 文化与 Retool 如何保护创新

Eeke de Milliano joined Stripe in 2013, when it had around 50 people and no product managers, as its first account executive. Because most volume was inbound, she spent her days talking to customers about their payment flows, which she says is much of what PMs do and the reason she recommends a sales job as preparation for product management.

Marketplaces with complicated payment needs, such as one payer to many recipients or many payers to one, led her and engineer Brian Krausz to explore the problem together. That grew into Stripe Connect, one of her favorite products. She also loved building Radar, where machine-learning models predict fraudulent payments and the product helps merchants understand why payments were blocked and write their own rules.

She says Stripe got its first PM at about 100 employees, and that Retool similarly delayed PMs because both companies build for developers, who are effectively the customers. The need appeared when the customer base spread into different segments and the organization became heavily matrixed across countries, compliance and legal. At that point PMs help tie the whole story together and make the machine move.

De Milliano wrote a quick guide to Stripe's culture when the company was around 150 people, intended for candidates; its success metric was that more of the right people would apply and fewer of the wrong ones, including a frank section on working hard. She argues there was no single pivotal decision behind Stripe's success, and cautions that her view dates from before 2019.

What she saw instead was a respect for thinking, with "think rigorously" as a principle and meetings that asked about first principles rather than accepting "best practices". Stripe also had a strong writing culture, in business reviews, strategy memos and product reviews, and she believes writing builds clear thinkers. To this day she imagines presenting her memos to Patrick Collison, and her ideas get sharper.

The counterweight was not overthinking: Stripe asked whether a decision was a one-way door and was rigorous about which really were. Pricing, she notes, is often thought irreversible but you can grandfather existing users, while titles genuinely are hard to take back, which is why Stripe famously avoided them. Questioning the status quo, such as why merchant accounts took weeks to set up, was how Stripe started.

She says the principles were top-down from founders and hiring, and that Stripe called them operating principles rather than values, since you cannot tell people what to value. Favorites included moving with urgency and focus, users first, and "micro pessimists, macro optimists".

Rather than ask why some teams are innovative, de Milliano asks why not all of them are, since nobody wakes up wanting boring incremental work. She names fear of failure first: leaders want the upside of innovation but not its costs. Her remedies are to normalize failure, for instance by calling a postmortem a retrospective, and to lower the stakes with small teams and early customer feedback instead of year-long builds.

The other blockers are practical and human. Teams drowning in tech debt, product debt, bugs and incidents cannot focus on bigger ideas, and thinking big is hard and time-consuming for most people, especially at a startup treading water. Her answer is to give teams permission to think, building it into culture and planning.

At Retool each team charter ends with a "Think Bigger" section asking what the team would do with 20% more time, or if the team doubled. Every year the org also receives a blank doc titled Crazy Ideas, described as ideas with a 90% chance of making no sense but a 10% chance of making a 10x to 100x difference. Each year Retool ends up doing three to eight things from the previous list, and at least one or two of last year's three products came from it, including Retool Workflows.

To make failure safe, de Milliano suggests sharing learnings publicly, such as a note to the shipped email list or an all-hands presentation. On the time-to-think question, she wants leaders willing to fund a person who digs into an idea, as happened informally with Brian Krausz and the marketplace problem, and to use hackathons and planning questions to help.

The three launches were Retool Workflows, for scheduled and triggered tasks with a visual builder plus custom code; Retool Mobile, for native apps; and Retool Database, a managed Postgres database inside Retool for people without production database access. In hindsight de Milliano would sequence them, because all matured at around the same time and consumed a lot of organizational headspace, even though the teams were small. She also notes she is building on others' earlier work.

How did a company of about 300 people do it? Each effort started with one or two people for the first six months, such as an engineer and a designer, who spent time with customers and prototyping, and only got more people once there was a clear signal. Retool acted as the VC, supplying resources and its customer base, but teams had to prove ROI in engagement or revenue.

Third, and she admits controversially, the teams were deliberately kept separate from the rest of the org and the core product early on, meeting independently with a few leaders. Retool Mobile was debated as a feature of the web builder, but a separate team moved faster and discovered a different target customer halfway through. The cost is that the products must now be integrated.

De Milliano wishes companies kept time capsules of their processes at 20, 50, 100 and 500 people, because Stripe could learn from Amazon, Atlassian and Apple only where it had to go, not how to get there. Her core point is that process is variance-reducing: it brings low performers up but also pulls high performers and creative thinkers down, which is why companies add more as they grow.

Citing an EconTalk conversation between Russ Roberts and Ian Leslie, she observes that while we marvel at ChatGPT producing human-like writing, we have been asking humans to be more robot-like and formulaic, as with five-paragraph essays. If you introduce process, be sure you accept that cost and give escape hatches. She now writes on templates that this is the "minimum viable process" and that people should go higher if they can.

Still, she says every level of an organization needs three documents: a charter (mission, vision, strategy), goals and a roadmap. Teams tend to start at the roadmap and build up, but starting top-down is worth it, and the charter's horizon should shift from about three months before product-market fit to perhaps a decade for a humming company.

To let the best people work freely, she says the unlock is managers who spot who does not need process and give them air cover. She recalls a Stripe COO named Claire asking, "Are you willing to break the org for this person?"

De Milliano's first mental model is to build for your best user, not your worst: teams over-focus on abuse and misfit users, such as in onboarding, and warp the product. When a new idea at Retool raised abuse worries, founder Anthony asked whether that wouldn't be an amazing problem to have. The second, from design head Ryan, is to build the scooter and not the axle, shipping a thin slice that delivers complete value; Retool Mobile began only with field workers doing inventory management.

Her third is a 70/20/10 split of building time: 70% for the core product with product-market fit, including maintenance and tech debt, 20% for strategic non-core initiatives and 10% for bets. Lenny puts bugs and maintenance in the 20% and bets in the 10%, with similar ratios. At Retool, bug time is left to teams, from Friday bug bashes to fixing as they go.

Retool's PMs stay close to customers in four ways, she says: many were previously in customer-facing roles, all are technical, hundreds of customer Slack channels provide off-the-cuff feedback, and Retool builds Retool, with roadmaps, feature flags and task tracking living in Retool apps. On growth motion, she notes teams with product-led growth always want sales-led and vice versa. For Retool, the key is tight interaction mechanisms with sales, success and support, in both directions, and this year the team is trying an internal "science fair" with a booth per product.

On talent, de Milliano warns managers against building a team in their own image and instead recommends a balanced portfolio: homegrown PMs who carry the culture alongside PMs from other companies who bring rigor, plus execution machines and visionaries. She asks pillar leads to hire people who do not look like them and, every six months, charts the team's strengths and weaknesses and hires for the weaknesses.

In the lightning round she recommends Bird by Bird by Anne Lamott and Claire's book Scaling People, and her favorite interview question is asking candidates what they attribute their success to, without allowing "luck", to probe self-awareness and curiosity.