Invest Profits, Not People: What Mixpanel Learned from Spreading Too Thin and Refocusing on Its Core

Asked what he had to unlearn moving from engineering to product, Iyengar points to the engineer's reflex of saying no to new ideas. After years of maintaining half-thought-out ideas, he says, engineers build scar tissue and an immune response that produces a hard no.

In product, ideas arrive from many more places and are fragile, so a hard no can kill a direction that might have had high reach and impact. His fix is to put his problem-solving skills toward making yes work, document how he tried, and do it sincerely rather than as a token alternative. If the path truly doesn't exist, he says, you can then say no with confidence.

For engineers who want to change, he suggests taking ten minutes to genuinely consider how an idea might work, while keeping empathy for the on-call engineer who was woken at 3 a.m. by bad ideas. Saying no after those ten minutes is still a good instinct in many cases, and he says he wishes he had learned the habit sooner.

Mixpanel launched in 2009 to provide product analytics to engineering, product and design teams. Its early growth rested on an in-house database called Arb (arbitrary segmentation) built to handle event data that is orders of magnitude larger than most data, at a time when everyone was shipping mobile apps. With an SDK installed in many apps, expanding into messaging and data infrastructure seemed natural.

By 2018 the company faced roughly 40% revenue churn in its core product. Customers still needed product analytics but were leaving for competitors with stronger features, because about half of engineering was split across three domains and could not close the gaps. Iyengar says the team made a hard call to say no to the other two categories and focus the whole engineering team on analytics.

Operationally, they discarded their existing plans, grouped the churn reasons that customer success and sales had collected, sorted them by ARR, and made the top ten items the roadmap. Engineers got direct access to the affected customers and a bucket of problems to solve. He calls it a violation of many product best practices, but says speed needed clarity and focus, and the team shipped about a hundred features in the first year.

Win rates and retention improved quickly. Iyengar says that from 2018 to roughly late 2021 or 2022, retention rose from about 60% to 90% and NPS from 16 to 50. He also notes that feature count is a vanity metric.

Iyengar says the feature sprint's downside was neglecting the product's overall design. Shipping so quickly left no time to think about system architecture, features hit diminishing returns, and each one had to be rebuilt for every part of the product, which kept its reach low. So Mixpanel started a second, design-led stream, around the time it adopted Figma, with the goal of making design a differentiator.

That stream asked what the product's key building blocks are, how few there could be, and how users discover them and how they relate. He cites Notion's pages-and-blocks architecture as an example of how a strong foundation lets many features hang off it. A concrete result was consistent, interactive visualizations, so any new chart improvement applied everywhere, and the work also made it possible to add dark mode.

Getting there was messier than the story suggests. Designers were being asked to make engineering-led projects look nice, with no time to come up for air. He and others agreed that the next three months of projects would wait on designers thinking through the product's architecture, giving them room to do it without disrupting delivery, and then the teams regrouped on strategy.

Asked whether the feature-checklist approach was right, Iyengar says it depends on context. In a competitive market with validated table-stakes features, speed matters most, but the approach outlives its usefulness quickly. His analogy: you can't mow your lawn while your house is on fire.

Iyengar says there's no hard rule about when to add product lines, but he offers Mixpanel's lesson. The company moved people away from the core to work on other things, which leaves the core open to disruption by someone who out-invests you. His takeaway: keep out-investing everyone in your core and put the profits it generates into the next venture, but don't move people.

The secondary products also sat in categories of their own, such as CDPs, message targeting or feature flagging, where Mixpanel would at best be a bolt-on. He says few customers need the sixth-best CDP or the eighth-best feature-flagging tool, and that such products might add 5 to 10% of revenue without accelerating growth, while taking engineers off the core.

If you are seeing churn to a competitor in your core and aren't best in class elsewhere, he says it may be time to reevaluate. He also warns that cutting mild successes is ten times more painful than you expect, because teams have whole roadmaps built around them. Lenny adds that a bundle needs an anchor product people won't give up.

Mixpanel organizes cross-functional teams around long-lived customer problems such as power, simplicity, data trust, onboarding, collaboration and price performance. Paired problems that trade off, like power and simplicity, go to one team so it must confront the tension. Planning runs on a six-month horizon, beginning with a leadership strategy memo, after which teams develop bets, each defined by a problem, a hypothesis, a plan to win and a way to measure it.

Unusually, Iyengar and the head of design join each team's ideation and solution discovery, collapsing the middle of the usual "W" planning process. Because the product team is small and does about ten to twelve things, he says it's worth doing something that doesn't scale. The cycle is two weeks of pre-work in December and a two-week sprint in January, ending with a roadshow to the company, and bets live in a Notion database, a one-slide-per-bet deck and an execution plan.

On prioritization, he says the trap in RICE is that confidence and effort push murky, high-reach ideas to the bottom. His teams ignore those two factors longer than is comfortable and spend a week with engineers and designers trying to solve the big ideas, often finding a higher-confidence, lower-effort route, before scoring as usual and aiming for a mix of innovative, incremental and debt work.

For estimation he likes Basecamp's *Shape Up* idea of fixing an appetite instead of estimating. Mixpanel doesn't use a fixed six weeks, but picks a reasonable appetite and asks what they would do with less or more time, then checks in at the end. Iyengar notes the full Shape Up process only works if you apply it completely, with finished scope rather than milestone one.

Around 2018, a sales engineer named Aaron automated a feed of customer gaps reported by sales and customer success into Slack. Every engineer and designer could read the raw feed with no gatekeeper, no process and no pre-aggregation. Iyengar says that at their scale it takes about 20 minutes a day, and he read it daily for years.

A ritual grew from it: engineers react with an email emoji, then email the customer, introduce themselves as the person who built the feature, and ask the "five whys." The feed now also includes Twitter, NPS feedback and win-loss notes, flowing to Slack and Notion. Records are enriched with account, ARR and CSM data from the warehouse, so engineers contact big accounts with context and tag the right people.

He cautions that you can't react to everything, especially in the first two weeks after a redesign when feedback says "I hate this, go back." Balancing that reaction is an organizational muscle his team has had to practice.

The tool stack is standard, with Slack, Zoom, Notion and Figma. The unlock is the data stack, where data from all systems lands in BigQuery, is modeled, and is pushed out with Census, so anyone who can write SQL can build internal tools and automate qualitative signals.

Iyengar's hot take is that the biggest analytics mistake is relying on client-side SDKs. On the web, ad blockers and similar issues mean you drop 20 to 30% of events, so data doesn't match internal databases. On mobile, you have to reinvent tracking for iOS and Android, creating duplicates, and you depend on users updating the app, so old broken tracking lingers.

He recommends tracking from your servers instead. That gives instant cross-platform coverage, an environment you control so updates reach every user, and a lower burden for developers, since events are just logs with structure and a user ID. Client-side context can be added later, but server-side should be the default.

On trends, he sees the data warehouse (Snowflake, BigQuery, Redshift) becoming the center of gravity for product, marketing and sales data, and events as a universal data model for analytics. Warehouses are poor tools for sequences of events, so Mixpanel is working on bringing trusted warehouse data into an event-optimized interface and tightening integrations with reverse ETL tools like Census and Hightouch.

He closes by encouraging engineers to adopt a product mindset by consuming raw customer context and talking to customers. In the lightning round he recommends *The Goal*, about the theory of constraints, and likes asking candidates to walk through their story from college to now.