Howbout's COO Needed an Analyst. Mixpanel Agent Got the Job

Howbout is the shared calendar app for friends, couples, and families — the app millions of people use to see when everyone’s free and lock in plans without the endless group chats. Founded in 2020, the London-based startup has grown to over 10 million downloads and 200 million plans made.

Howbout runs lean: three founders, a product designer, and two engineers. There’s no dedicated analyst. Instead, anything related to Mixpanel usually goes to or through Jake – he championed the Mixpanel rollout back in 2022 and knows his way around it very well. But because he’s also a busy co-founder and COO, he’s always on the lookout for ways to make it faster.

Strategically, the team decided that some features in the Howbout app likely weren’t being used enough to justify the engineering cost of maintaining bugs within these features. So the team needed to know actual adoption and usage feature by feature, to see what was underused and could safely be cut. Done properly across an entire platform, that kind of analysis normally takes a full day. Given the outcome was to improve engineering efficiency, the CTO, Duncan, volunteered to lead on the initial analysis, with Jake to review later. 

Duncan asked Claude, via the Mixpanel MCP server, to build a full feature adoption and retention board, covering every feature across the platform. Rather than just ranking by usage, it also took into account the context about what each feature was for — directly taken from the Mixpanel Lexicon. A niche feature tied to the product’s social side, for instance, could get a ‘worth keeping’ even with low usage, while one with the same numbers but no other reason to keep it got flagged to cut. Duncan also used Mixpanel Agent to quickly create cohorts at pace, so he could further refine the analysis. The whole thing came together in about 30 minutes. Jake expected to find gaps that would send him back to the drawing board. Instead, it took him five minutes to review, and he didn’t disagree with a single recommendation.

Jake uses Mixpanel Agent as a thought partner, in addition to being a time saver. When running an experiment, for instance, Jake had to come up with a creative solution to A/B test their onboarding flow — a difficult task because social platforms come with peculiarities that make it tricky to compare clean data. Often, a user’s outcome depends on how both they and their friends engage with the platform. It’s tough to scale. So Jake formed his own hypothesis on grouping friends to get better comparison data, then he asked the agent for its take. It talked through its reasoning, and even built a scorecard showing which variant was ahead. That gave him an answer he trusted.

1. A full day of analysis, delivered in 30mn

An analysis of that scale would normally have fallen on Jake to deliver, and taken up to a full day. Instead, Duncan was able to self-serve in 30 minutes, with just a quick sense-check from Jake. That freed Jake up for the rest of his job.

2. An ambiguous experiment, turned into a confident decision

The friend-group experiment from earlier came with real ambiguity built in — a comparison over time, shifting seasonal noise, a network effect layered on top. Rather than sit on a result he wasn’t fully sure of, Jake could ask the agent to check it again a few days later and confirm nothing had moved. That’s the difference between shipping a decision and hedging on it.

3. A team empowered to self-serve with confidence

Jake knows the product well — which is exactly why he wasn’t convinced the Agent would really save him time, and he thought he’d spend his time fixing its output. He was happy to be proven wrong.

Howbout leaned into AI to move faster: the CTO now builds his own boards instead of waiting on the in-house expert, whilst the COO turns a half-day feature experimentation into a quick conversation with Mixpanel Agent.