Joyn's product analytics bill grew with every login and few people trusted the numbers. Moving the product analytics platform from Amplitude to Mixpanel through the existing data layer cut projected cost from $220K to $179K a year and made it the group's daily decision tool.
Role Led the migration end to end: vendor evaluation and negotiation, phased delivery, and a year of hands-on adoption
The product organisation wanted PMs, CRM and marketing to explore data on their own, without the cost rising every time traffic did.
Joyn had 5–7M monthly users in 2021 and has grown to about 20M. Teams across the group needed one product analytics tool they could open daily and trust.
Amplitude charged for every event. Browsers had become far less persistent with cookies, so users were re-identified constantly and each identify call counted toward the bill.
Fixing one without the others would have left the platform expensive, ignored, or stuck behind a release queue.
Every new anonymous-ID and user-ID pairing, and every session start for a registered user, needed an identify call. Misconfigured identity events fired several times over, using the event budget without adding information.
Teams struggled with numbers that looked too high and had no training on the tool, while the product organisation wanted PMs to self-serve.
A migration that needed app and web releases would have competed with product work for engineering time. Segment was already in place as the CDP, which opened another route.
Four steps: price it, swap the destination, prove the numbers, then run adoption as a product. The track turns purple where the work moves from the build to the people.
I ran a vendor evaluation, negotiated directly with both vendors and built a cost model comparing event-based with user-based pricing. The key finding for the product analytics contract: identity calls drive cost under event pricing and are free under user pricing. The Head of Product Ops approved the decision and owned the budget.
Cost model · Vendor negotiationThe data already flowed through Segment, so apps and web needed no re-instrumentation and no releases. A working group of a data engineer, a warehousing expert, a product analyst, the Head of Reporting and a Mixpanel consultant delivered screen views, identity and video views first, then lane views and click-throughs, then data models for marketing.
Segment · Backend uploadsAn automated daily check in Prefect compared Mixpanel user counts with Snowflake, the source of truth. Early gaps swung between 2% and 19%; the daily check and a phased ramp-up of data brought them to about 5%.
Prefect · SnowflakeAfter a consultant-led kickoff per user group, I ran adoption myself: office hours every second Friday, in-person help, first-line support on Slack, FAQs written from recurring questions, and the Mixpanel Lexicon so every event and property carried one business definition.
Office hours · Slack · LexiconEngineering tracks errors, drop-offs and feature-flag results. PMs use it for problem discovery and experiment analysis. Data science visualises experiment results. Marketing builds on the models set up for it.
Content curators review it daily and weekly and use it to promote or demote lanes.
Mixpanel training is now part of onboarding for every new engineer, product person and content curator.
Lexicon definitions and the daily Snowflake check give self-service users a reason to trust what they find.
Full-volume tracking stopped scaling, so Joyn moved to 30% sampling, which cut cost about 70% against full volume. PMs want full volume back, and that decision is still open.
Let's talk
Tell me your biggest data or AI bottleneck and I'll send you how I'd approach it — the same way I worked this problem. No résumé dump, no pitch. Just a useful conversation.
Start the conversation →