HubSpot’s product analytics went from Mixpanel, to a tool they built, to Amplitude. Dan Wolchonok wrote the timeline on 16 January 2018, on his own site. He ran the analysts and data scientists who answer “how many people use this feature” for HubSpot’s product teams.
Amplitude later published a customer story from an interview with him. That page is a vendor case study. Where the two disagree in emphasis, I stay with his post and label the interview lines.
The Mixpanel → Amplitude playbook is about historical events, identity, and dashboards. Wolchonok does not describe an event export or a dual-run. This case is the buy-versus-build decision, and a cutover off the internal tool.
The path off Mixpanel
His timeline:
-
- He joins HubSpot and builds a new product with Mixpanel. The kind of analysis it allows changes how he thinks about product work.
-
- That product has hundreds of thousands of users. He gets nervous about the Mixpanel bill. He calls that nervousness a mistake in hindsight. He does not publish the amount.
- 2014 to 2015. HubSpot builds its own behavioral analytics system. The reasons he lists: they are a public company and want to own the system, an engineer’s salary might match the vendor bill, the tool can be shaped for HubSpot, and maybe they can sell the result.
-
- They compare the internal tool with a vendor. They pick Amplitude. He says the longer vendor-assessment story was going to be a later post. It is not this one.
-
- The migration finishes. He says 250 to 300 HubSpotters use Amplitude in a month.
The Amplitude page says they implemented Mixpanel and then uninstalled it, built the internal system and then scrapped it, and kept Amplitude. It says about 300 people use Amplitude. That matches his range. The page’s “more than 30,000 businesses” is HubSpot’s customer count in Amplitude’s intro, not a user count for the analytics tool.
Why the internal tool lost
On his site, the internal system was slow and it had outages. People who built it had left, and the team was understaffed. A team of three engineers, plus some of a PM’s time, was maintaining it. Amplitude’s engineering team was larger. He did not think HubSpot’s custom work would beat that, and he wanted the product engineers on HubSpot’s product.
The Amplitude interview pushes the same failure harder, in his words as they quote him. Outages and reliability were “the monster”. He calls a “death spiral”: people stopped trusting the data and stopped opening the tool. He says product people were frustrated by slow loads, changing data, and unreliable reports, and that the builders had been talented. I am using those sentences as the vendor page’s quotes, not as lines from the January post.
They ran a cost and benefit comparison. Put more people on the internal tool, or buy. Amplitude won. In the interview he says they had believed they were a special case, and that a perfect vendor does not exist. They took the tool that covered most questions, and they would customize the rest.
What Amplitude had that they wanted
Two product facts decided it, in the January post.
HubSpot is B2B. They need charts of companies, not only of users. Conversion and retention for everyone in a company. Amplitude let them flip one control on an existing chart between users and organizations. Other vendors could do account-level analysis. He thought their versions were too cumbersome.
They also wanted the behavioral data in a SQL warehouse, so the business intelligence team could join it to financial, CRM, and support data in Looker. At the time of the post that store was Redshift. He adds that it was Snowflake by the time he was writing. The Amplitude page mentions the Redshift integration and does not mention Snowflake. Use his sentence for the later warehouse.
He also cites a product-analytics roadmap that matched HubSpot’s, dashboards, and behavioral cohorts. Speed mattered because the internal system had been slow.
The interview adds that cohorts were already showing which kinds of accounts did better. He calls that discovery exciting. There is still no event volume, no instrumentation plan, and no month where both systems were scored against each other.
What I would copy
Read the bill panic against the five years that followed. He got nervous about Mixpanel in 2014, HubSpot built a tool, the tool’s team thinned out, and they bought Amplitude anyway. The January post calls the original nervousness a mistake. He still does not print the Mixpanel number, so the lesson is the sequence, not a price.
Staff the internal tool for the outages, or do not build it. Three engineers and a slice of a PM were not enough once the first authors left. The interview’s “death spiral” is what distrust looks like after slow, shifting reports. The January post already had the slowness and the turnover.
Write the account-level question down before you pick a vendor. One toggle between user and organization was the feature he could not get without pain elsewhere. The warehouse export was the second. Redshift then, Snowflake when he published.
Neither source describes a Mixpanel event import. If you need the old years, the Mixpanel → Amplitude playbook is that project. HubSpot’s public record is the decision to stop running their own system.