2 Mixpanel Pitfalls & How to Get it Right

PUBLISHED June 3 2014
by Mark Saldaña

This post is part of our Distribution series, presenting actionable material your startup can use today.

More marketing tips @500Distribution on Twitter and via Distrosnack, our daily email of bite-sized marketing tips and awesome GIFs.

Why are some users visiting, but not registering? Why are others registering, but not buying?

Here at 500 Distribution, we are big fans of Mixpanel for tracking data on user behavior in your conversion funnel. We’ve helped dozens of companies through the 500 Distribution Program and our Accelerator use Mixpanel to better understand the true shape of their funnel, address drop-off points, and make more money (or just lose less of it).

That said, despite its friendly demeanor, Mixpanel can has common pitfalls that can trip-up even moderately experienced users or technical ‘numbers’ people.

This post looks at the two most costly pitfalls that we’ve seen too many times (shakes head) over the course of helping 100s of companies understand and implement analytics-driven distribution.

Finally, we also spoke to our friends at Mixpanel to get their input on these caution areas, so many of these takeaways have come from seeing their share of smart founders and data teams make the same mistakes.

The common mistakes we see at 500 fall into fall into one of two categories:

1. Technical misunderstanding
2. Not thinking about the tool — or analytics in general — in the right way

Let’s first look at the top technical pitfall: the distinct_id.
 

Distinct ID

Mixpanel recently told us that handling the distinct_id is the single most common issue they amongst their users.

This mirrors what we see in most of our distribution work through the 500 Distribution Program, and in the guidance and office hours we do for the 500 Accelerator Program.

distinct_id is Mixpanel’s key unique identifier that’s used to track all events and people (aka, all actions and their doers) in your reports.

It’s easy to get this basic building block wrong, which launches a domino effect of useless data and bugs. In our experience through 500 Distribution, 90% of bugs can be traced to the distinct_id being wrong, or incorrectly tracked.

If your Mixpanel setup isn’t tracking the correct distinct_id — for example, if you’re not sending the right distinct ID for events — you’ll end up with cumulative event data. However, you won’t be able to tie that data to a specific user, and you won’t be able to visualize conversion actions as a funnel.

Why is the Mixpanel distinct_id so freaking confusing, and how do you solve this problem once and for all?

We went to Mixpanel to demand an answer (I mean, ask for their input with a friendly smile).

What’s the deal with distinct_id?

“Doing distinct_id correctly requires a fundamental understanding of how cookie based tracking works. Once you understand that at distinct_id is required for every event, and is the primary variable used for all our calculations, you can customize your tracking however you want — cross-platform, client or server, pretty much any configuration.

Now, it’s up to you to keep the distinct_id consistent across all of your tracking, and it’s very important that you do this correctly. It can get a little complicated but the basic structure is the same no matter what.”

Here’s how to keep the distinct_id consistent:

“The second you know who your user is (i.e., you’ve created some sort of ID for that user within your database), you need to chain your ID to Mixpanel’s distinct_id. Only then will Mixpanel will know who that user is and can proceed to show you all sorts of cool things about what they’re doing or not doing.”

Why is distinct_id such a common pitfall? Mixpanel says there are a few main reasons why people continue to have this problem.

“When people set up robust analytics, our first instinct is to start getting creative and putting our new toy together without reading the docs and setting it up correctly.

I can’t tell you how many times people say, “We’re just going to track for a month and let it pile up and look at it!”

Followed by, “Hey why doesn’t my funnel look right??”

If you mess up distinct_id, it’s hard to go back and fix it because it is the ONLY identifier Mixpanel uses. You can’t link profiles and know who is who without it, and it’s very hard to backsolve.”

The simplest fix is to take the time to QA the connection between your own identifiers and Mixpanel’s, and sidestep the distinct_id pitfall completely.

We just learned about distinct_id, the ‘minor’ detail that also happens to be the most common pitfall causing issues for new Mixpanel implementations.

But we all know that it’s not only the tool, it’s how you use it.
 

“Track Everything”

The second common no-no we see here at 500 Distribution is an approach we call “track everything.”

Data — especially loads new data — can be lull you into a drooling trance.

When you’re suddenly presented with a way to measure pretty much everything you want on your app or site, it can be tempting to do just that: track everything.

But, there’s a big problem with the “track everything” approach.

Over-tracking makes you lose sight of what you really need to watch. The shape of your conversion funnel gets obscured by a flood of data points that don’t tell you the most meaningful and actionable things you need to know — why people visit but don’t register, why people register but don’t convert, and the biggest opportunities to dial down churn.

By the way, it’s also worth noting that properly implementing analytics, sifting through data, and extracting meaningful insights takes time. If you’re an early stage company, then you already know that you have to prioritize your smart people and their good time towards the activities that are most likely to deliver the most impact.

In other words, you don’t want to “track everything.” You just want to track what matters.

We wanted to get Mixpanel’s take on this: why are smart teams so easily seduced by the promise of big data?

“Ok, please please please do not use the strategy of we’re just going to track everything and look at it later.

Two reasons:

1. First, you pay for the number of events you send to Mixpanel, so it’s best to be strategic.

2. Second, you should start with your 3 to 5 most important business questions. Only after that, then think about what metrics will help you make progress on those questions. Is it onboarding, is it user activation? Once you have metrics you’re sure of, that tie directly to those 3 – 5 most important questions about your business — not about data for data’s sake — it’s really easy to use Mixpanel.

What Mixpanel is really good at, what it lets you do, is measure your experiments and tests.

Mixpanel is an agnostic platform — you have to tell it what to do. If you don’t have a question that you’re looking to answer, it’s just going to give you pretty graphs.

We really caution against BS metrics that don’t tell you whether you’re doing well or better. In some cases, this can be pageviews or even uniques.”

How much intervention do you offer to define those questions?

“Every business is unique, but support is also part of what we do at Mixpanel. We offer it for free. Our biggest issue is people don’t leverage it enough.

There’s a surprisingly steep learning curve to this. We get it. This is the first time, EVER, that people have had this amount of data available to them. It’s so easy to send data to yourself. But what are you going to do with 10 million data points?

The most important thing is to have a question. It’s hard to get young companies to think about this up front because they’re doing so many things. They say, I don’t have time to think through Mixpanel so I’m just going to implement it get a bunch of numbers and look at it later.”

But when ‘later’ comes around, companies may find that an early bug (i.e., with distinct_id) corrupted that massive set of data, turning it into useless mush. Or, they might just have too many dots to plot out the line.

Good analytics can give you clues on your hardest questions, and even change the course of your business.

To get good means asking the right questions, even they’re broad or basic at first. For example, if you’re an ecommerce site, you might ask which customer segments are buying stuff on my site? That breaks down into, WHO’s buying WHAT on my site.

You don’t need to have analytics experience to use Mixpanel — or any analytics tool — effectively. You do need to have the right question. Don’t spray and pray, get some data around it before you spend that money!

This post was written with contributions from Robbie Faibish at Mixpanel and Matt Berman at 500 Distribution.

Get even more growth and marketing goodness from the 500 Distribution team by subscribing to Distrosnack, our daily bite-sized email of actionable growth tips and awesome GIFs.

Published by Mark Saldaña

Mark spreads the 500 message all over the world using a deadly combination of social media, PR, and baked goods. Before 500, he spent time at sexy enterprise startup Box.