Select Page

Brian asks, “What are the most useful analytics you track that have the biggest impact on conversions?”

Can’t see anything? Watch it on YouTube here.

Listen to the audio here:

## Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.

In today’s You ask I answer Brian asks, what are the most useful analytics you track that had the biggest impact on conversion or goal attainment? It depends, it depends.

One of the things that we have to do if we want to become proficient at analytics, is get away from the idea that there is an always guaranteed number, or an always guaranteed formula that will tell us how we’re doing that will tell us how things are going.

There isn’t.

It vary so wildly, not only from company to company, but even in your own data, to the point where you may have to do more sophisticated analysis on a regular and frequent basis to determine whether or not something’s actually working.

So let me give you an example.

What are the things that alone or in combination have the highest mathematical relationship a correlation to the outcome, you care about conversions? The answer to that requires some math.

Specifically, there’s a few different techniques you can use to get that not answered.

The first is, if you if there’s a defined time based series of data, you can do what’s called Markov chain modeling, where you are testing to see if the addition or subtraction of a channel in a chain of events, influences conversion.

And the analogy that we often use for this is something like a basketball game, right, you have a bunch of players on the court, some of those players are really good shooters, right, they can score a three point shot halfway across the court.

And they’re really, really good at getting the ball in the basket.

However, in order for them to do their job, somebody needs to get them the ball.

So there are also an equal number of players who have or don’t have good passing game.

And their passing game enables the shooters to win.

So who’s more important? Logically, we understand that there is no who’s more important, right? So those roles is vital to winning the game.

The question is, how much does a player contribute? If you pulled this player out of the game, would you score less or more baskets, right, and you swap them in with a different player.

That’s what Markov chain modeling does.

And if you have those time series based chains of evidence, you can detect that with the help of machine learning, and be very effective at modeling to understand what it is that you should be doing more of.

So when you run this, you would run an attribution analysis and figure out social media is working, or emails working, or SEO isn’t working, and so on, and so forth.

And that gives you some good guidance.

If you can’t do that, because you don’t have the data, you don’t have that chain of evidence.

Your next best bet is something like multiple regression analysis.

And this is where, imagine that all your data is on that big huge spreadsheet, you have what’s called a response column, the variable that you care about conversions, then you look at what are the mathematical correlations of every column alone or together that has the highest relationship to that conversions number? That’s a second way of doing that analysis to figure out okay, well, what are those things that indicate a relationship that you could then go test? So it may be it’s sending email on Tuesdays, and five tweets a day? I’m just making this up.

But if you did that, and he did a multiple regression analysis, and you found those things that would then give you a testing plan, right? Let’s say if you say, email on Tuesdays five tweets a day.

What happens if we go to 10 tweets a day on Tuesdays? Do we see lift right, we see a change.

That’s us another method to determine what’s working.

A third method is to use what’s called principal component analysis where you take the correlations of all these data points and you slim them down until You figure out what are the remaining variables left that make a difference of some kind that’s statistically relevant.

All three of these techniques are good.

They all have situational use.

They’re not no one technique is perfect for every single use case.

## Christopher Penn 5:22

But what this does is it lets you select the metrics that matter.

Because again, sometimes things will matter that you don’t think matter.

I ran this recently found my own data.

And I found that Twitter followers, the number of new Twitter followers, I got actually mattered, which was like all those months ago, really, but turns out it, it had a correlation.

So I went out and started some do some Twitter campaigns to try and see if I could influence the number of followers I get, and then look to see, is there a relationship, a change a lift from doing that? If there is a lift, then I know it’s causal.

more followers equals more conversions.

If I do that, and there is no lift, then I know it’s just a correlation, perhaps spurious but there’s no causal effect between followers and conversions.

So that’s the kind of thing you need to do.

That means that there is no most useful analytics, like web or social or email is there isn’t? Because it varies.

And it depends on time, it depends on who was involved, the offers being made.

There’s a lot of extra variables, but that would be the first thing I would suggest from a an analytics perspective to figure out how do we tease out this information in such a way that we can then have something to test to prove that we know what is and is not working? Really good question is a challenging question to answer.

You need some firepower, some analytical firepower to pull it off.

It’s not something that you can just do quickly in Excel.