Predict the Future of Your Mobile App Business | AppsFlyer
app marketing predictive modelling

Predict the Future of Your Mobile App Business Without a Crystal Ball [MAMA Board]

Welcome to Edition 35 of MAMA Boards, an AppsFlyer video project featuring leading mobile marketing experts on camera. For today’s mini whiteboard master class, we have Dr. Tim Wiegels, Head of Data at FREE NOW, one of the biggest ride hailing apps in Europe.

ROI is one of the biggest metrics on which marketing performance is judged, but how can marketers get ahead of their revenue to optimize their efforts? Tim reveals one very simple secret to transforming ordinary user cohorts into anticipated profitability using his RFM predictive modeling method. Don’t know what that is? You’re about to find out! 

Real experts, real growth. That’s our motto.

Enjoy!

 

predict the future of your app business

Click the image above for a high-resolution version in a new tab.

 

Transcription

Hello, and welcome to a new edition of MAMA Boards by AppsFlyer. I’m Tim. I’m the Head of Data at FREE NOW, one of the biggest ride hailers in Europe. What am I doing there? I’m the guy that pretty much takes care of all the data stuff: data engineering, data science, business intelligence, but also marketing intelligence, which I did for most of my professional life. That’s why we today talk about “Predict The Future of Your App Business Without a Crystal Ball.” You can also do it with a crystal ball, but maybe let’s do it a bit more accurately.

Cool, let’s start! I would like to start with how to use your data. I came to many companies and one thing that quite often happens is that people don’t use their data, which led to stuff like calculating ROI on a monthly basis. That is, they would look at the campaign cost versus CLV for January, which is good, because January might have gone good or bad, but you don’t know anything about the campaigns.

So imagine you are much more advanced, and that you are actually very good data-wise, which means you have your own data. In eCommerce, this could be like basket case, conversions, whatever you want. In my case, as a Mobility app, it’s going to be registrations and the amount of times a single user takes a car, for example.

So you have this data. You also have your mobile measurement partner data, which tells you everything about your marketing parameters. (You might use AppsFlyer, because you watch AppsFlyer and this is what we’re talking about.) And what I also always recommend to people is to use a marketing cost data aggregator, because you could integrate the Google API, the Facebook API, but this is just going to be a pain for you if you do it yourself. There are companies that sell this for little money: Funnel.io, Adverity, loads more. They just give you the data. It’s very easy. Get this.

So, you have now a standard marketing data source which tells you everything about how much your campaign costs, how many conversions you had, how much CLV you actually got, and what the ROI is, which means you can say Facebook campaign Y performed better than Google campaign X. You also know how much money a certain customer actually cost you, and you can then really calculate their value.

You should actually use this to go a little bit further, which would be performing cohort reporting. What you should do is differentiate your customers, or whatever you use in your business, KPI-wise, into groups. These groups can then be compared. In our case, what we did was use acquisition cohorts, which means people that registered or people that became active in a certain month ­– in our case, between September and December 2019.

And what you can now look at is how much money they made you in the first month, how much money this became in the second month, and the third month, and so on. Therefore, this stat is actually cumulative, so it means the campaign, or the cohort of September 2019, made you 10k in the first month, 5k more in the second month, and 3k more in the month after that.

This is quite nice because you can now look at the second cohort, which is October, and you can see October performs much better – much more money the first month, much higher increase in the second month, and so on. If you now look further down into November, yeah, okay, they make more money in the first month, but then they, percentage-wise, didn’t make as much more money as the other two cohorts.

So, what does this mean? Is this actually a good cohort? Is it a bad cohort? Should you cut something in your marketing budget for the stuff you did in November? So, you should actually go into more detail. For this, it will be quite important to know what happens in the future. So, do you make more money with this, or are the campaigns useless, and just burn it? So, you now know all this.

Now, let’s answer the question, what does the future hold? So, it will be really, really good if you can have some easy method to actually find how much money your campaign makes, and how much money your partner makes in the future.

So, just imagine your data team can actually help you to look at people that have been in the system for seven days now, and can predict, on the seventh day, how much money they will make on the 30th day, on the 90th day. This would actually be great because you can understand campaign X on Facebook looks very good on day 7, but if we look into the future, these guys are one-time users. They will never make you more money.

On the other hand, you might have a campaign which doesn’t look that awesome compared to campaign X on Facebook, but will have much better users, which you know due to your predictive revenue model.

So, predicting the future. Many people think about artificial intelligence, machine learning, support vector machines, all this hot shit, all these buzzwords and data, maybe even deep learning, which is my favorite because no one knows actually what it is, and it’s way too advanced for most of this stuff. You don’t need that. You only need to have some kind of historical data, which is pretty good, so just get a bunch of data. Half a year, one year, whatever you have of all your users, and use an RFM model.

RFM sounds also fancy. Abbreviations are always cool. But this acronym actually just stands for recency, frequency, and monetary. This is data you have, so what you need to know is how to look at it. Look at what people actually did in the first seven days, looking at how many interactions each customer had. You also look at how many conversions they made after these interactions. And, finally, you look at the CLV after seven days, so how much money they made you, the actual monetary value. And since you have historical data, you already know how much money they also make on day 30, on day 90. You could even go further, but let’s just keep it these two.

And now you apply this in the RFM black box, which is a very easy model. It gives you a certain amount of clusters, which actually tell you, given this, this, and this, how much money they will make you.

As a very easy example, you have customers that don’t make any conversions in the first seven days. They also don’t make any interactions. The amount of monetary value is also zero. It’s very unlikely that these guys actually come back on day 90 and suddenly make you 100 euros. So, first cluster, flat line. You might also have a cluster where you can see, well, these guys, on average, make five conversions in the first week. They have 20 interactions with the app. They make you like 100 euros. They are probably a very, very good cluster because they will always come back. And you will know how much money they make, on average.

Don’t expect to have real numbers here. So, it’s more like, cluster one makes you between five and ten euros on day 30, cluster two makes you between 50 and 100, and cluster four makes you something between five thousand and 100,000. So, it goes way, way broader, but you just have a rough outline, and you know, ten people of cluster zero is not good, but one person of cluster four, awesome.

So, you now have the RFM model. You should apply your new precognitive powers at this point. Some people use the crystal ball. You use an RFM magic box, which is much, much more advanced. What you should do now is, every time you have people that have been in your system for seven days, you just look at the three numbers you have, and put them in one of your clusters, or buckets.

So, if you look at this, we have four different buckets here, though note this is just a rough outline, something very simple. One bucket will make you zero euro in the future on day 30. One bucket will make you something between zero and ten. The third one will make you up to 50 euro, and the last one will make you between 50 and 200 euros. Now, if you take the seven day old customers and apply your RFM magic, every one of them will fall into one of the buckets.

This is an example of a very average case. You have 30% of users that are completely useless. They will never look back at your app. They just test it out. Fine, good, cool. Then, you have some people that actually make a little bit of money. They might come back from time to time, but not really awesome users. You might have one person that is average. Lastly, in this case, we are quite lucky because we got two people that are quite good and will make you between 50 and 200 euros. That’s all, actually.

So, now you have your data set, and for every user you have in the system, after seven days, you know the amount of money they make after 30 and 90 days. Now, if we go back to the cohorts we had over here, we can actually improve them further and apply all this future knowledge. What you can see now is how much money you expect in month one and month three, which is not that important for the first cohort because you already know what happens in month three because we’re far enough along.

What’s interesting is the cohort from November because, as you might remember, we said it doesn’t look that great in the beginning. It does not look as good as October. But you now know, due to the awesome prediction power and prophecy you have, that the November cohort on day 90 will behave quite well. Maybe this is because you had people in there that, in the beginning, engage in the first month, don’t come back that often in the third and the second month, but then really become more and more engaged with time. Just an example.

And also, you can see that, for example, December is a really, really good time for engagement. It’s actually cheating a little bit because, as a ride hailer, December always works well for us because of Christmas parties, and people just get engaged with the app a lot. But as you can see, they make a lot of money. So, actually, if they come back in month one, they make loads of tours, and if they come back in month three, they made even more money in between. And that’s about it.

So, you now have this power. Obviously, don’t just use it on a monthly basis. You should actually use this on the campaign level. What you should also do is take your standard marketing data source and plug the data into this model, which means you would have the information of every campaign – and what the CLV will be after 30 days and 90 days – and can make much better decisions.

Cool. I told you everything. So, that’s it for today. I hope you enjoyed this edition. If you want to see more of MAMA Boards, go there. There, somewhere here. It will be there at some point. That’s it. So, I hope to see you soon, see you at a conference, see you wherever else. Bye!