Using predictive models to triple ad performance. Bidease & inDriver case study

Using predictive models to triple ad performance. Bidease & inDriver case study

If you’re interested by diversifying your advert visitors or have already determined to take action however are nonetheless hesitant about utilizing in-app instruments, this case research clearly exhibits that in-app adverts are a stable and forward-looking resolution.



On this article, we are going to discover the challenges and skepticism almost definitely to be confronted by each advertiser and share the outcomes now we have achieved.



inDriver app was launched in 2013, adopted by an immense quantity of meticulous work carried out by their extraordinarily bold group over time: inDriver went all the best way from a brilliant native ridesharing service to a worldwide product.

In April 2018, the service went worldwide, which turned out to be an enormous success, the present variety of inDriver customers being over 80 million in 33 international locations.

inDriver has a fairly unusual technique for getting into new markets. The service begins with small cities, steadily transferring on to giant cities and megalopolises.

Right here is the place we stumble upon Fable #1: DSPs have actually huge protection. Small GEOs are a real weak spot of programmatic platforms. 

This brings us to Fable #2: it takes an infinitely great amount of money and time to coach the predictive fashions, or extra particularly, to gather the required quantity of information to coach the algorithms if it is available in small parts.

So, allow us to bust these myths, we could?


The Process

The important thing activity for Bidease was to arrange efficient consumer acquisition based mostly on the CPA mannequin (signups) with a day by day cap (a sure variety of signups in sure places).

GEO: Colombia, Ecuador, Brazil, Dominicana, Chile, El Salvador, Panama, Russia;

Platforms: Android;

Sources: Bidease;

Tracker: AppsFlyer;

Timeframe: January 2021 – current.


Why In-app Visitors?

On the one hand, cell entrepreneurs know their approach across the good previous (and most significantly, time-proven) visitors sources – social media – fairly nicely. Nevertheless, as everyone knows, conventional sources are likely to run out ultimately.

Alternatively, it’s worthwhile to step up product development. This mission stays topical always, not to mention throughout fast growth.

So the place do apps discover new customers, ideally these offering the specified unit economics?

At that time, inDriver laid their eye on in-app visitors and programmatic media shopping for. And right here is why:

  • In-app visitors exhibits exponential annual development;
  • Most Android smartphone customers have a minimum of 10 apps put in moreover the default or pre-installed ones;
  • About 100,000 new apps are launched on Google Play each month;
  • Android is the world’s hottest OS, with over 2.8 billion energetic customers in 190+ international locations;
  • Individuals spend a median of 4.2 hours a day utilizing their smartphone, principally utilizing a wide range of apps (not simply social media).

A pleasant alternative to broaden the scope of your advert actions and impression, isn’t it?

We had by no means handled in-app stock earlier than partnering with Bidease, so we had some reservations about its high quality and general efficiency. We additionally heard lots about in-app visitors typically being fraudulent. In follow, nonetheless, it turned out to be fairly the alternative, and the outcomes have been past our expectations.

Aisen Syromyatnikov, Accomplice Relations Staff Chief, inDriver



  • Step 1: Goal Viewers and Visitors Sources

Initially, we determined to not restrict our potential TA, making it as huge as attainable. In any case, everybody can use a ridesharing service with none limitations.

We selected the 2 largest in-app stock exchanges, Google AdX and MoPub, as the primary visitors sources, as each of them had confirmed to work very nicely for service app promotion.

However then got here the query of GEO, as for a number of the in-app media shopping for platforms the minimal geotargeting unit is a rustic. 

We’ve got no such problem, having the ability to goal a small city or perhaps a tiny village. The nuance, nonetheless, is that you just want a significantly great amount of impressions to coach a predictive mannequin. The higher and cheaper option to cope with this activity is through the use of giant cities or grouping a number of related GEOs into one.

One other option to save money and time on coaching the mannequin by making a lookalike mannequin based mostly on the information the shopper already has about their viewers.


  • Step 2: Holistic Method to Creatives

That is the place the enjoyable half begins. 

The in-app visitors/programmatic shopping for combo makes cell app promotion extra complete, providing a chance to point out adverts to most related customers, flexibility, and customized selection of advert codecs.

We began concentrating on the sources with video adverts, that are usually extra participating for the viewers and supply extra details about the product. 

Based mostly on the outcomes, we solely left the best-performing video adverts, upgraded them visually, and reviewed the product description, making it extra detailed. We supported the movies with static native banners, which amplified the video advert message by complementary optimistic claims.

This video/native duo appeared there for a cause – it totally embraces the precise nature of the service, since normally potential prospects don’t have to go someplace urgently proper right here proper now.

Individuals did see our video adverts and even interacted with it. Nevertheless, they nonetheless hesitated to put in a brand new unfamiliar app. So we launched retargeting campaigns with native banners that labored like reminders stressing the service’s distinctive proposition as soon as once more: “You supply your individual fare”.


Let’s take a look at the figures.

Through the take a look at campaigns after we solely labored with video adverts, the i2c (impression to conversion) fee for installs averaged at 0.06%, which implies 1 set up per 1,600 impressions, and the i2c for signups (subsequent step down the app funnel) averaged at 0.05%, i.e. 1 signup per 2,000 impressions.

The video advert/native banner combo, nonetheless, improved these metrics lots: i2c for installs now averaged at 0.08%, i.e. 1 set up per 1,250 impressions, whereas the common i2c for signups reached 0.07%, i.e. 1 signup per 1,400 impressions.

We additionally used Static Interstitial and Playable Adverts infrequently to fluctuate the codecs and keep away from getting the viewers uninterested in our adverts.


  • Step 3: Including Predictive Fashions Skilled for Signups

As a subsequent step, we added the predictive fashions. By then we had already collected the required quantity of information to coach predictive algorithms on. This allowed us to optimize the prices of shopping for extremely related visitors and improve the conversion fee correspondingly.

The ultimate figures have been as follows:

I2c (impression to conversion) fee for installs at that stage was 0.2%, i.e. 1 app installs per 500 impressions. 

I2c (impression to conversion) fee for signups was 0.17%, i.e. 1 signup per 600 impressions.



In January – June 2021, the shopper acquired 200,000 installs, of which 120,000 occurred from Could to June 2021. By tough estimates, month-to-month installs and signups have tripled.

Why did advert campaigns carry out so excessive unexpectedly? 

It was all fairly easy: the AI was skilled based mostly on the incoming information, which it used to construct most correct predictive algorithms. After that, it began performing at full power whereas continuously studying and bettering the metrics.



Properly earlier than the beginning of the marketing campaign, our colleagues knew it was hardly going to be a straightforward mission and the outcomes would take time to realize. We knew we must face some difficulties, however the problem solely made the duty extra thrilling. 

Over a comparatively brief time we collected the information wanted to coach the predictive mannequin and managed to get near our shopper’s enterprise aims proper from the primary iterations. As we continued working with the fashions, our outcomes improved by the day. 

This case research is a vivid instance of how programmatic adverts can present nice outcomes given the best strategy! 

Aleksey Bashutskiy, Head of UA, Bidease


Lastly, a few phrases for individuals who are nonetheless hesitant about attempting out in-app promotion:

  • If you happen to see that your viewers from conventional sources is rising scarce and burning out, go proper for in-app;
  • In-app offers excessive viewers high quality, however the price is excessive as nicely, so that you should be ready to funds accordingly;
  • In-app visitors requires profound analytics and super-fast response to any statistical modifications;
  • In-app has a unique shopping for and rotation logic, so be affected person and don’t anticipate the outcomes straight away.

Leave a Reply