How IMVU Uses Machine Learning to Scale Mobile User Acquisition

By Lomit Patel | September 12, 2018

If you are still manually optimizing campaigns the same way it was done half a decade ago, you may be a quickly disappearing breed in the mobile user acquisition space.

More than ever, machine learning (ML) is being incorporated in one form or another into almost every level of the mobile acquisition value chain, from distributing inventory more efficiently to allowing advertisers to scale their decision making. Any process that is still being done manually is likely less effective than it could be as new solutions are quickly emerging to attack inefficiencies. Today’s tech layers and solutions can potentially allow us to:

  • Increase efficiency of UA operations
  • Reduce unnecessary work and hires
  • Focus on higher level KPIs and make broader decisions
  • Focus on developing compelling messaging and creative
  • Find unexpected hidden opportunities

Machine learning is such a ubiquitous term these days. At its core, machine learning is a type of artificial intelligence (AI) developed around allowing computer system to progressively improve performance on a task by “learning” through statistical approaches. Put another way, machine learning is the development of algorithms that allow for more and more accurate prediction with incremental collection of data.

Let’s dive into how machine learning approaches are implemented in the RTB and supply side of the ecosystem and how IMVU uses them to maximize UA efforts.

Mobile Demand-Side-Platforms

The mobile advertising exchange space was the original wild west of mobile UA as conversion optimization wasn’t regulated by any single player. The leaders in this vertical originally stayed ahead by rapid opportunity arbitrage, fast optimization algorithms, and deep knowledge of publishers. However, these competitive factors have quickly evolved as machine learning has become a critical cornerstone of each successful Demand Side Platform (DSP). All DSPs now need to be able to make rapid evaluations of the real-time-bidding (RTB) landscape to get to advertiser goals as quickly as possible with minimal waste in order to stand out.

Liftoff is an example of a DSP that IMVU works with to leverage a machine learning approach across RTB. To run effective campaigns with Liftoff, it is important for us to pass sequential major app events to them in order to allow their algorithm to learn as quickly as possible about which targets work and which do not. By sharing exclusion lists with Liftoff, they are able to build an app graph to understand which types of devices and apps are most correlated to success for our apps. Over time, as more data is collected and Liftoff’s optimization predictions improve, we are able to gradually focus on targeting events further and further down the funnel.

Supply Partners

The supply side of the marketing ecosystem is perhaps the most motivated to maximize the efficiency of its inventory because every improperly served or targeted impression is effectively an economic loss. Traditionally, closed advertising networks have been quick to develop effective conversion optimization techniques, initially focused on driving installs from their publishers. In recent years, leading publishers such as Facebook and Google have taken the lead in taking advertising machine learning to the next level.

Facebook has implemented machine learning extensively for both conversion optimization as well as user targeting with the goal of maximizing its effective CPM payout while concurrently achieving advertiser goals. To improve conversion optimization, Facebook’s algorithm modulates a slew of variables, from time and frequency to creative placement and affinity to perform a desired action. User targeting based off of Facebook’s lookalike modeling takes into account countless user specific, social, engagement, and probabilistic attributes to produce an incredibly precise prediction of similarly valuable users. Ultimately, the combination of these approaches allow Facebook to not only be effective at driving results, but to continuously be improving over time.

Google is the other supply side giant using complex machine learning models to drive success for advertisers. Most marketers familiar with Google are deeply familiar with Conversion Optimizer, its automated bid strategy that focuses on maximizing conversions according to advertiser goals. Conversion Optimizer has been around for nearly a decade and has evolved gradually over the years, most recently becoming Target CPA bidding (tCPA). Though Google Universal App Campaigns (UAC) rely on machine learning just as much as Facebook ads do, Google takes a different approach to training its optimization algorithms for each advertiser.

To build a comprehensive model of what types of publishers, placements, and demographics convert for an app, UAC begins from the ground up by testing all factors far and wide and gradually narrowing down efficient targets conversion by conversion. After uncovering which attributes do and do not resonate with the app, the tCPA algorithm then focuses on driving conversions at the desired cost based on its cumulative learnings, after which, conversion optimization can then proceed to a conversion event further down the funnel. While a well optimized UAC campaign can drive consistent results over time, its approach requires a considerable training budget to reach maturity.

Key Differentiators Between Liftoff vs. Google UAC & Facebook

At the core, Liftoff, Google UAC & Facebook approaches to UA are similar – they each use machine learning technology and dynamic creative optimization to target post-install event conversions and acquire users that are most likely to convert and perform valuable actions inside our apps. The biggest key differences are below:

  • Google and Facebook each have access to a wealth of precise and deterministic user data which is a primary factor in their ability to perform well at scale. This advantage allows them to layer on powerful signals on top of advertisers’ own first party data (for example, search is strong intent signal to Google that is used to enhance targeting).
  • Liftoff focuses on mobile ad exchange traffic, reaching users in over 500,000 mobile apps, while Google and Facebook both access inventory from their direct and extended networks.
  • Liftoff provides advertisers with full transparency of which sub-publisher sources their ad are served on, while Facebook and Google are black-box outside of their owned inventory.
  • Liftoff is a fully managed service, while Google UAC and Facebook only offer self service platforms, which make managing campaigns much easier with a lean in-house team.
  • With enough conversions, Liftoff offers CPA pricing that allows marketers to de-risk their spend unlike Google & Facebook who bill (primarily) on upper funnel CPC and CPM models.
  • Liftoff offers the ability to group events internally to expedite optimization. If they see that gaming app users who complete level 10 within day 1 have a higher LTV, for example, they can group the ‘level 10’ event with a D1 time layer to create a new optimization event. They can also combine events to a virtual event that fires when a user either completes level 10 or completes a purchase.

The mobile advertising ecosystem is evolving rapidly from both a supply and technology standpoint year over year. While there are certainly pros and cons to using one vs another, the best approach is the one that allows you to maximize your team’s time and effectiveness by eliminating key bottlenecks and inefficiencies. Our recommendation is to include a DSP like Liftoff as part of a diversified UA strategy to compliment the duopoly of Google and Facebook in your channel mix.

Lomit Patel heads up the Growth team at IMVU which is responsible for driving user acquisition, retention and monetization across all platforms (iOS, Android, and Web). Lomit is a seasoned growth marketing executive with expertise in building and scaling up customer acquisition, retention and monetization channels at early and mid-stage consumer tech startups.

Learn more from his Mobile Hero profile.