How the Machine Learning Team Enhances Performance

I’m Gwennaëlle Mabon, Senior Machine Learning Engineer at Adikteev. My work focuses on designing, training, deploying and maintaining the machine learning models that power our retargeting solutions. Alongside a small but highly skilled team, I translate complex user behavior data into actionable predictions that directly improve campaign outcomes.

At Adikteev, retargeting is a science, grounded in rigorous experimentation and data-driven modeling. Our models are not off-the-shelf; they’re custom-built and constantly optimized to fit each campaign’s goals—whether that means boosting ROAS, driving in-app purchases, or reactivating dormant users.

Who We Are

Our machine learning team is still growing and currently includes four members, two of whom hold PhDs. We combine academic rigor with real-world engineering to develop models that bridge the gap between theory and business impact.

Data That Matters

The model is only as good as the data, as I like to say—your meal is only as good as the ingredients you put in it.

We don’t operate on a “more data is better” philosophy. Instead, we carefully select and engineer features most relevant to each client’s KPIs.

We start from baseline bid request data, which is available across the programmatic ecosystem. From there, we build more than 40 features that capture user activity across multiple time windows, including clicks, impressions, installs, purchases, and app opens.

This multi-window, multi-signal approach lets us capture the full user funnel—from initial exposure to long-term monetization—and gives our models the context they need to make accurate predictions.

Models Built for Your Goals

No two campaigns are the same. Each requires a tailored machine learning model aligned with specific KPIs.

We train models on timestamped event sequences using action counters—metrics that track concrete user behaviors such as tutorial completions, add-to-cart events, first purchases, or app reactivations.

This lets us incorporate critical factors like recency, frequency, and engagement decay, all of which shape user behavior on mobile apps.

Models are retrained continuously to reflect seasonality, product changes, and shifting user patterns. They’re designed to be auditable and explainable, ensuring transparency and trust in their outputs.

Training at Scale

Managing data from millions of devices daily demands a scalable and reliable tech stack—and that’s a team effort.

The engineering team has built and maintains a robust infrastructure capable of handling high-volume, high-velocity data streams. This setup allows us to train complex models efficiently across about 2 million devices every day.

Thanks to their work, we can iterate and update models rapidly without passing the computational costs onto our clients. This strong infrastructure underpins our ability to deliver high-performance machine learning solutions at scale and cost-effectively.

Real-Time Impact

The ultimate goal of our machine learning work is to produce measurable improvements, not just technical achievements:

• More precise user targeting based on behavioral signals
• Smarter ad placements that translate to higher ROI
• Reduced wasted budget thanks to targeted, data-driven algorithms

This approach aligns retargeting efforts directly with client business objectives, not just superficial metrics.

What’s Next

There’s a lot more to share about the machine learning infrastructure and algorithms behind Adikteev’s retargeting technology. Stay tuned.