How AppsFlyer built AI into their platform

It's easy to talk about AI strategy in the abstract. It's harder to explain exactly how you're shipping it inside a platform used by thousands of companies. At HumanX 2026, Michael Grinich sat down with Eran Dunsky from AppsFlyer to dig into the specifics — how AppsFlyer is integrating AI across their marketing analytics platform, what's working, and what they've learned along the way.

What AppsFlyer does

AppsFlyer is a marketing measurement and analytics platform. If you've ever installed an app because of an ad, AppsFlyer is likely the system tracking whether that ad actually worked. They process large volumes of attribution data across mobile, web, and CTV — giving marketers visibility into what's driving real results.

That scale of data creates a strong foundation for AI. But a strong foundation doesn't mean implementation is straightforward.

Where AI fits in

Eran described two distinct tracks for AI adoption at AppsFlyer: internal tooling and customer-facing product features.

On the internal side, AppsFlyer rolled out AI-powered development tools to improve engineering productivity. Code generation, automated testing, and workflow acceleration — the kinds of improvements that compound across a large engineering org.

The customer-facing side is where things get more interesting. AppsFlyer has been building AI features that help marketers make sense of their data faster. Instead of manually slicing dashboards and hunting for anomalies, AI surfaces insights proactively. The goal isn't to replace the marketer's judgment — it's to eliminate the repetitive work that sits between a question and an answer.

The practical challenges

Shipping AI in production at AppsFlyer's scale comes with real constraints. Eran was candid about the challenges:

• Data privacy and trust. AppsFlyer handles sensitive marketing data for major brands. Every AI feature has to respect strict data boundaries — including tenant isolation and contractual data-use restrictions. You can't just throw everything into a model and hope for the best.
• Accuracy matters more than speed. In marketing attribution, a wrong answer is worse than a slow one. AppsFlyer has to validate AI outputs rigorously before surfacing them to customers.
• Adoption isn't automatic. Even when the features work well, getting customers to change their workflows takes deliberate effort — clear UX, good defaults, and gradual rollout.

What's working

Eran highlighted that the biggest wins have come from AI features that meet users where they already are. Rather than building standalone AI tools that require new workflows, AppsFlyer embedded intelligence directly into existing surfaces — dashboards, reports, and alerting systems.

This mirrors a pattern seen across enterprise software: the AI features that get the highest adoption integrate into existing workflows rather than requiring users to learn new ones. The product just gets faster and smarter.

Looking ahead

AppsFlyer is continuing to invest in AI across both tracks. Eran emphasized that the company sees AI not as a separate product line but as a capability layer that improves everything they already do. The focus remains on practical, measurable improvements — not AI for its own sake.

For teams building AI into their own platforms: start with the problems your users already have, embed intelligence into workflows they already use, and be ruthless about accuracy before you optimize for anything else.

This interview was recorded at HumanX 2026 in San Francisco.

Authors: Conner Simmons and Noelle Festa