Ad fraud evolves alongside the technologies that advertising itself is built on. One of the biggest shifts today is driven by AI: fraudulent schemes are easier to scale, and bot behavior can be tuned faster to evade the very systems designed to catch it.
This changes how we need to think about protection. It used to be enough to accurately identify invalid traffic based on how it behaved. Now, when you manage to do it matters just as much.
According to FraudScore data, the average fraud rate in analyzed traffic reached 54.3% in the first half of 2026, up from 39.7% a year earlier. For the first time since we began tracking, every month of the half-year came in above 50%. Mobile traffic drove most of the growth: 55.2%, compared to 40.0% in H1 2025. On the web, the rate actually fell from 13.4% to 10.0%.
The problem is that traditional checks work after the fact. By the time enough statistics have accumulated and fraud has been confirmed, adaptive schemes have already done their work and changed tactics. That makes it logical to look for additional control points - in particular, to evaluate traffic before attribution.
What pre-attribution brings to the table
Pre-attribution moves fraud checks to an earlier stage: the moment of the click. This changes how you deal with risk. Instead of receiving and attributing traffic first and only then analyzing its quality, some decisions can be made before a click ever enters the rest of the chain.
At the click level, a range of signals can point to likely invalid interactions: technical characteristics, behavioral patterns, repeated events, and more. But the goal isn’t simply to block as aggressively as possible. Advertisers also need to avoid affecting good traffic, so filtering has to account for the specifics of each campaign and source.
That’s the main advantage of early control: if risk can be identified before attribution, the decision is made while the traffic can still be stopped from going any further.
SafeClick: protection at the click level
To address this, we’re developing SafeClick - a pre-attribution protection tool that works directly with clicks and filters out invalid traffic before it converts. Its purpose is straightforward: to help advertisers decide on click quality as early as possible.
For mobile advertisers, this approach also has a direct financial impact. In many pricing models, mobile measurement partners (MMPs) charge based on the volume of installs and events, so every invalid click that reaches attribution can potentially drive those costs up. If part of that traffic is cut off before it enters the chain, fewer invalid installs and events end up on the bill - which means advertisers pay less not only to the publisher, but also for attribution itself.
That doesn’t mean post-analysis becomes any less important.
Why post-analysis is still essential
Early filtering has an obvious limitation: decisions are based on the signals available at the moment of the click, while the full picture often only becomes clear once enough data has built up.
This is where post-analysis is irreplaceable. It makes it possible to evaluate traffic at the campaign or source level, spot patterns that can’t be seen from a single click, classify fraud, and determine its true scale.
The results of this analysis are also highly practical. They serve as evidence when working with ad networks and traffic suppliers: to confirm an issue, file claims, and recover budget. And finally, post-analysis is what helps improve protection itself - newly discovered fraud scenarios and signals become the basis for tuning rules and models.
That’s why setting pre-attribution against post-analysis is a false choice. They look at the same problem from different angles and at different points in time.
When the two approaches work together
This is how the post + pre model emerges. With FraudScore and SafeClick working together, one layer deals with traffic that has already happened and helps you understand what went on, while the other applies that accumulated knowledge earlier - at the click level.
On one project, we manually compared post-analysis results with the results of pre-attribution filtering. A significant share of the traffic that post-analysis had flagged as risky and recommended rejecting could have been stopped at the click level. At the same time, traffic that post-analysis considered acceptable wasn’t affected by the filtering.
This is a single case, and its results can’t be automatically applied to every campaign, so we’re continuing to test the approach across a wider range of scenarios.
Next step: connecting the data
The most exciting prospect, in our view, goes beyond simply running two tools in parallel. When post-analysis uncovers a new fraud scenario, that information should flow back into the early protection system and shape how it works. In turn, pre-attribution results can become an additional data source for analyzing traffic after the fact.
The result is a loop in which each stage improves the one before it: detection → analysis → early filtering → repeat analysis. This kind of feedback matters all the more while AI is speeding up how quickly fraud scenarios adapt. Protecting ad traffic won’t come down to a single “perfect” tool. It will rely on a system’s ability to learn quickly from new data and apply it where decisions can be made sooner.
Want to see how early filtering performs on your traffic? Request early access to SafeClick on our website.
