Ad Tech Measures Every Impression but Misses Where B2B Deals Close

Attribution models can now credit a B2B campaign with 50 conversions, while an incrementality test might show only five of them were actually caused by the campaign. That gap, argued by Margarita Savytska in a Data-Driven Thinking column on AdExchanger, is the ad tech industry's most consequential blind spot.

The industry built the most precise measurement infrastructure in media history, Savytska writes. Every impression is counted, every click tracked, every conversion attributed. Cross-device, cross-platform, cross-session measurement became the industry's defining capability and its core value proposition. But that infrastructure does not measure influence.

Where do B2B decisions actually happen?

The fully measured buyer is the one who sees a display ad, clicks through, visits a landing page and fills out a form. Every touch point is tracked, timestamped and attributed.

Savytska then describes a buyer the industry cannot see at all: one who asks an AI assistant for a vendor recommendation, receives an answer that does not mention a given brand, forms a shortlist without visiting any website and fills out a competitor's form. That buyer is invisible to every measurement system the industry has built.

A third buyer receives a forwarded email from a colleague, clicks through from that personal message and converts. The measurement system records a direct visit with no campaign attribution. In reality, the most influential touch point — the peer recommendation — happened in a channel no ad tech platform can observe.

Each scenario represents a real buying decision. The proportion of decisions made in invisible channels is growing, while the proportion in measurable channels shrinks.

Why does incrementality matter more than attribution?

Attribution can only credit what it can see, Savytska argues. When a large part of the buyer's journey happens in AI conversations, private messages, dark social (sharing that happens through private channels such as messaging apps rather than public feeds) and peer recommendations, the model distributes full credit across a partial journey.

The visible touch points get overcredited. The invisible touch points get zero credit. The resulting picture of what is working skews toward measurable channels regardless of whether they are the influential ones. This bias compounds over time and shapes the budget.

Incrementality (the difference between what happened with the marketing activity and what would have happened without it) measures actual impact instead. Savytska gives a concrete example. A campaign reaches 1,000 buyers already in-market and attributes 50 conversions, which looks strong. An incrementality test might reveal that 45 of those 50 would have converted anyway. The real contribution was five conversions, not 50.

The industry already has the tools: holdout groups, geographic splits and matched market analysis. What it lacks, Savytska writes, is an industrywide commitment to use them as the primary measure, because incrementality almost always produces lower numbers than attribution, and lower numbers are harder to report. "The incentive structure rewards the metric that makes performance look best, rather than the metric that reflects reality," she writes.

How does AI-mediated buying widen the gap?

The problem is set to grow. When a buyer researches a purchase through an AI assistant, the research happens inside a conversation no ad tech platform can observe. The AI synthesizes information from across the web, forms a recommendation and delivers it in a chat window. No impression was served. No click happened. No cookie was set. The recommendation shaped the buyer's shortlist based on content the buyer never directly visited.

As AI purchasing agents begin evaluating B2B vendors, the top of the funnel moves into a channel current measurement cannot capture. An industry that built its value proposition on measurability now faces a future where its most consequential buyer interactions are unmeasurable by existing infrastructure.

What does decision measurement look like?

Savytska calls for a measurement paradigm built around decisions rather than impressions. Her framework has four components:

• Self-reported attribution at every high-intent conversion point. A free-text field asking "How did you hear about us?" captures channels technology cannot.
• Incrementality as the primary metric for paid activity, asking whether the activity changed behavior that would not have happened otherwise.
• Pipeline and revenue as anchor metrics: did the activity produce a pipeline, and did that pipeline close?
• Buyer research as a measurement input. Post-deal interviews and win/loss research reveal the actual decision journey. "That qualitative data is the highest-fidelity measurement available and the most underused," Savytska writes.

The industry that solved measurement for impressions now needs to solve it for decisions. That is the harder problem, Savytska concludes, and the only one that matters as the buyer's journey moves beyond the reach of the infrastructure built to measure it.