The AI Gap and How to Measure It

Billions of dollars have poured into AI over the past two years, and the returns are arriving slower, and more unevenly, than the investment case promised. [1] Marketing teams are restructuring around AI-native workflows. Production, selling, and distribution are all being re-architected at once. And yet, in category after category, adoption is not tracking with spending on AI. The gap between what companies are building and what their customers are willing to do with it keeps showing up as a surprise.

Consumers’ lack of trust in AI further complicates the commercial potential of AI. According to Iris Flex's AI Gap Study, a nationally representative study of 1,003 U.S. adults, [2] four out of five U.S. adults said they don't trust AI-driven advertising or product recommendations — a deficit that can explain why AI investment and AI return are moving on different timelines.

This is a measurement problem, hiding in plain sight.

Most of what passes for AI research right now is reach and sentiment: how many people have heard of generative tools, how many say they're “interested in” AI-powered experiences, how favorably a brand's AI initiative polls in a tracker. That's basic, needed, but surface-level. It tells you whether people have noticed AI. It tells you almost nothing about whether they'll let it operate on their behalf, in their categories, with their money — a key factor for return on investment.

Real measurement has to go deeper than reach or sentiment alone. Four dimensions matter here, and they matter together:

Exposure — how much AI a person actually encounters day to day

Comfort — how at ease they are with it once they do

Context — does a person's willingness to engage with AI shift depending on the category, the stakes, the moment

Delegation — will they actually hand a decision to AI, or do they want AI's input while keeping the final call themselves.

These four dimensions come out of two national studies I fielded through Iris Flex over the past year.[3] Neither dimension on its own paints the full picture. Someone can be highly exposed to AI and comfortable with it in the abstract, and still refuse to delegate a purchase decision in a category that feels personal or high-stakes (e.g., a financial services product related to retirement). Someone else may resist AI everywhere except the one place where it demonstrably saves them time – work. The interaction between exposure, context, and willingness to delegate that determines real-world adoption.[4] Averaging it away is how companies end up investing in “AI experiences” that a majority of their actual customers won't use.

I brought these four factors together in a framework I call the AI Proximity Index (APX), as a companion to my book, AI for Marketing: The Consumer Perspective. [5] APX measures those four dimensions and maps people into eight behavioral segments, from Power Users who delegate freely across categories, to Avoiders who resist even after repeated exposure. [6] Overall, US adults are highly exposed to AI, but not so ready to delegate to it. There is a wide gap (roughly 30 points of Net APX Score) between how exposed people are to AI and how far they're actually willing to let it go.

Understanding this gap is useful, but knowing where it shows up is actionable. What makes this actionable rather than merely descriptive is what happens when you apply it to a real audience. Mapping [7] APX scores onto a national media-use dataset reveals how AI proximity varies by the media people consume weekly.

The pattern reveals a divide between digital and broadcast media. Podcast listeners, gamers, streaming and online-video audiences skew above the population average on AI proximity — more willing to let AI operate in their decisions. Traditional broadcast and radio audiences skew at or below average. That's a behavioral difference, and it's exactly the kind of signal that reach metrics and sentiment trackers miss, because it shows up when you measure context and delegation together, zoomed in on specific audiences.

There is an opportunity in front of the insights profession right now. Researchers are particularly well positioned to tell a Chief Marketing Officer, Chief Data Officer, or CEO specifically where their customers sit on the spectrum between AI exposure and AI delegation — and therefore where to invest, where to hold back, and where the next competitor is likely to get it wrong first. That is applied behavioral research. It draws on the same rigor our profession already brings to segmentation, message testing, and journey mapping. It just needs to be pointed at a new, more urgent question.

The harder part isn't the methodology. It's the translation. A segmentation model with eight named types and a proximity score is a research deliverable. A roadmap that tells a C-suite which categories are ready for AI-native experiences, which need a human-in-the-loop bridge, and which should stay untouched for another eighteen months — that's strategy. Insights professionals don't always take that next step from measurement to strategy, but this moment rewards the ones who do. The organizations currently restructuring their production, sales, and distribution models around AI need someone in the room who can say, with evidence, not just what their audience feels about AI, but how far that audience will actually let it go — and be willing to put that finding in front of the board, not just in an appendix.

The business of AI is not going to slow down to wait for measurement to catch up on its own. It's on us to close that gap, and to walk into the C-suite with a strategy.

References

[1] BCG's most recent research on AI value found that a majority of companies report little or no measurable revenue or cost benefit from their AI investments despite substantial spending, and that only a small fraction have advanced beyond proof-of-concept to real value creation at scale. See BCG, “Are You Generating Value from AI? The Widening Gap.”

[2] Iris Flex AI Gap Study, n=1,003, nationally representative U.S. sample.

[3] Iris Flex AI Gap Study (n=1,003, nationally representative U.S. sample) and APX Proof-of-Concept Study (n=300, pre-registered on OSF, fielded via Pollfish).

[4] “Interaction” here refers to the combined, non-additive effect of exposure, context, and willingness to delegate — not any single dimension considered on its own.

[5] Idil Cakim, AI for Marketing: The Consumer Perspective (CRC Press/Taylor & Francis, July 29, 2026). https://a.co/d/09dEnkwA

[6] APX Proof-of-Concept Study, n=300, pre-registered on OSF, fielded via Pollfish.

[7] APX proxy scores were appended to AI Gap Study respondents using a crosswalk methodology matched on shared demographic and behavioral markers, allowing APX behavioral segmentation to be projected onto a broader national media-use dataset.