The Jevons Paradox of Insights: Cheap AI Access Won’t Shrink the Industry, It Will Rebuild It as Decision Infrastructure
AI made consulting the data corpus nearly free. Now the real premium belongs to those who can turn that abundance into decisions that actually move the business.
A pattern is getting harder to ignore in all of the data we are analyzing and synthesizing: AI adoption has gone universal. Yet the insights & analytics industry is not contracting. It is structurally upgrading, bifurcating, and re-sorting around a new control point.
As one sharp observer recently framed it (drawing on William Stanley Jevons’ 1865 The Coal Question), efficiency gains in using an input do not reduce total consumption of that input. They explode it, because cheaper access enables entirely new applications, scales, and economic activity that were previously impossible. Britain’s coal use rose more than tenfold after Watt’s more efficient steam engine, not because people used less coal per engine, but because the savings made railroads, iron smelting, and ocean shipping viable at previously unthinkable volumes.
The same dynamic is now playing out with knowledge itself in the insights industry.
AI has collapsed the cost of “consulting the corpus”: data access, basic cleaning, analysis, reporting, even initial synthesis by orders of magnitude. A reasonable observer might have predicted fewer researchers, smaller budgets, and contracting supplier revenue. The data shows the opposite. According to our sneak peek review of the forthcoming GRIT Insights Practice Report, Brand-side analytics professionals are expanding research-project budgets to their strongest level on record, with the share holding budgets above $15MM jumping from 27% to 40% since 2023. Technology spending is at new highs across every segment, with zero decreases among larger players. Mid-size service-led suppliers (101–500 FTE) are simultaneously growing revenue, capability, governance, and technology investment. The 95% of researchers already using or experimenting with AI tools (up 12 points in regular use) are not replacing the function; they are revealing how much more the function can do when the friction of basic execution disappears.
The bottleneck has migrated. AI now supplies the abundant “wheat” of raw signals, correlations, summaries, even synthetic respondents capped at the realistic ~10% augmentation level after early 2025 persona-drift failures. The value now accrues to those who “bake” it into something human organizations can actually eat and act on: novel synthesis, contextual judgment, defensible governance, workflow orchestration, and last-mile activation into decisions that change behavior, pricing, product, or media.
This is the Jevons Paradox of insights. Cheap knowledge does not shrink demand for intelligence. It explodes it, and reclassifies who captures the premium.
The Evidence Is in the Bifurcation
The cross-source synthesis make the split unmistakable:
Brand-side analytics is the expansion engine: budgets, staff, outsourcing, and technology all at record highs; method portfolios broadening; optimism highest where proximity to actual business decisions is greatest.
Brand-side researchers face the first negative staffing index in the tracked period, project spending near stagnant, and method portfolios consolidating around controllable, desk-executable approaches. This is not failure. It is the end of a particular operating model (commissioning + execution) and the beginning of another (judgment + governance + synthesis + activation).
Suppliers sort the same way. The 101–500 FTE service-led segment leads on every dimension that matters: revenue growth, formal AI governance (83%, highest in the industry), confidence in risk minimization (77%), technology spending with zero decreases, and staffing increases holding in the 40% range for a third year. The 500+ FTE segment is in managed transition: revenue flat, staffing stagnant, but technology spending at its highest since 22A and offerings shifting toward consulting, analytics, and activation. Tech-led suppliers have cooled revenue growth but sustained technology investment, deliberately retrenching toward defensible infrastructure (fraud detection, AI-enabled data collection platforms, sample rails) that mass-market generative AI cannot yet commoditize. Smaller suppliers carry principal-level expertise but face the sharpest infrastructure and governance constraints.
Only 1-in-10 teams say AI is fully integrated into workflows. That integration gap, not adoption itself, is now the primary differentiator. Organizations that merely adopt tools are being outpaced by those that orchestrate agents + human judgment across the full lifecycle. Agentic users report 84% significantly higher team efficiency and 72% say their organization now depends on research/insights significantly more than a year ago.
As the emerging decision intelligence stack makes visible, the industry is climbing from traditional market research (collect/consumer data) through unified data + AI infrastructure and predictive analytics into simulation/digital twins and ultimately AI-driven decision intelligence (recommend). AI-native entrants are attacking from the top layers. Traditional firms must either climb the stack or become the substrate that feeds it.
Implications for Research Buyers
For brand-side leaders, the implications are direct:
Let analytics lead the AI/orchestration agenda. They are already doing it with expanding budgets and clearer line of sight to commercial outcomes.
Researchers must accelerate self-service + automated capabilities or risk marginalization. The version of the role organized around project execution is under pressure. The version organized around specifying the right question, validating evidence, interpreting ambiguity, governing risk, and translating findings into decisions is becoming more central than ever.
Consolidate spend on fewer platform partners that offer clean, governable data, agent orchestration with human checkpoints, formal AI governance, full DQ transparency, and voice/AI follow-ups. Demand the three-layer model (proprietary/permissioned data + workflow-embedded technology + value-accretive services) rather than project-by-project execution.
Implications for Suppliers of All Kinds
The reclassification is already visible in the numbers:
Mid-size service-led (101–500 FTE): This is currently the winning model. Protect it aggressively and scale with broad method + tech + governance expansion. It is the cleanest pool of platform-grade targets in the market.
Large service-led (500+ FTE): Execute the repositioning to high-value data/analytics/consulting/activation platforms now. PE will accelerate the transition for those who move deliberately.
Tech-led and small: Own defensible vertical niches, sample integrity rails, or marketplace infrastructure. Avoid generic AI claims without proof of governance and data substrate.
All: Reframe your offering as “insight operating systems,” “customer intelligence platforms,” or “decision infrastructure” (not research services!) to attract capital and command premiums. The companies that translate themselves into the value frameworks capital markets actually reward will have more options, whether the counterpart is PE, strategic, or growth equity.
What Senior Leaders Should Do Now
The 2026–2027 window is decisive. To optimize for the emerging stack and organizational model:
Pick a control point and resource it ruthlessly. Trust & quality infrastructure (sample provenance, fraud detection, synthetic validation, AI audit trails), decision orchestration/workflow embedding, verified human data substrate, or last-mile activation/synthesis. Do not under-fund every layer by trying to own them all.
Treat AI as the operating system, not a tool. Set explicit task-by-task automation policies. Decide what to automate fully, what to govern tightly with human checkpoints, and what stays human-led. The people closest to the workflow AI is absorbing must be in the room where governance policy is set.
Build governance as a commercial moat. Formal guidelines + clear expectations roughly double confidence in risk minimization, and confidence correlates strongly with exceeding goals (57% vs 34% for brand-side researchers). Buyers and capital providers will reward it.
Invest in the three-layer model aligned to your control point: proprietary or hard-to-replicate data + workflow depth and APIs + services that accelerate adoption, trust, customization, and activation (not just labor).
Climb or partner into the decision intelligence stack. Move beyond collect/predict layers into simulation and recommendation — through build, buy, or deep integration — or become the high-quality substrate that AI-native platforms call.
Where the Growth Opportunities Lie
The premium is shifting to:
Trust and quality infrastructure as a standalone category, underwriteable because buyer confidence in AI risk minimization remains below 50%.
Decision-orchestration platforms that embed in workflows and become the system agents mediate.
Verified human-origin data + governed synthetic blending is the substrate on which trustworthy AI rests.
Vertical and decision-domain specialization (pricing, CX/UX, brand strategy, shopper insights, etc.) rather than horizontal breadth.
Marketplaces and activation services that become the connective rails of agentic procurement and last-mile outcome accountability.
Hub-and-spoke organizational models with insights operations as the connective tissue coordinating embedded specialists across the enterprise.
The old roll-up model of buy agencies, cut costs, call it scale is not the real opportunity. The real opportunity is taking assets the market still prices as fragmented research businesses and rebuilding them into insight infrastructure platforms. That is exactly what the Jevons dynamic demands: when the cost of the input collapses, the organizations that own the new production function (orchestration, trust, activation) capture disproportionate value.
The insights industry is not disappearing.
It is being reclassified into decision intelligence.
The leaders, whether inside brands, inside suppliers, or inside the capital that is already moving who understand this first and act on it will not just survive the transition. They will define what the next version of the industry actually looks like.
The data is clear. The window is open. The question is no longer whether to transform. It is whether you will lead the bake or be left holding the wheat.