Signal Scanner · AI & AUTOMATION

Breadth Without Depth: Why AI's Adoption Numbers Diverge Four-Fold

The debate fixates on how fast AI adoption is rising. The quieter signal: the three best US measures of it disagree fourfold, and the gap reveals AI that is widely touched but rarely material to output.

Every board deck opens with an AI adoption statistic, and the direction looks settled: usage is climbing toward near-universal. Beneath it sits a fact the headline hides. The United States runs three high-quality surveys of business AI use, and in 2026 they disagree by a factor of four: roughly 18% of firms, about 41% of workers, and around 78% on an employment-weighted basis. The spread is not noise; it reflects what each survey counts. Read carefully, it says AI has spread wide and shallow, and that your own adoption metric may measure nothing that moves output.

Signal Identification

This is an emerging measurement inflection, not a capability story. The signal is that the "AI adoption rate" executives and investors lean on has stopped meaning one thing: identical metrics now range from single digits to nearly four-fifths depending on who is asked. The Federal Reserve has begun documenting the divergence, exposing a widening gap between using AI somewhere and using it where it changes production.

Time horizon: 2–4 years (the measurement gap is visible now; the depth question resolves as 2026-2028 integration data accrues) Plausibility band: High Geographic / Jurisdictional Scope: United States primary (Census BTOS and Federal Reserve surveys); read-across to OECD economies running parallel firm-level AI surveys Sectors exposed: Corporate strategy and finance functions; enterprise software vendors; investors pricing AI exposure; information, professional services and finance as lead adopters; public statistical agencies

What's Changing

The three surveys meant to agree do not. The Federal Reserve's April note put Census firm adoption at about 18 percent at the end of 2025, work-related use in the Real-Time Population Survey at about 41 percent of the workforce, and the Survey of Business Uncertainty at an employment-weighted 78 percent (Federal Reserve Board, 03/04/2026): three answers to one question, spanning a factor of four.

The gap is a measurement effect, not a contradiction. The St. Louis Fed found worker surveys near 35 to 40 percent while the main firm survey sat at just 5 to 7 percent, most of the difference coming from how the question is asked (Federal Reserve Bank of St. Louis, 01/06/2026). The Census Bureau itself broadened its question in November 2025 from AI in producing goods or services to AI in any business function; on the wider wording, use ran 17 to 20 percent and reached 37 percent at firms with at least 250 employees (US Census Bureau, 26/05/2026).

Under the headline, use is thin. New Census research finds 18 percent of firms using AI in at least one function, 32 percent employment-weighted, yet 57 percent of adopters confine it to three or fewer functions, and only very large information, professional-services and finance firms reach 50 to 60 percent (NBER, 04/05/2026). Trade coverage reads the same data as adoption spreading yet narrow in scope (Modern Distribution Management, 27/05/2026).

Four US surveys, four "AI adoption rates"

Firm survey (narrow question) 5–7% Firms, any function (Census BTOS) 18% Workers (RPS) 41% Employment-weighted (SBU) 78%

Source basis: Federal Reserve Board FEDS Note (03/04/2026) and Federal Reserve Bank of St. Louis (01/06/2026). Bars show the 2026 estimate from each survey.

Disruption Pathway

The pathway runs in three stages. Now, the divergence is documented and the gap between headline adoption and material use becomes visible to anyone reading past the top line. Across 2026-2028, as firms move from access to integration, the measures that matter (functions touched, tasks reassigned, cost per finished output) separate from the ones that flatter and begin to sort leaders from the rest. Toward 2029, adoption breadth saturates and stops discriminating, so advantage and macro productivity turn on depth the headline surveys never captured.

Stress concentrates at three points: investors and boards that priced AI exposure off breadth and now face a depth-based repricing; statistical agencies as one "adoption rate" loses meaning and users demand intensity measures; and enterprise vendors that sold seats on universal uptake, now asked what actually moved. Two adaptations follow. Leading firms retire adoption as a target for depth metrics tied to output and cost, as buyers question spend that outran returns (Axios, 28/05/2026). And agencies extend surveys toward functions, tasks and outcomes, as the Census supplement has begun to do (US Census Bureau, 26/05/2026).

Why This Matters

For boards, CFOs and investors, the metric to distrust is the adoption headline. A number that ranges from 5 to 78 percent depending on the survey cannot anchor a capital or competitive judgement, yet it routinely does. The scorecard needs rebuilding: replace "are we using AI" with "where has AI changed a function, a cost, or an output", because only the second predicts returns. Firms that read high reported adoption as progress will over-invest in breadth and under-invest in the reorganisation that creates value; those that measure depth learn early whether their spend is landing. The near-term risk is not falling behind on adoption, but mistaking adoption for advantage.

Decision-action posture for this signal: Prepare — the measurement gap is visible now, so boards should retool their AI scorecards this cycle and commit capital on depth-of-integration evidence rather than headline adoption.

Counter-Argument

The strongest objection is that this is a lag, not a mirage. The J-curve of general-purpose technologies predicts this shape: broad early adoption, thin measured impact, then a sharp payoff as firms reorganise. Harvard's Belfer Center argues real value is already there for firms that aim AI at innovation rather than cost-cutting, and that they out-earn the rest (Belfer Center, 25/06/2026). On this reading the 78 percent is the leading indicator and the shallow-use numbers are simply early.

The objection is right that depth will come for some, but it sharpens the signal rather than dissolving it. If value accrues only where AI is genuinely integrated, an adoption rate that counts a marketing chatbot the same as a re-engineered claims process is the wrong gauge, and its four-fold spread proves it. Lagged or absent, the decision is identical: track depth, not breadth. The mirage is not that AI does nothing; it is that "adoption" tells you whether it is working for you.

Implications

This is a durable measurement problem, not a passing data quirk. While AI stays easy to touch and hard to embed, breadth and depth will part company, and any single adoption figure will flatter or frighten depending on its wording. The inflection window is 2026-2028, as integration data accrues and the depth distribution reveals itself. Those positioned to gain run intensity metrics; those exposed manage to an adoption headline, or priced AI exposure off it. The number to watch stopped being how many have adopted; it is how much of the work has actually changed.

Early Indicators to Monitor

Disconfirming Signals

Strategic Questions

Keywords

AI adoption measurement; Business Trends and Outlook Survey; employment-weighted adoption; AI productivity paradox; enterprise AI integration; depth versus breadth; Federal Reserve AI surveys; survey question framing; AI value capture; automation metrics

Bibliography

Source tiers: Tier 1, governments, regulators and intergovernmental bodies. Tier 2, think-tanks, academic institutes, major consultancies and quality data providers. Tier 3, quality journalism and specialist trade press. Tier 4, vendor, company and practitioner sources, used only as directional corroboration.

This Signal Scan is one component of a Shaping Tomorrow intelligence cycle on AI & Automation. Every monitored topic also includes monthly Change Trackers that assess what is materially shifting in momentum, and a quarterly Decision Intelligence briefing that converts the cycle into board-level strategic implications. Cycles may also close with Action Triggers, an end-of-cycle synthesis that reads across the three formats and surfaces patterns no single deliverable carries on its own. Continuous intelligence on this topic is available as a Single-Topic or Multi-Topic programme.

Discuss a Tailored Programme

Prepared by Shaping Tomorrow: 18 July 2026