Skip to main content
PrototypeIndependent Research Lab
Data Vintage: Blue Book 2025 / SUT 2023
AI
UK AI Economic Measurement Lab
National Accounts Thematic Research Prototype
RESEARCH AGENDA

Six Core AI Economic Measurement Challenges

Roadmap Horizon: 2026–2028

In its September 2026 methodology publication (S1), the Office for National Statistics explicitly outlined six central measurement challenges that prevent AI from being directly observed in official National Accounts. This research prototype implements empirical workflows to address each of these challenges.

1

Disaggregating Broad CPA Product Groups

Critical
ONS Measurement Problem:

Broad CPA categories (such as CPA 62.0 for computer programming or CPA 26.2 for computer manufacturing) contain both AI and conventional non-AI products. National Accounts currently cannot isolate the AI component directly.

Currently Observable Evidence:Published broad Supply & Use table totals (£101.8bn in CPA J62, £21.0bn in CPA C262).
What Remains Unmeasured:The precise monetary ratio ($s_{AI}$) of AI software/goods to non-AI software/goods across all 23 relevant CPA groups.
Proposed Empirical Test / Solution:Proportional allocation calibrated against audited corporate turnover breakdowns and specialized tech survey modules.
Q1 2027: Methodology options paper; Q2–Q4 2027: Test disaggregationInspect Workflow
2

Embedded AI and Blurred Technology Boundaries

High
ONS Measurement Problem:

AI capabilities are increasingly bundled into standard consumer goods, avionics, motor vehicles, medical scanners, and generic SaaS platforms without separate itemized pricing.

Currently Observable Evidence:Broad electronics (CPA C26.4/C26.5) and aerospace (CPA C30.3) published output.
What Remains Unmeasured:The value added by the AI algorithmic sub-component versus physical manufacturing chassis.
Proposed Empirical Test / Solution:Hedonic regression modeling decomposing price premiums of AI-enabled versus non-AI baseline products.
Q2 2027: Conceptual decomposition trialsInspect Workflow
3

AI Business Population Identification & Validation

Critical
ONS Measurement Problem:

UK SIC does not contain an AI-specific division. Experimental company lists (such as the 5,860 DSIT cohort) require ongoing validation against false positives and marketing buzzwords.

Currently Observable Evidence:Curated 60-company research benchmark; DSIT AI sector study population estimates.
What Remains Unmeasured:A dynamic, verifiable register of active UK AI producers distinguishing core developers from superficial adopters.
Proposed Empirical Test / Solution:Supervised multi-label text classification with mandatory human expert review and Companies House filing validation.
Q4 2026: Evidence and data landscape reviewInspect Workflow
4

Separating AI Revenue from Non-AI Revenue in Diversified Firms

High
ONS Measurement Problem:

Large diversified multinationals (e.g. global tech conglomerates, big-4 consultancies) generate substantial AI turnover alongside legacy consulting, advertising, and hardware sales.

Currently Observable Evidence:Total firm-level turnover and consolidated group balance sheets.
What Remains Unmeasured:Granular segment-level AI revenue attribution ratios ($r_{AI}$) for diversified corporations.
Proposed Empirical Test / Solution:Hierarchical Bayesian estimation separating firm-level AI probability ($p_{AI}$) from revenue attribution ratio ($r_{AI}$).
Q1 2027: Firm-level disaggregation methodologyInspect Workflow
5

Exploiting Alternative & Non-Survey Data Sources

Medium
ONS Measurement Problem:

Traditional annual surveys suffer from 12–24 month reporting lags. Non-survey sources (web scraping, job vacancies, API token telemetry, code repos) lack standard National Accounts sampling frames.

Currently Observable Evidence:Lightcast AI job postings indices; GitHub repository activity; public LLM token benchmarks.
What Remains Unmeasured:Representative sampling weights and price deflators for non-survey alternative metrics.
Proposed Empirical Test / Solution:Benchmarking alternative web metrics against audited Annual Business Survey (ABS) returns.
Q4 2026 – Q1 2027: Data landscape scopingInspect Workflow
6

Producing Timely, Decision-Useful Quarterly Estimates

Medium
ONS Measurement Problem:

AI technology shifts rapidly across quarterly cycles, whereas benchmark Supply and Use tables operate on annual structural releases with substantial revisions.

Currently Observable Evidence:Monthly Index of Production/Services; Quarterly BICS survey waves.
What Remains Unmeasured:High-frequency quarterly nowcasting models for AI domestic output and capital formation.
Proposed Empirical Test / Solution:Dynamic factor models and mixed-frequency nowcasting linking high-frequency cloud compute and trade flows to annual SUTs.
Q1 2028: Experimental satellite account readinessInspect Workflow