If you believe something on this site is factually wrong, miscited or misleadingly framed, email contact@aieconomicshub.com with the page URL and the issue. See the methodology for how corrections are handled.
- · AI TCO Worksheet: The Seven-Sheet Model
The worked example deducted the AI cost twice, and priced rework at about £11 an hour instead of the £45 review rate. Recalculated: Sheet 3 is £6,688 a month, fully loaded cost £3.03 per ticket (12.3 times the naive £0.25), saving £15.47 per ticket, and net value £37,122.50 a month, not £31,476. The Sheet 6 forecast, sensitivities and variance trigger now follow from the inputs, the unquantified pilot-to-production multiplier is now described qualitatively, and the graphic is updated. View
- · Token Economics
Corrected the billing mechanics: models bill tokens sent, not context-window size; output costs commonly four to six times input; cached reads cost 2.5% to 10% of input on one major list, with a write premium. Added reasoning tokens, context re-sent on every call, tokeniser differences and the Tokenomics Foundation's 4 August 2026 launch. The worked example now compares cost per query including escalation, not token counts. View
- · Token (glossary)
The entry called the token the atomic unit of AI value, contradicting the site's own position. It now defines the token as a unit of consumption and pricing, not value. A Reasoning Tokens entry was added. View
- · The AI Value Ledger
The template's illustrative row showed claimable value below total cost yet a Scale decision. Version 1.1 clarifies the total cost and net outcome columns, adds a column comparing claimable value with total cost, and changes the example so its decision follows from its figures. Version 1.0 remains available. View
- · What CFOs Should Ask of AI ROI Claims
Bain's 44% had been described as funding AI from savings that had not yet materialised. Bain says 44% fund it from savings of earlier automation programmes, a wave that mostly fell short of target; the finding was published in June, not April. The UK government Copilot experiment had been described as rigorous baseline measurement; it relied on self-reported savings with no control group, and is now described as such. IBM Institute for Business Value and APQC finance figures (8%, 18%, six to eight months) were removed because they could not be found at source. View
- · What AI Actually Costs
Removed findings attributed to AWS, BCG and the FinOps Foundation that could not be traced, an unsourced example of Copilot subscription cancellations and an unsupported 10 to 25% hyperscaler platform margin. Rewrote the Starbucks example from press reports: the company retired an AI inventory counting tool after miscounts, not an agent overridden by managers. Marked the Uber per-engineer cost range as single-source. Replaced unlinked references with sourced ones. View
- · Productivity J-curve
The page attributed the J-curve to a 2023 study of customer-support agents, which found productivity gains rather than a dip. It now credits Brynjolfsson, Rock and Syverson (2021) for the concept and describes the 2023 study accurately. View
- · AI Value Gap and AI Economics in 5 Minutes
Removed figures that read as benchmarks but were house assumptions: a 10 to 20% starting attribution coverage, a 50 to 60% one-year target, and a claim that about twenty pence of each pound of time saved reaches the P&L. Replaced with guidance to measure the starting point and set a target trajectory. View
- · AI value-gap statistics (McKinsey)
The McKinsey State of AI 2025 finding had been described as 5 to 6% seeing clear financial ROI. The source says 39% report any EBIT impact and about 6% attribute 5% or more of EBIT to AI and report significant value. Corrected on the Five Levels page, the Value Management Layers page and diagram, and What CFOs Should Ask of AI ROI Claims. View
- · Executive confidence in measuring AI ROI
A 29% figure attributed to IBM could only be traced to a blog post. Replaced with IBM Institute for Business Value's 25% of executives who say their organisations consistently measure business value and AI ROI, labelled as an IBM-owned source. View
- · Inference forecast gap and price spread
A vendor's 30 to 50% gap between forecasts and bills had been presented as a finding, and a 9x spread had been described as fully loaded cost. The first is the vendor's own estimate with a worked example, and the second is a spread in list input prices for one model. Corrected in six articles and the waterfall graphic replaced with the vendor's worked example. View
- · Bain survey figures
Removed an unverifiable 4% over 30% figure and a revenue threshold. Restated the finding as in Bain's report: nearly 40% of those who measured landed at 0 to 10%, and 44% are funding the next wave from prior savings that fell short. Corrected on the homepage graphic and caption. View
- · Token price ranges
Ranges were out of date. Replaced with current list prices for Anthropic and Google read on 6 October 2026, and the chart rebuilt. Self-hosted range remains an unverified estimate and is labelled as such. View
- · AI billing APIs reference
The Anthropic usage endpoint was wrong, the GitHub Copilot usage endpoint had been replaced by usage metrics endpoints, and Cursor now documents an Admin API. Added a last verified table. Five providers are marked not re-verified. View
- · IDC finding on FinOps in AI projects
Wording changed from 'meaningfully integrated' to 'embedding', matching the IDC statement as reported in April 2026. View
- · Claim labels
186 claims labelled Evidence cited no source and are relabelled Interpretation. Evidence is now reserved for claims that link to a source. View
- · Flagship framework pages
Removed or relabelled figures that could not be traced: IDC 56% outside IT budgets, a 70/20/10 split attributed to McKinsey, Deloitte 16%, IBM and APQC finance benchmarks, Deloitte 2026 decision maturity, Apptio 35% and a BCG function-share figure. Verified figures from Deloitte, BCG, Kyndryl, McKinsey and Gartner now link to their sources. Layer cost ranges are labelled as this publication's estimates. View
- · Industry perspectives
Removed percentages attributed to Accenture, Microsoft and BCG that could not be traced to a source. View
- · In five minutes
Replaced an unverifiable 79% figure and corrected the Bain wording and quote. View