AI Users

Methodology — adoption first, then measurement

A company qualifies when it deploys AI at scale in its own business — documented, dated, and large enough to matter — and passes an entry screen: a large cap whose revenue grows faster than the median of its own industry, so insurers and banks can qualify next to platforms. After entry, membership follows AI adoption alone; the business numbers only confirm it, and count only through the Coincident exit. AI must make an existing product better; where AI is the product the company belongs to Agentic Winners instead, and suppliers of AI tools (Unity, Symbotic) stay out. No autos. Membership is never decided by price: the index measures afterwards whether early, scaled adopters outperform.

Equal weight (1/N), quarterly anchor rebalance, USD; EUR, GBP, CHF, DKK, HKD and INR closes converted daily. 5 constituents are also HALO members (marked HALO, capped at a fifth). 31 names carry adoption evidence dated before the base; the remaining one is marked PIT — no qualifying evidence by the base date — and stays until the November 2026 review decides under point-in-time rules (see the membership rules below). No constituent sits in Rubin, AW40 or the Agentic Ecosystem. Part of the AI index family: Rubin builds the factory · Agentic Ecosystem operates it · Agentic Winners sell AI products · AI Users put it to work.

Membership rules — reviewed four times a year

Constituents change only at a review, on the third Friday after each reporting season. Next review: 20 November 2026, counting evidence published by that day.

Five stages

Early Adopter: the company put new-generation AI into production early — generative or agentic AI since 2022, or the new AI its sector needs — has AI that acts in a core process, through agents or automated decisions under controls, and has stated how it will scale: targets, rollout plans for 2026 to 2028, deployment teams, next-generation AI computing. Adopter: new-generation AI is in production in a core process and does the work — AI that acts, or agents; a stated plan alone is not enough. In use: AI is in production, but only the earlier generation (fraud models, image recognition), or new-generation AI that assists, even with a stated plan. Coincident: the benefit has converged with the industry, so the edge is gone. Laggard: evidence that AI is not used, or that the programme was given up. New means anything put into production since 2022, not only generative AI: in industry and in credit, underwriting and pricing that means deterministic models with a documented learned component — certified models, autonomous inspection, new decision models — and in drug discovery AI that finds targets, predicts structures and chemistry, designs molecules and runs laboratory loops, counted by the drug programmes it reaches. The first three stay in the index. A company that has not disclosed enough is marked insufficient evidence, never Laggard by default.

Entry

A large cap enters when it has dated evidence of AI in production in a core process — the scope, a number and a date (adoption score 2 or higher) — and AI improves a business it already runs rather than being its product. Entry screen: revenue growing faster than the median of its own industry, checked when a name joins; after that, membership follows adoption alone.

Point in time

Only evidence published on or before a review date counts at that review. A name whose first qualifying evidence is dated after the base joins at the first review after that date; nothing learnt later is written back into the history.

Hold

Members stay while they are Early Adopter, Adopter or In use (adoption score 2 or higher). Price performance plays no part in membership.

Exit

A member leaves as Coincident when its benefit has converged with its industry in two reviews running; with insufficient evidence when its adoption score falls below 2; and as Laggard on evidence that the AI programme was given up, which acts at once. Evidence that is not refreshed for two reviews is flagged as stale, never downgraded for that alone.

Numbers confirm

Share prices move before profits do, so membership follows adoption and stated plans, not results. Whether revenue grows faster than headcount and whether margins hold, each against the company’s own industry, is measured for every member as confirmation. It decides only one thing: when the edge has converged with the industry, the member leaves as Coincident.

Two reviews to change

A stage change needs confirmation in two consecutive reviews, so a single quarter does not churn the index. The exception is a write-off, which acts at once.

Size and caps

Target 20 to 30 names (32 since the two additions of October 2026), floor 12 — the index is not filled up when fewer qualify. A section holds at most a quarter of the names; HALO members at most a fifth.

Weights

Equal weight, reset to 1/N at each quarter end (30 March, 30 June, 30 September, 30 December). Cap-weight, drift and Active Momentum variants run alongside.

Inditex had no qualifying evidence at the base date; it stays in the index, marked PIT, until the November 2026 review decides. Capital One and SoFi qualify on evidence published before the base and stay.

Where AI pays first — the sector maturity atlas

The research layer behind the sections: a quantitative pass over 1,931 companies above $5 billion, information date 30 Sep 2026 — how far each sector's leaders pull ahead on decoupling, revenue that grows faster than headcount.

Decoupling: annual growth of net revenue minus growth of headcount — revenue that scales without matching hiring. Leader premium = top quartile minus median of each sector; trend = change of that premium against the universe. Price-driven sectors carry no AI reading and are left out: Energy, Equity Real Estate Investment Trusts (REITs), Utilities, Materials, Real Estate Management & Development.