
ASOS Credited Personalised Search With £31m. Its Try-On Got Adjectives.
ASOS's FY26 results put a pound figure on one AI result, personalised search at 'c.£31m of annualised revenue', a number that 1.4% of the year's entire revenue reproduces. The try-on gets adjectives and the rest get percentages and counts, while the cyber incident's 2–4% hit is a share of a few weeks of billed GMV, not of the year's net figure.
Neritus Vale
ASOS’s results for the year to 30 August put a pound figure on one AI result. Personalised search “alone lifted conversion by 140bps, equivalent to c.£31m of annualised revenue”, the statement says. Every other result gets a count, a percentage or, in the try-on’s case, adjectives. The cyber incident’s trading impact, the obvious figure to set against the £31m, sits on a different base from headline GMV.
The statement does not show its working, but one sum reproduces the figure: 1.4% of the year’s entire adjusted revenue, £2.24bn. If that is the method, the lift has been applied to every pound ASOS took, whether or not the shopper searched, and the match may be coincidence until ASOS confirms it. That is sound only if the lift applied to all shoppers, and the statement gives the 140bps for users “exposed to the updated app features”.
Six weeks earlier, ASOS’s pre-close update described the same work as a comparison, crediting recommendations within search with “materially higher conversion rates relative to traditional customer journeys”. Shoppers who pick a recommendation select themselves, so on that reading the gap measures them as well as the feature. The results statement says “lifted” and names no randomised test behind it.
Drapers and City AM led on growth and profit, and City AM added that investors “have been buoyed by the group’s use of artificial intelligence to get ahead of fashion trends”. That is trend-spotting, a different use from search. Sorted by what they measure, the statement’s AI and personalisation claims leave one column nearly empty.
| Priced in pounds | Measured, not priced |
|---|---|
| Personalised search: +140bps conversion, c.£31m annualised revenue | AI Assistant: 4m conversations, containment +24ppt |
| Experiments: more than doubled, positive share almost fivefold | |
| Missed Product Delivery Promise: c.40% better, with AI “increasingly supporting the team” | |
| AI Studios: in production since H2 | |
| Virtual Try-On: higher conversion, no figure | |
| ChatGPT app: traffic and bounce rate, base unstated | |
| Lifecycle CRM: visits +30%, reactivated engagement +28% |
The table covers the statement’s named AI and personalisation initiatives. It leaves out efficiency claims such as about 500 hours of manual work removed each week and development effort cut by c.87%, which are given in hours and percentages, not pounds.
The assistant’s measures changed between April and November. The interim statement said the cost per customer contact the assistant contained had fallen “by more than 90%”. The full-year statement keeps the volume, adds containment and, on cost, says only “reducing cost to serve”. It also expects AI translation to cut localisation costs by more than 90%, a forecast with no pound figure attached. The one reported cost per contact dropped out.
The try-on test grew without acquiring a number. ASOS reports “higher conversion and revenue per customer and a directional reduction in returns”, with the strongest gains among newer customers. “Directional” suggests a result that points the right way without reaching significance, though the statement does not define it. In October the test widened to all UK Womenswear, about a third of the catalogue, still without a figure.
The ChatGPT app, launched on 20 May, has figures without a denominator. Its traffic and bounce-rate gains, reported earlier today, do not say whose visits they count: ASOS.com’s, or only those ChatGPT sends. The same sentence announces a dedicated agentic commerce team, a cost that arrives before any sales figure.
The CRM programme, which the statement measures only in growth rates, was among the activity ASOS paused after the incident. Note 20 describes that incident as unauthorised access to “certain customer information held on third party platforms”. The pause covered “certain marketing and CRM activity” and ended within two days. The intruders’ alert had come through the platforms ASOS uses to message customers, as we reported.
The incident’s estimated trading impact carries a footnote that changes its base. The hit was “most notably in the UK, with the latest impact since 6 October estimated at c.2–4% of Group GMV”, the statement says. Footnote 6 defines that GMV as “GMV excluding tax on a billed basis”, the measure the statement also uses for category and brand growth, so returns are not netted off. Headline GMV, under footnote 4, is “net of returns”.
Read with its date and footnote, the percentage covers a few weeks of billed sales, not a year. Note 20 says the incident “is not expected to have a significant impact on the business”, so the 2–4% is a share of sales affected, not an estimate of cost. City AM’s story sat at a web address reading “up to £94m”, which is 4% of the year’s £2.35bn of GMV.
City AM’s current version puts October’s hit at £4m to £8m, the same rates applied to a typical month’s GMV. A month of the search gain is about £2.6m, so on that arithmetic the incident’s GMV hit exceeded it. The bases still differ: one figure is annualised revenue, the other a share of billed sales that the statement never sizes.
The statement’s grammar hangs the lost trade on the pause: “This has resulted in” follows the sentence about it. Yet the pause lasted two days at most, and the impact runs for weeks. The breach itself may have kept customers away, and the statement does not separate the two.
The CRM never gets a sales figure of its own: the 2–4% in the sentence after the pause is a group-level trading impact, with marketing and CRM taken together.
The strongest objection is that activity measures are the honest record for tools this young. On this view, features launched beside a rebuilt app and a bigger marketing budget cannot be priced without extrapolation. For the thesis to fail, the AI’s money would have to sit unlabelled in the accounts, in a gross margin up 340bps with a fifth of buying “AI-led and buyer-refined”.
The answer sits in the same statement. ASOS credits the margin to better buying and sourcing, flexible fulfilment and full-price sales, without naming AI. Elsewhere it uses controls readily: loyalty Tier 3 members delivered 17% higher Net ACV “versus comparable controls”. Exchanges, launched in March, gets pounds: it has “retained c.£5m of GMV”. For the try-on, the ChatGPT app and the CRM, it published adjectives and activity measures.
The FY27 guidance gives AI no figure either: low-single-digit GMV growth and adjusted EBITDA of £180–200m, with The Stylist’s growth “from FY28 onwards”. The promised Capital Markets Event, with “an updated medium-term financial framework”, gives ASOS a choice of columns for The Stylist.
If ASOS prices the try-on and the ChatGPT app as it priced search, with the working shown, investors can weigh The Stylist against its cost. If it brings another page of conversations and containment rates, it will be asking them to fund an ambition measured in activity, with one unexplained £31m standing in for the rest.