Notes
Four short reads
How to read the evidence; one product record from the mill to the repair bench; who is allowed to say no; and what lands on the factory when a house buys software.
What I believe
An announcement is not evidence.
Technology has to justify its place, and what usually gets in the way is a claim nobody checked, an announcement treated as proof. These five rules are how I read past that. I learned them on the product side.
-
Name the job before you claim the benefit.
A tag retrieves information. Predictive AI ranks options. Generative AI drafts text. Say which one you are buying, then ask whether better training, a correct catalogue or one owner for the information would do the same job.
-
A launch is not the same as use, and use is not the same as benefit.
Most of the evidence you will be shown in luxury stops at “delivered”. Read every claim against the ladder before you pay for it.
-
The system prepares. A person decides.
The adviser can turn down a suggestion, the artisan checks the cut, and somebody can switch the feature off. If none of those people exist, the design has a problem.
-
The promise is made in development and kept at the repair bench.
One product record, kept up to date, has to run from the yarn to the repair bench. If the campaign says “heavy rain” and the record says “water resistant”, a faster channel only spreads the contradiction faster.
-
Set the stop rule before you start.
Measure after returns and running costs, against a comparison, with the reason you would stop agreed before you start. An announcement tells you nothing about any of that.
Where AI is already doing a real job
Assist
Tiffany advisers, as LVMH has described it: an AI interface holding client preferences and product facts.
Person decides: the adviser keeps the relationship.
Make
Louis Vuitton with Comelz, shown at VivaTech 2026: hides digitised and cuts proposed.
Person decides: the artisan validates the cut.
Discover
Conversational and visual search, and shopping agents that compare on the customer’s behalf. If an agent can do the comparing, the store has to offer something an agent cannot.
Person decides: the handover to an adviser, with consent.
Create
Generative tools widen exploration. In a 2024 experiment with 293 writers, assisted work scored as more creative and became more alike.
Person decides: creative direction and the review for distinctiveness.
How I read every claim
- AnnouncedA press release, a keynote, a vendor case study.
- DeliveredIt exists and runs, somewhere, for someone.
- UsedStaff and clients return to it without being asked.
- BeneficialSomething improved against a comparison group.
- Worth the costAfter service time, returns and the licence, it still pays.
Evidence for one rung tells you nothing about the next. Too many decisions are made on the first rung as if they had reached the fifth.
Luxury by Design
Notes, talks and one annual read.
Notes, talks and one annual read, all using the same ladder on what luxury houses announce. The method is public so you can check the advice against it.
Talks and workshops
Reading the evidence
A two-hour session for a leadership or product team: the ladder, the five rules, and your own last three technology announcements read against them. Also given as guest lectures: Lectra, October 2026; Kingston University ([course], [month year]); London College of Fashion, November 2026.
- Fee
- Fixed, quoted on request for up to twelve people, excluding VAT and travel
Annual read
What was announced, and what was delivered
Once a year, the technology luxury houses have announced, read against the ladder using their own published documents: what was delivered, what is in use, what has shown a benefit, and what has shown nothing so far.
- First edition
- Thursday 25 March 2027, London, reading the 2026 announcements