Cases

The method, applied to real decisions.

Three kinds of case, each labelled for what it is: decisions read from a house’s own published documents; one invented decision worked through in full; and, as soon as a client allows it, engagements described with permission. Nothing here is dressed up as more than it is.

Read from public documents

Four technology decisions by one house, read on the ladder using only what the company itself has published. Burberry is the example because it has published more, for longer, than any comparable house, and because in 2016 it let its finance chief say in public that a much-publicised change had not moved the numbers. I developed product at Burberry; nothing below draws on anything learned inside the company.

2012 · Regent Street

A flagship with a hundred screens and tagged merchandise

Pick up a trench coat near a mirror and the mirror showed the garment being made or walking the runway. The tag identified the item and the content appeared.

Rung reached: delivered.

No public source shows whether customers understood the product better, whether they felt served or competed with, or what it cost to keep every product linked to the right film as collections changed. Fourteen years on, the company’s own account of store productivity names clienteling, visual merchandising and category destinations. The screens are gone from the story, and the flagship is to reopen in 2027 as Burberry House, with an archive, hospitality and private rooms for aftercare.

Burberry, Annual Review 2012/13; Evening Standard, 13 September 2012; Burberry, 26 October 2026.

2016 · See now, buy now

Runway collections made purchasable the day they were shown

By February 2017 the full collection was in 80 stores and three looks in more than 300, online as shown.

Rung reached: delivered, and partly used.

In October 2016 the finance chief said the first show had not meaningfully affected quarterly sales. In May 2017 February show sales were reported 25 per cent above September’s. By September 2018, under a new creative lead, the house was back on the conventional calendar. None of the figures shows the incremental profit from the model after the inventory risk of producing a full collection ahead of demand. Removing a delay for the customer became a synchronisation job for merchandising, production, allocation and fulfilment.

Burberry, Annual Report 2016/17; Fashionista, 18 October 2016 and 17 September 2018; Glossy, 18 May 2017.

2020 · Shenzhen

A social retail store built with Tencent

A WeChat mini-programme linked product information, appointments and a café, with social currency earned through participation.

Rung reached: delivered.

In practice: platform integration, interactive content and a game. No public source reports attended appointments, repeat visits or extra full-price sales. The question that carries to other cities is whether a reward for scanning measures interest in the product or interest in the reward.

Burberry, Annual Report 2020/21; launch release, 31 July 2020.

2025/26 · AI clienteling

An AI-driven clienteling programme, governed by a steering committee

The annual report to March 2026 says the programme is in use to improve personalisation, engagement and retention. In May 2026 the company showed the associate tool and an AI-drafted outreach message; in July it said store productivity had improved through clienteling, among other things.

Rung reached: delivered, with use claimed.

The first of the four that involves a system that learns rather than retrieves, and the company does draw a line from clienteling to productivity. Still open, and what an evaluation would need: the kind of model, who uses it and how widely, what an associate may decline, and what changed for a client compared with good clienteling without it. A reader who takes “using an AI-driven clienteling programme” as evidence of benefit has gone two rungs further than the document.

Burberry, Annual Report 2025/26; preliminary results, 14 May 2026; Q1 trading update, 17 July 2026.

For the year to March 2026 Burberry reported revenue of £2,420 million and adjusted operating profit of £160 million. Those are group results, not a return on any decision above. The temptation to draw that line anyway is the habit the ladder is meant to break. The full reading, with every source, is note 1 of Luxury by Design.

Two more, from the maker’s side

2026 · Louis Vuitton with Comelz

Leather hides digitised and cuts proposed by the system

Shown at VivaTech 2026: the hide is scanned, the cutting plan is proposed, and the artisan validates the cut before it is made.

Rung reached: delivered, as demonstrated.

What makes it worth watching is the design, not the demonstration: the decision stays with the person who knows the leather. What is not public is yield against the same artisans cutting by eye, or what the system costs to keep calibrated as suppliers and hides change.

VivaTech 2026, as shown and reported.

Tiffany · as LVMH describes it

An AI interface holding client preferences and product facts for advisers

The adviser keeps the relationship; the system holds what the adviser would otherwise have to remember or look up.

Rung reached: delivered, with use described.

The right shape for assistance in a luxury store, on the group’s own description. The evaluation question is the same as for every clienteling tool: what the adviser may decline, and what changed for the client against good clienteling without it.

LVMH, as published; see the beliefs page for the full panel.

Worked through in full

Two invented decisions: a luxury house’s AI clienteling rollout, read first as a two-page Report Read and then as a four-week Decision Review; and a knitwear manufacturer deciding whether to put AI into product development, read as a Decision Review. Neither is a client project and the figures are made up.

See the worked examples

From my own engagements

Client work is described here only with the client’s permission, anonymised to the kind of business and the decision. The first case will appear when it can.

Example output

The same decision, read two ways.

Two made-up decisions. The first, a luxury house’s AI clienteling rollout, is shown as a Report Read and then as a Decision Review, so you can see what each one gives you. The second, a manufacturer deciding whether to put AI into product development, is shown as a Decision Review. Neither is a client project, and the figures are invented.

Illustrative Report Read

The clienteling pilot report, read in two pages

Document read: the vendor’s 14-page pilot report and the house’s cover memo recommending rollout to all 38 stores · Nothing else read, nobody interviewed

The read in one line

The tool was delivered and used in six stores for twelve weeks. The report does not show that it helped, because nothing was compared, and it treats the activity the tool generates as the benefit.

What the document proves, rung by rung

RungWhat the document showsReached?
Announced“Personalised outreach at scale; measurable uplift in engagement”, p. 2
Yes
DeliveredLive in six stores from March; 41 advisers onboarded; 2,310 messages drafted, p. 4
Yes
Used28 of 41 advisers sent a drafted message in week 12; the other 13 are not asked why, p. 6
Partly
BeneficialReply rate 19% “up from 11%”: the same stores’ previous quarter, before the new collection; no comparison stores, p. 8
Not shown
Worth the costLicence per store, “no incremental headcount”; adviser minutes per message and CRM clean-up absent, p. 12
Not shown

Rung reached: delivered, and partly used. The cover memo treats it as beneficial. The report does not support that.

The simpler fix it did not compare itself with

The drafts draw on CRM preference fields the report itself calls “incomplete for a majority of returning clients”, which is why advisers rewrote most of them. Filling those fields for each store’s top 200 clients is about four weeks of the existing team’s time and improves outreach with or without a drafting tool. The report does not test this, because the vendor does not sell it.

The one question to ask before you sign

“Which stores were the comparison, and what did their advisers’ reply rate do over the same twelve weeks?”

If there were none, the report cannot support a rollout and the honest next step is a second pilot with comparison stores. If there were, ask why they are not in the report.

If the question cannot be answered, the decision is bigger than a read. This is what the Decision Review returns for the same case.

Illustrative Decision Review

Roll out the AI clienteling tool, or not

A luxury house · Retail, CRM and clienteling teams · A vendor’s pilot report is positive and the budget round is in six weeks

What we found

  • The pilot report measures messages sent and reply rate. There are no comparison stores, so nothing shows what the same advisers would have achieved without the tool.
  • The suggestions draw on CRM preference fields that are empty or stale for most returning clients, so advisers rewrite most messages by hand.
  • Nobody has decided what an adviser may decline, or who can switch the feature off if clients complain.

Options compared

OptionTypeEffortRecommendation
Agree what advisers may decline, and name the person who can suspend the feature Process Low
Do now
Fix the preference and consent fields in the CRM you already have, for returning clients first Existing tool Low
Do now
Second pilot with comparison stores, a measure agreed first and a stop rule AI Medium
Pilot nextOnce the record is clean and the adviser’s authority is written down
Roll out to all stores on the current pilot report New software High
Not advisedThe report shows the tool was delivered. It does not show that it helped

Recommendation

Do not roll out on this report. Fix the record and the adviser’s authority first, then run a second pilot against comparison stores with the measure and the stop rule agreed before it starts.

Next-step brief

Owner
Head of clienteling
Change
Clean preference and consent fields for returning clients; a one-page rule on what advisers decline and who suspends the tool
Measure
Voluntary adviser use, unwanted-contact complaints and repeat appointments, against comparison stores over twelve weeks
Stop rule
If advisers stop using it by week eight, or complaints rise against the comparison stores, stop before the budget round
Support
The CRM owner and the vendor’s account lead
Illustrative Decision Review · a manufacturer

Put AI into product development, or fix the intake first

A knitwear manufacturer supplying premium and luxury brands · Product development and sampling · A vendor proposes an AI platform that drafts tech packs and knit programmes; one client house is asking for digital samples from next season. Invented; the figures are made up.

What we found

  • Sample rounds average 3.4 per style. The team’s own log shows most second and third rounds come from client spec changes arriving by email after the first proto, not from slow drafting.
  • The vendor’s case study reaches “delivered”: drafting time fell at a footwear company. It gives no figure for sample rounds, errors caught, or technician hours after correction, and has no knitwear reference.
  • The factory already holds a complete record for every style, yarn lot, gauge, programme, lab dip, wash recipe and measurement chart, in the technician’s files and a spreadsheet that no client system reads.
  • The client’s digital-sample request covers hand and drape, which do not render. A 3D knit sample would still need a physical one, and the cost of making both is on nobody’s business case.

Options compared

OptionTypeEffortRecommendation
One intake template for spec changes, agreed with the top three clients, and one owner for changes after the first proto Process Low
Do now
Move the style record into the ERP the factory already pays for, so technicians and clients read one version Existing tool Medium
Do now
Pilot AI drafting on twenty styles for one client, against twenty comparable styles, measuring sample rounds and hours after correction AI Medium
Pilot nextOnce the intake is fixed, so the pilot measures the tool and not the email thread
Subscribe to the platform for all clients on the vendor’s case study New software High
Not advisedThe case study shows drafting got faster at a footwear company. It does not show fewer samples anywhere

Recommendation

Do not subscribe on the case study. Fix the intake and the record first, then pilot AI drafting on one client’s styles against comparable styles, with the measure and the stop rule agreed before it starts. Answer the client’s digital-sample request with a cost per style and one line on what a render cannot show.

Next-step brief

Owner
Head of product development
Change
One intake template and one owner for spec changes; the style record into the ERP; a per-style price for digital samples sent to the client
Measure
Sample rounds per style and technician hours after the first proto, for the twenty piloted styles against twenty comparable ones, over one season
Stop rule
If rounds have not fallen by at least one per style by mid-season, or technicians stop using the drafts, stop before the renewal date
Support
The ERP owner, the vendor’s account lead and the client’s technical team

Wherever you are, start with a conversation.

Get in touch