The same use case, built by several brands
Each brand has its own assistant to answer the same customer questions. The group pays several times for the same function.
Sector · Retail and luxury
Your retail AI runs brand by brand. Who knows what it returns?
AI in retail and luxury has settled in brand by brand: shopping assistants, tools for in-store advisors, product content, copilots at head office. Each project came with its own objective, technology and budget. Leadership sees the total cost; the value of the whole is measured nowhere. We measure it from the logs your platforms already produce and from interviews with your leadership teams, brand by brand, then for the group.
- 40+AI diagnostics delivered in a year
- 41,200users measured
- 218qualified use cases
- 20 to 40%of paid AI seats unused across the estates analysed (an engagement finding, not a study)
Across all its engagements, Idun Group has carried out more than forty AI diagnostics in a year and measured the usage of 41,200 users. Across the estates analysed, 20 to 40% of paid AI seats go unused: an engagement finding, not a study.
Company markers, across all our engagements: none is attributed to a client, none is a result guaranteed in advance.
In short
In retail and luxury, AI serves customer relations first (shopping assistants, conversational commerce), then sales advisors, product content and head-office functions. It was rolled out brand by brand, rarely with a common measure. The audit of real usage by Idun Group, an AI consulting firm, reads the usage logs these tools already produce, holds interviews with leadership teams, then says what to industrialise, consolidate or cut.
Where does AI stand in retail and luxury?
Across all sectors, of the AI-equipped companies Idun Group has met, 93 said how they measure their AI usage: 39 have no measurement at all, and in large companies (5,000 employees and more), 18 of 29 measure nothing. AI in French retail lags behind Europe: in 2025, 9.9% of the sector's companies with ten or more employees used at least one AI technology, against 15.5% across the European Union, according to Eurostat. AI in luxury has become a priority, rarely a result: according to the Comité Colbert, the association of French luxury houses, in its "Luxe et technologie" report (June 2026), 22% of houses rank it among their top three priorities, up from 5% in 2024, but fewer than 20% of luxury executives say they have seen a significant impact from it.
Large groups, though, have already equipped their customers, their advisors, their marketing and their head office. In luxury, according to the Comité Colbert, head office is where AI is advancing fastest: its penetration in the houses' support functions has risen fivefold, from 6% to 31%, against 21% in customer-facing functions (16% in 2024).
In a retailer or a luxury house, the AI teams use covers six areas: customer assistants on the website and in conversational commerce, tools for sales advisors and beauty advisors, product content and brand voice, head-office copilots, research and product development, and coding assistance for e-commerce teams. Demand forecasting, replenishment, pricing, markdowns, logistics, in-store computer vision and creative work also fall under artificial intelligence in retail, but they are not our trade: Idun Group does not touch them.
Each brand measured its project with its own indicators, when it measured it at all. The overall view is missing almost everywhere: nobody knows what the brands return taken together, or which uses duplicate one another from one brand to the next.
Why a retailer or a luxury house calls us
A retailer calls Idun Group when its AI has piled up with no overall view: each brand and each department launched its projects with its own objective, technology and budget. The problem is never a lack of AI. Six situations come up again and again.
An assistant error is a brand incident
Elsewhere, a wrong answer is a quality incident. In front of a customer, a wrong recommendation puts the house's image on the line, immediately and in public.
Head-office seats nobody can account for
Marketing, procurement or finance, at head office and in the brands, have been given Microsoft 365 Copilot, ChatGPT or Claude seats. The spend is committed and recurring; the vendor's console shows who logs in, not for which tasks or with what result.
Content rework time, never counted
Product sheets, campaign variants per market and per language: teams turn out volumes they never reached before. The time spent reworking what the model produces appears nowhere.
Proof before extending to the network
Rolling out an advisor assistant to every store is expensive. The pilot convinced people, but nobody can say whether advisors still use it after the first few weeks. Leadership wants proof of usage before signing.
An advisor tool the advisors do not open
Advisor support is the most advanced use of AI in luxury, according to the Comité Colbert. When advisors doubt the tool, they stop opening it, and a team's drop in usage only shows up at the budget review.
Our references in the sector
In retail and luxury, Idun Group has worked for LVMH and for L'Oréal. Idun Group is an independent consulting firm with its own tooling: it measures with Idun Cockpit, its own tool, then says what to do, without reselling any platform or selling agent building.
At LVMH, for all the perfume and cosmetics brands, an Idun engagement on their generative AI platform in production.
At L'Oréal, the deployment and governance of usages in the group's Microsoft ecosystem.
AI use cases in a retailer, from store to head office
In retail, the ways teams use AI fall into six areas: customer assistants, sales advisors, product content and brand voice, head-office functions, research and product development, and e-commerce teams. For each one, Idun Group gives what measuring usage reveals there and the decision it makes possible.
| Use | What it is | What measurement reveals | The decision |
|---|---|---|---|
| Customer assistants and conversational commerce | Product recommendation, choice guidance, gift advice, questions before and after purchase, on the website and in messaging apps. | How relevant the answers are in production, the conversations that stop, and unserved demand: the questions the assistant cannot answer. Those hidden needs shape the next version better than an internal workshop. | Make it reliable before extending it. This is the most exposed area: here the assistant answers the customer on behalf of the house. |
| Sales advisors and beauty advisors | Client follow-up, appointment preparation and reminders, what the trade calls clienteling, as well as product knowledge and personalised advice. It is the most advanced use in luxury: according to the Comité Colbert's "Luxe et technologie" report (June 2026), 9% of houses have deployed it at scale with a measurable impact, and 22% are rolling it out with first results. | Real adoption per team and per brand, usage that holds or falls away after launch, the questions asked most often. | Extend what proves its worth, target training where adoption drops. Active users, far more than declared satisfaction, decide the rollout. |
| Product content and brand voice | Multilingual product sheets, campaign variants per channel and per market, copy in the house tone. | The rewrites requested from the tool before a text is kept, the ones that recur and signal a badly tuned tool, and duplicates between brands of the same group. The manual rework done afterwards leaves no trace: the interviews put a figure on it. | Consolidate onto a shared base rather than maintain one tool per brand, and automate only what deserves it. |
| Head-office functions | Generative AI copilots and assistants, such as Microsoft 365 Copilot, ChatGPT or Claude, opened to support teams at head office and in the brands (marketing, procurement, finance, HR), often seat by seat and with no defined scope. | The gap between paid seats and used seats, from one brand to the next, heavy usage nobody scoped, and what teams go looking for in other tools because they cannot find it in the group's. | Industrialise the use cases that pay off, adjust the number of seats at each renewal, choose which teams to train and on what. |
| Research and product development | State of the art, document exploration, fact-checking, especially in beauty. | Usage against allocated seats, and the internal sources teams stitch together by hand for lack of a connector. | Connect internal sources before changing tools. |
| E-commerce teams | Coding assistance for the teams that build and maintain the websites, apps and purchase journeys. | Adoption per team and the seats that sit idle, on a tool that affects how fast changes go live. | Right-size seats at renewal, spread the practices of the teams that ship fastest. |
Measuring the ROI of AI in retail, brand by brand
The return on investment (ROI) of a retailer's AI is measured by combining the usage logs of its platforms with interviews with its leadership teams, brand by brand, then at group level. Idun Group sets the cost of each use case against its observed usage: advisors who actually open their assistant, shopping conversations that find their answer or stop, rewrites requested from the content tool. Idun Group carries out the audit of real usage within 15 days, with at least twelve interviews.
- 01
What your platforms already record
The analysis starts from the logs and exports of the platforms you administer: Microsoft 365 Copilot, ChatGPT Enterprise, Claude, Azure OpenAI, Dust, custom assistants and agents, and, for your vendors' tools (shopping assistant, advisor tool, content tool), the exports their contract gives you. Nothing is installed on workstations: no instrumentation of your platforms, not a line of code on your side. Your teams hand over the exports or open a read-only connection; the scoping meeting decides which.
- 02
Interviews say what the logs leave out
A log says which teams and which brands use which tool, how often and at what cost, and which assistant conversations fail or get cut short. It does not say what the team was trying to do. Interviews with IT, digital and the business units are a mandatory step: interviews and data, never one in place of the other.
- 03
Brand by brand, then consolidated
The exercise runs brand by brand, then consolidates at group level. Duplicates only appear at that level, and closing them brings the quickest savings.
- 04
A read-out you can take to a committee
The document needs no reworking: cost per use case for finance, detail per tool and per brand for technical teams, training needs for HR.
The measurement loop
The three deliverables of the audit
Idun Group's audit of real usage returns three deliverables: the map, what teams do with AI, brand by brand; the account, what each use case costs and what it returns; the course, what to industrialise, consolidate or cut, and which teams to train.
The map: usage
For each brand, the detected use cases, real adoption per team and per department, and the inventory of assistants and instructions teams have created, shared or kept to themselves.
The account: ROI
For each use case, its full cost, often split across the brands, digital and IT, set against what it returns: observed usage against paid seats, and the time given back, converted into cost at a rate set with you from your job profiles.
The course: strategy
The order of decisions: what to industrialise first, what to consolidate across brands, what to cut, which teams to train and on what. Then what falls to your teams: publishing the assets kept private, and plugging AI into the tools where advisors and head office already work.
From pilot to network, three outcomes
An assistant tested in a few stores only goes out to the whole network on usage signals, not on the impression the pilot left: advisors who keep opening it over time, conversations that find their answer. Once usage is measured, the decision comes down to three outcomes: industrialise, consolidate or cut. Cutting is not a failure: at renewal, the budget moves to a better-measured use case.
| Outcome | What usage shows | What follows |
|---|---|---|
| Industrialise | Broad, repeated usage that holds after the first few weeks, conversations that find their answer | Extend to the network or to the other brands, with a dedicated budget |
| Consolidate | The same use case built by several brands or on several tools | Keep one implementation for the whole group, close the others |
| Cut | Marginal usage, seats or conversations paid for with no usage to match, rewrites so frequent that the tool saves no time | Terminate, or move the budget to a better-measured use case |
What the analysis reads, and what it does not
Idun Group analyses usage, never people: results are aggregated by team, by brand and by tool. The analysis is read-only, with nothing installed on workstations. It does not read your customer databases; the conversations of a customer assistant, however, do contain your customers' messages, and how they are processed is agreed with your data protection officer (DPO). Processing is hosted in France, at Scaleway in Paris, or on your premises.
What the data shows
- aggregated usage, by team, by brand and by tool
- cost set against observed usage
- customer-assistant conversations that get their answer, stop along the way or go unanswered
- the rewrites requested from content tools
- unserved demand: what customers and teams look for without finding it
What it does not show
- the activity of a named employee
- your customer databases
- demand forecasting, pricing, replenishment
- what the exports do not contain: the report says so in its appendix of unknowns
The kind of situation the analysis handles
At a luxury house that is an Idun Group client, the customer service assistant had been designed without gift sets, which had been left out of its scope: the customers' conversations are what revealed this demand. It is the first of the three situations in this section, and the only real one; the other two are illustrative and describe no client.
The gift sets the assistant could not handle
The assistant's scope stopped short of something customers were asking for; only their conversations showed it.
"Three brands, three assistants, one question."
In one group, three brands each built an assistant to answer the same questions about delivery and returns. Each brand judges its assistant a success by its own indicators. Side by side, the logs show three costs for the same service, and answers that do not say the same thing.
"Product sheets written fast, slow to rework."
A content team writes its product sheets with an assistant and turns out far more than before. Nobody was counting the successive rewrites requested from the assistant before a text was kept; the logs show them. The time saved writing is measured alongside the time spent correcting.
The shape an estate takes
One measured case, with its figures
Consulting and training firm
At a consulting and training firm, the analysis of real Claude usage established 85% real activation across around forty seats, false starts included. With the seats that left no trace at all, 31% of seats, nearly one in three, were to be reassigned or activated. At the firm, the three heaviest users' consumption is worth, at API prices, 12 to 25 times the price of a premium seat.
The limit The limit, written at the top and never in an appendix: the platform export had a rolling thirty-day retention while the rollout had been running for five months. One month out of five was measured.
The firm is not a retailer. The same reading applies to a retailer's estate: how many paid seats are active, how many leave no trace.
What retailers and luxury houses ask us
What are the uses of AI in retail?
In retail and luxury, the AI teams use covers six areas: customer assistants and conversational commerce, sales-advisor tools, product content and brand voice, head-office copilots, research and product development, and e-commerce teams. Demand forecasting, pricing and logistics have their own AI too; Idun Group does not work on them.
Where does AI stand in luxury?
According to the Comité Colbert's "Luxe et technologie" report (June 2026), 22% of houses rank AI among their top three priorities, up from 5% in 2024. The most advanced use is support for sales advisors, but only 9% of houses have deployed it at scale with a measurable impact, and fewer than 20% of luxury executives say they have seen a significant impact. These are self-reported survey answers; Idun Group reads real usage in the logs of each house's tools.
How do you measure the ROI of AI in retail?
The ROI of AI in retail is read first in observed usage, not only in the time savings teams declare or in a vendor's statistics. Idun Group combines usage logs with interviews, and each finding carries its source: "measured" when a log establishes it, "declared" when it comes from an interview. We then set out, for each use case and brand by brand, its full cost, often split across the brands, digital and IT, and what it returns: observed usage against paid seats, and the time given back, converted into cost at a rate set with you. Sales and conversion are read in your commerce tools: Idun Group does not link them to conversations. No gain is quantified in advance: it is measured in your organisation.
Do our sales advisors really use their AI assistant?
In luxury, support for sales advisors is the most advanced use of AI, according to the Comité Colbert. Real usage of an advisor assistant is read in the tool's logs, by team and by brand: the share of equipped advisors who open it each week, the usage that holds after launch, the questions asked most often. Interviews explain why a team drops off.
Should we extend a pilot assistant to the whole network?
A pilot assistant only goes out to the whole network on usage signals: advisors who keep opening it over time, conversations that find their answer. Without those signals, you fix the assistant or cut it, and the extension waits. When several brands in the group have built the same use case, you keep only one implementation.
How does the exercise run across a multi-brand group?
In a multi-brand group, AI usage is measured brand by brand, then at group level. The exact scope is set during discovery, before any commitment. Consolidation is what brings out the duplicates between brands.
Do we have to give you access to our customer data?
Idun Group does not ask for access to your customer databases. We read the usage logs of your AI tools, conversations included: when an assistant talks with your customers, those logs contain their messages and, sometimes, health data (an allergy, a pregnancy). When the analysis detects such data, it stops at the finding. Nothing needs to be installed, no access is requested before signature, and the exact status of the processing is validated with your DPO.
Our vendors already provide dashboards. What do you add?
A vendor's dashboard stops at its own tool, and does not put two brands side by side. Idun Group brings the group's tools and brands together in a single reading: the same use case can be compared from one tool or one brand to another, and interviews explain the gaps.
Do you know what your AI estate already returns?
Thirty minutes to choose the brand or the function to start with, and the platforms to read. At the end of the half hour, we tell you whether the exercise is worth doing in your organisation.
Read-only. We analyse usage, never people.
Supported by
