From Selling Shelves to Selling Data: How AI Turned a Loss-Making Business into a Subscription Model
Issue 016 · Published 6 October 2026 · The AI Executive Briefing
IN ONE MINUTE
Most companies use AI to do the same things more cheaply. The bigger prize is changing what you sell. PwC's 2026 survey of 4,454 CEOs found that only 12% have seen AI deliver both cost and revenue benefits, while 56% report no significant financial benefit yet. In 2017–18, I led the turnaround of a loss-making display-fixtures business by changing its business model: from one-off hardware sales to a subscription data service built on computer vision and AR. It worked because it created value for all three parties in the chain: the brand owners, their consumers and us.
THE OBJECTIVE
Most AI programmes start with a cost line: which tasks can we automate, how many hours can we save? That is a reasonable start, but it is rarely where the largest value sits.
The question I find more useful is different. What does each party along your value chain need that nobody is providing today, and could AI let you provide it?
This issue walks through a real case, a simple test for spotting the opportunity in your own business, and what it means for each member of the leadership team.
WHAT THE RESEARCH SAYS
PwC's 2026 Global CEO Survey, of 4,454 CEOs in 95 countries and territories, shows how uneven AI's returns still are. Only 12% of CEOs say AI has delivered both cost and revenue benefits. A further 33% report gains in one or the other, and 56% say they have seen no significant financial benefit to date.
At the same time, business models are moving. In the same survey, 42% of CEOs say their company has begun competing in new sectors over the past five years.
Taken together, the message is clear. Efficiency gains matter, but they are finite. The companies that capture AI's full value use it to create new revenue: new services, new customers and new ways of being paid. The barrier is rarely the technology. It is seeing your own business differently.
I learned that lesson before generative AI, with a display shelf.
THE CASE: FROM SHELVES TO DATA
The background
Earlier in my career, I managed a global luxury division at Li & Fung, one of the world's largest supply chain management companies. It spanned seven verticals: colour cosmetics, fragrance, personal care, wine and spirits, packaging, retail management, and retail display fixtures.
The fixtures business, with revenue of more than US$40 million, sat in the middle of a simple value chain:
- Li & Fung, the manufacturer: we designed and built display stands and shelves for top-tier colour cosmetics brands, each carrying the brand's logo and fitted with containers tailored to its products.
- The brand owner: installed the fixtures at its counters in retail stores.
- The consumer: visited the counter and tried the products with the help of a sales assistant, using shared testers.
The fixture stood exactly where the buying decision was made. Yet our part of the chain ended the day it was delivered: we were paid once, for hardware, and saw nothing of what happened at the counter afterwards.
The problem
For years, the business had been unprofitable. The executive consensus was to keep it as a loss leader: a way to offer brand owners a one-stop shop. The team, meanwhile, was proud of its success two decades earlier and wanted to stay a pure manufacturing workshop, building shelves exactly to clients' drawings.
The real bottleneck was not factory capacity. It was mindset.
At the same time, consumer behaviour was shifting. Rising hygiene expectations meant shoppers hesitated to apply shared testers to their faces. That was a problem for every brand we served, because in colour cosmetics, trying a product is often what sells it.
The solution
Cutting costs or pushing the old model harder would not have fixed either problem. So we changed the destination: from selling display hardware to providing high-value data services to brand owners, on a subscription basis.
We launched the Magic Mirror project, building cameras, computer vision and augmented reality into the display fixtures themselves.
- For consumers: AR let shoppers "try on" cosmetics virtually, instantly and hygienically, without touching a tester.
- For brand owners: the mirror measured store footfall, dwell time, consumer attention and conversion, turning a shelf into a source of insight.
We then turned the try-on into a moment shoppers wanted to keep. Each consumer could take away an instant printed photo of their virtual look, a product sample, and a QR code linking to a digital copy of the photo to share with friends and on social media. In return, shoppers who opted in shared their store check-in and product preferences. That simple value exchange deepened every interaction between the consumer and the brand, and gave brand owners something a shelf had never offered: a direct way to strengthen their CRM funnel at the point of sale.
The people side was handled as deliberately as the technology. New technology always raises the same question: will this replace my job? You cannot inspire people who fear they are building their replacement. So we spoke openly with our craftsmen:
- Their manufacturing skills remained vital. They still built the same high-quality fixtures.
- We explained why the business had to change, and how technology would upgrade their work rather than replace it.
- A separate team added the "eyes and brain", the cameras and sensors, to the structures the craftsmen built.
The craftsmen were not asked to become technologists. They were asked to build the foundation of a more valuable product, and they could see their work in it.
The results: win, win, win
We deployed the first 20 Magic Mirror units across seven major airport duty-free stores in early 2018.
| For | What changed |
|---|---|
| Brand owners | In the stores with the mirrors, footfall rose 80% within two weeks of installation, dwell time rose 50%, and conversion rose 120%, compared with before. Brands also gained data on consumer attention they had never had, opted-in consumer profiles with store check-ins and product preferences to feed their CRM, and organic reach as shoppers shared their photos on social media. |
| Consumers | A hygienic, instant and more engaging way to try products, with no shared testers, plus personalised gifts: a photo of their look and a sample to take home. |
| Us, the fixtures supplier | Gross margin rose from 15% to 42%, the highest gross profit the business had earned, driven by the new service. A long-term loss-making unit became a profitable subscription business: the retail audit service became a new subscription line, generating stable annual recurring revenue (ARR). With our data and analytics platform integrated with brand owners' marketing insight teams, customers also had stronger reasons to stay with us. |
The value of the idea grew further during Covid-19. With lockdowns, colour cosmetics sales fell sharply worldwide, and few consumers were willing to put a shared tester on their face. Virtual try-on worked exceptionally well in that market, because it solved a real consumer pain point even in the toughest conditions.
THE FRAMEWORK: THE WIN-WIN-WIN TEST
The Magic Mirror worked because it passed four tests. They apply to almost any business looking to use AI for growth rather than cost alone.
| Test | Signals to look for | In the Magic Mirror case |
|---|---|---|
| 1. Is this the moment to change the model? | End consumers' expectations are rising; your customers want more engagement and insight than your product gives them; your core product is losing value and needs to move up the chain | Shoppers were wary of shared testers; brands wanted more engagement and retail statistics; our fixtures business was loss-making |
| 2. What needs are unmet along the value chain? | Requirements that no party in the chain is meeting today, at any level: consumer, customer or supplier | Consumers wanted to try products safely; brands wanted to know their consumers and stand out; we needed to escape one-off hardware economics |
| 3. Does everyone in the chain gain? | A clear gain for you, your customer and their customer, with no party paying for another's gain | Consumers gained a hygienic, engaging digital experience and personalised gifts; brands gained consumer insight and a stand-out brand experience; our upgraded product earned a higher gross profit than ever, driven by the new service |
| 4. Can we be paid for the outcome, repeatedly? | Recurring revenue tied to the value created, not a one-off sale | Yes: one-off hardware sales became a subscription retail audit service, generating stable ARR, and integration with brands' marketing insight teams deepened customer loyalty |
The tests build on each other. The first tells you when to act, the second where the opportunity lies, and the last two whether it will last. Value that flows to only one party is a short-term deal, not a business model.
WHAT THIS MEANS FOR YOUR LEADERSHIP TEAM
For CEOs and boards: don't let a loss leader become an identity
The most important decision in this case was refusing to accept "loss leader" as permanent. Every portfolio has units kept for strategic reasons rather than returns. Ask the four questions of each one. Some will pass, and the answer to an underperforming business may be a different business model rather than a cost programme or an exit.
For CFOs and P&L owners: change the revenue model, not just the cost base
The financial shift here was structural: from one-off hardware revenue to recurring subscription revenue, and from a 15% to a 42% gross margin. The subscription did more than add ARR. Once our analytics platform was integrated with brand owners' marketing insight teams, it became part of how they ran their business, which made each customer relationship more valuable and more durable. Treat AI business cases the same way. Model the new revenue line, its margin and its recurrence, not just the hours saved. And agree outcome measures with customers before launch. A pilot of 20 units was enough to prove the case.
For CMOs and commercial leaders: design a value exchange, not a data grab
Consumers shared their data because they received something they valued: a hygienic try-on, a photo of their look and a sample to take home. That is the basis of any sustainable first-party data strategy. Make the exchange explicit, make consent a genuine choice, and connect the data to the CRM so every interaction builds the customer relationship.
For CIOs, CDOs and technology leaders: put the intelligence where the decision happens
The technology was not exotic: cameras, computer vision and AR built into an existing physical product. The value came from placing it at the point of decision and connecting it to the brands' systems. Start with the decision you want to improve, then choose the technology. Build privacy in from the first design, not after launch.
And for every leader: a new business model only works if your people come with you. That is the subject of next week's issue.
Questions leaders ask
How can AI create new revenue, not just cut costs?
Map the unmet needs along your value chain, from your customers to their customers. Where AI lets you meet a need that nobody else is meeting, there is usually a new service your customers will pay for, and it is often worth more to them than the product itself.
How do you know if a business model idea is worth piloting?
Apply the four tests. If the timing signals are there, the need is real, every party gains and you can be paid for the outcome repeatedly, run a small pilot with measures agreed in advance.
What makes a data-driven business model sustainable?
A fair value exchange. People share data when they receive something they value in return, and when the choice is genuinely theirs. Without that, the model will not survive regulation or public trust.
THE LAST WORD
AI can make your current business cheaper. It can also make it different. The second is where the value multiplies.
Which needs along your value chain are still unmet today, and who will meet them first: you, or someone else?
—
Until next week,
Artificial Ian
→ Found this useful? Forward it to a leader with a business that needs a new model, not just a new budget.
→ Subscribe to The AI Executive Briefing, weekly in your LinkedIn inbox.
Sources: PwC, 29th Annual Global CEO Survey (4,454 CEOs, 95 countries and territories, fieldwork 30 September–10 November 2025, published January 2026). Magic Mirror results: the author's own programme data, 2018.