Southeast Asian Retailers Are Buying Chinese AI Tools Without Understanding the Chinese Question That Built Them

I was sitting in a conference room somewhere in the region with the head of retail operations, two regional directors, and a technology lead who had just returned from a vendor roadshow in Shanghai.

The PowerPoint deck on the screen was impressive. Sensor technology. Behavioural analytics. AI-powered recommendation engines. Smart hangers that detect when a customer picks up a garment. Dashboards that track store traffic, try-on rates, conversion, and sales associate performance — all in real time, all the way down to individual product level.

The head of retail operations leaned forward. “This is exactly what we need,” she said. “When can we implement?”

Nobody in the room asked what problem they were solving.

I have sat in versions of this meeting more times than I can count. Different brands. Different markets. Different price points and store formats and organisational structures. The technology changes slightly. The vendor accent shifts. The enthusiasm is consistent. And the missing question is always the same.

Not: what can this tool do?

But: what question are we actually trying to answer?

That distinction between buying a tool and building from a question is precisely the gap between how most Southeast Asian fashion retailers are approaching AI and how China’s most sophisticated retail operations built the systems that are now being exported to this region.

And until that gap closes, the tools will keep arriving. The results will keep disappointing. And the question will keep going unasked.

(I write a fortnightly series on continuing education, capability development, AI, and the messy realities of organisational transformation. I’m also the author of Making S.E.N.S.E. of AI for SMEs. You can find his full archive on Medium.)

The Number That Stopped the Room

Let me share a number I encountered in a presentation by Austin Lee — named IDC’s Best CIO in China in 2024, former digital transformation lead for China’s largest chain retail group, and a pioneer of AI transformation across the fashion industry.

The number is 50.

Before digitisation with behavioural sensing technology, a single retail store generates approximately 50 POS transaction records, around 1,000 inventory entries, and 50 sales associate records per day. After digitisation, that same store generates data on 200 try-on actions, 1,000 pick-up events, 50,000 raw sensor readings, 500 display position coordinates, and the full movement and recommendation patterns of every sales associate on the floor.

Behavioural data — the data generated by customers touching, picking up, trying on, and putting back merchandise — is 50 times the volume of sales transaction data.

And here is the finding that should matter most to every retail leader in Southeast Asia: try-on rate shows a correlation coefficient of 0.9 with future sales performance. Two months before a product becomes a bestseller, the customers who pick it up and try it on are already telling you it will be.

The question is whether you are listening.

Most retailers in Southeast Asia are not. And the reason is not a technology problem. It is a question problem.

The Wrong Question Is Costing You Everything

According to a 2025 joint report by Lazada Group and Kantar, surveying 1,214 e-commerce sellers across six Southeast Asian markets: while 68% of sellers say they are familiar with AI, actual implementation across business operations averages just 37%. Sellers believe they have integrated AI into 47% of their processes — a notable gap between perceived and actual adoption.

The leading AI use case in Southeast Asian retail? Customer service chatbots, followed by marketing and advertising. These are not bad applications. But they are answers to a shallow question.

The question that built China’s most sophisticated retail AI systems is fundamentally different. It is not: what can AI do for us? It is: what does each individual store need to do differently this week — and what data would tell us that with certainty?

That distinction sounds subtle. It is not. It determines everything: what data you collect, what infrastructure you build, what your frontline staff are trained to do, and what your leadership team is held accountable for.

The AI system built for China’s largest chain retail group does not answer questions about the organisation. It answers questions about each store, each product category, each sales associate, and each square metre of shelf space — every week, automatically, and with enough specificity to generate actionable decisions before Monday morning.

Southeast Asian retailers are buying the tools. They are not yet asking the question.

What China Built And Why It Cannot Simply Be Imported

The framework is built on a deceptively simple insight: the three major costs of any retail store are merchandise, rent, and people. Everything else is a lever. And every lever can be measured, predicted, and optimised at the level of the individual store — not the region, not the brand, not the category average.

This produces three core metrics that sit at the centre of the system.

Merchandise efficiency measures sales minus inventory cost divided by the number of customer pick-up interactions — not sales volume alone, which tells you what happened, but pick-up rate as the leading indicator of what is about to happen.

Space efficiency measures sales output per display unit, minus the cost of that unit. Which products are earning their shelf space? Which display positions are generating traffic but not conversion?

Staff efficiency measures individual sales contribution minus individual cost, divided by customer interactions handled. Not a blunt productivity metric — a diagnostic. Where in the customer journey is each associate losing the sale?

Each metric generates a weekly decision. Not a quarterly review. Not an annual planning cycle. A decision, this week, that changes what happens in the store next week.

The results are documented. A store case study shows a product identified by high behavioural data but low initial sales — repositioned, given primary display placement, and supported by targeted sales associate coaching — moving from zero units sold to twelve units in four weeks. No price reduction. No promotional campaign. Just acting on the signal the customers were already generating.

In documented cases, products selected on the basis of behavioural data achieved 100% prediction accuracy for future sales growth, while products selected on sales data alone declined in the following week. Gross margin improved by 10% under equivalent investment conditions across stores implementing the full system.

As shared by Austin — this is not a pilot. The system was deployed across more than 977 stores in phase one alone, operating at national scale, weekly, with full closure of the data-to-decision-to-execution loop.

The McKinsey-BoF State of Fashion 2025 confirms that inventory remains one of the industry’s most persistent challenges globally — both excess stock and stockouts simultaneously impacting fashion brands. Fashion companies that have scaled AI adoption report 35 to 45% reductions in inventory costs and 20 to 30% improvements in sales forecasting accuracy.

China is achieving these numbers not in research papers but at scale, in thousands of stores, on a weekly cycle. Southeast Asia is not.

The Framework Problem

Here is where Southeast Asian retailers need to pay careful attention — because the lesson from China is not simply deploy more sensors or buy better analytics software.

The behavioural data system required a specific sequence of decisions that most Southeast Asian organisations have not yet made.

First, a decision about what to measure. Not sales outcomes — which are lagging indicators — but behavioural leading indicators: pick-up rates, try-on conversion, display position traffic, individual associate recommendation patterns.

Second, a decision about cadence. The system operates on a weekly cycle — every store, every week: review behavioural data, identify anomalies, generate strategic recommendations, assign tasks, track execution, measure results against the prior week. Nearly half of Southeast Asian companies have moved beyond AI pilots. Yet talent shortages, unclear ROI, and integration complexity remain the biggest structural barriers to scaling AI and delivering measurable impact. (Source: EDB Singapore/ McKinsey, “AI in Southeast Asia: An Era of Opportunity,” February 2026) The cadence problem is underappreciated. Most organisations are still thinking about AI transformation in quarters, not weeks.

Third, and most critically, a decision about ownership. The model operates across three organisational levels simultaneously: the individual store, the regional retail company, and group headquarters. Each level has specific data inputs, specific decision rights, and specific accountability for results. The technology serves a governance architecture. Without that architecture, the technology has nowhere to land.

A key challenge often absent from the discourse on AI adoption in Southeast Asia is determining specific use cases that actually add value. The excitement around AI’s potential is clear, but businesses need to ensure any implementation delivers tangible value, rather than just existing for the sake of it. (Source: TechWire Asia, March 2025)

The use case is not “AI for retail.” The use case is “pick-up rate as a two-month predictor of sales performance for a specific product in a specific store format.” That specificity is what produces the 0.9 correlation. That specificity is what the conference room missed.

The Dangerous Shortcut

In its 2026 budget, the Singapore government announced active co-funding of AI adoption among small- and medium-sized retailers through training support and introduced a 400% tax deduction on AI expenditures. Across the region, governments are pushing AI adoption with investment, incentives, and urgency.

This is well-intentioned. It is also accelerating a dangerous shortcut.

When adoption is incentivised without direction, organisations procure tools before they have answered the question those tools are supposed to serve. They build dashboards before they have decided what decisions those dashboards should support. They deploy recommendation engines before they have mapped the customer journey well enough to know where the recommendation belongs.

The most consequential AI applications in Southeast Asia are unfolding deeper in the operational stack — not in customer-facing personalisation and engagement. By 2026, the leaders in AI adoption will not be defined by a single location, but by their ability to conquer specific market challenges. (Source: Inside Retail Asia, February 2026)

The framework does not start with tools. It starts with a question about the store. What are the three major cost centres? What are the leading behavioural indicators that predict performance in each? What decisions need to be made weekly, and by whom, to move those indicators? And only then — what data infrastructure, what sensing technology, what analytical system would support those decisions?

That sequencing, question first, tool second, is the methodology that produced results in China. It is also the methodology that is almost entirely absent from Southeast Asian retail AI conversations, which tend to begin and end with the tool.

What Actually Needs to Change

Southeast Asia’s AI market is set to expand from $12.03 billion in 2026 to nearly $80 billion by 2031 — a 37.13% compound annual growth rate. (Source: Chatboq, “AI in Asia Statistics and Adoption Trends,” 2026) Offline retail still captures 80 to 85% of total fashion sales in the region, meaning the in-store behavioural data opportunity is enormous and largely uncaptured. (Source: Sellercraft, “Southeast Asia Online Retail Outlook 2025–2026”)

The investment is arriving. The tools are available. The vendors are in the conference rooms. And the question is still not being asked.

Southeast Asian retailers need to stop asking which AI tools China is using and start asking what question Chinese retail AI was built to answer.

The answer — what does each individual store need to do differently this week, and what data would tell us that with certainty — is not a Chinese answer. It is a retail answer. It applies here as precisely as it applies in Suzhou. The specific products, the specific customer behaviours, the specific seasonal patterns differ. The question does not.

By 2026, 78% of Asian workers are using AI tools weekly, surpassing the global average of 72%. The region is not behind on adoption. It is behind on direction.

The behavioural data is already being generated in every physical fashion retail store in Southeast Asia. Every customer who picks up a garment and puts it back is producing a signal. Every display position that attracts traffic but not purchase is generating a question. Every sales associate who converts at a different rate from her colleague in the next store is revealing something about the gap between what the organisation assumes and what the customer actually needs.

Most of that data is currently invisible. Not because the technology to capture it does not exist — it does, and it is increasingly accessible to mid-sized retailers across the region. But because no one has decided that the question it answers is worth asking.

Back to That Conference Room

After the head of retail operations asked when they could implement, the vendor started to speak.

I asked a question first.

“Before we talk about implementation — what decision are you trying to make that you cannot currently make? And what data would you need to make it with confidence?”

The room went quiet. Not because the question was difficult. Because nobody had been asked it before. That silence is the real gap between Southeast Asian retail and what China has built. Not the sensors. Not the dashboards. Not the AI models trained on millions of try-on events.

The technology to act on the answer, once you have it, already exists. The question is whether you are willing to ask it.

That decision is not a technology decision. It is a leadership decision. And it is the one that China made first.