Boring AI wins: here's why the retailers winning in this space are doing deeply unsexy work

Retail does not have an AI problem. It has a heroics problem, says Vineta Bajaj, Group CFO at Holland & Barrett, and a member of the RTIH AI in Retail Awards judging panel.

The industry has normalised firefighting. Being “always on”. Running on adrenaline. Celebrating teams who save the day when things go wrong. That culture is often framed as unavoidable. Retail is complex. Retail is fast. Retail is unpredictable.

But that story is convenient. And wrong. Retail is predictable most of the time. Roughly 80% of retail decisions repeat. Demand follows patterns. Replenishment cycles are known. Pricing mechanics are understood. Labour needs are forecastable. The same decisions are made every day, often by different people, with slightly different judgement, under pressure.

The real problem is that retailers treat the predictable 80% as if it were exceptional. They force humans to re-decide routine things. They rely on heroics instead of systems. And then they wonder why teams are exhausted and AI never scales. This is where boring AI wins.

Boring AI is embedded, not performed

The fastest way to kill AI in retail is to turn it into performance art. Dashboards. Scenarios. Workshops. Endless “decision support”. All of it looks sophisticated. None of it removes work.

In replenishment, boring AI means order proposals, not forecasts. In pricing, recommended price moves within clear guardrails, not scenario trees. In labour, stable shift plans, not prettier dashboards. This distinction matters.

Scenarios feel safe because they keep humans in control. In reality, they expose the flaw in how retail adopts AI. Humans are still being asked to make routine, repeatable decisions. Boring AI should make the boring decisions happen.

If a pricing move sits within agreed guardrails, the machine should execute it. No meeting. No approval. No dashboard review. Humans should only see exceptions. Cases that fall outside tolerance. Cases that genuinely require judgement. The machine under the hood should hum. Quietly. Reliably. Constantly. If AI’s primary output is a dashboard asking a human what to do, the value has already leaked out.

Vineta Bajaj: The fastest way to kill AI in retail is to turn it into performance art. Dashboards. Scenarios. Workshops. Endless “decision support”. All of it looks sophisticated. None of it removes work.

Heroics are a symptom, not a strength

Retail teams often take pride in heroics. Fixing problems at the last minute. Saving promotions. Chasing stock. Reworking labour plans. Pulling all-nighters to close gaps. But heroics are not a sign of excellence. They are a sign of system failure.

Every heroic intervention usually points to a decision that could have been automated earlier. A pattern that was known. A rule that was not enforced. Data that was not trusted. Boring AI changes the operating model.

It normalises the predictable. It absorbs routine variability. It makes “nothing happened today” the goal. Humans then focus on the genuinely exceptional 20%. Supplier failures. Weather shocks. Demand spikes. Strategic trade-offs.

Data discipline is the unavoidable foundation

There is an old phrase. “Shit in, shit out.” What we are seeing now is that principle playing out at industrial scale. Everyone is racing to adopt AI. Very few are clear on the problem they are trying to solve. Even fewer have the data discipline required to solve it well.

AI does not fix ambiguity. It amplifies it. Product hierarchies change mid-year. Promotions are coded differently by team or country. Substitution logic is unclear. Ownership of data is vague. Definitions drift. No model can compensate for this.

The retailers winning with AI invest heavily in data discipline. Clear ownership. Boring definitions. Master data that does not change unless it absolutely has to. Promo mechanics that mean the same thing everywhere. This work is slow. Political. Unsexy.

It is also where AI either becomes boring and valuable, or exciting and useless. GenAI layered on messy data does not create intelligence. It creates confident nonsense, faster.

Value lives in exceptions, not averages

Another retail myth is that improvement comes from optimising the middle. It doesn’t. A small percentage of SKUs drive most stock-outs. A handful of stores create disproportionate waste. A few promotions destroy margin assumptions. Averages hide all of this.

Boring AI focuses on exceptions. What changed. What broke. What needs attention now. AI does prioritisation. Humans apply judgement. This is how heroics disappear. Teams stop reacting to everything and focus on what actually matters.

Ownership beats innovation every time

AI does not fail because it is technically hard. It fails because nobody owns the outcome. Innovation teams do not run P&Ls. Data teams do not own margin. IT does not carry working capital targets. When AI sits outside the line, it stays in pilot mode forever.

Where AI works, ownership is explicit. Pricing AI is owned by Commercial. Finance defines guardrails, margin floors, and economic logic. AI proposes price moves automatically. Humans only see exceptions.

This only works if definitions are clear. Data is clean. Guardrails are stable. Ownership without discipline is theatre. Discipline without ownership is bureaucracy. You need both. And yes, this means fewer hero moments. That is the point.

Boring AI compounds. Heroics do not.

The most damaging question leaders ask about AI is “what’s the ROI this quarter?” The AI that creates durable advantage delivers small gains that compound. Slightly better forecast bias. Fewer emergency orders. Lower waste. More stable labour plans.

Week to week, it feels underwhelming. Year to year, it is material. Heroics feel good in the moment. They do not compound. They exhaust people and hide structural issues. From a CFO perspective, the choice is obvious. Compounding beats adrenaline. Every time.

From tools to a retail brain

The winners are not building use cases. They are building systems. Demand feeds supply. Supply informs pricing. Pricing reshapes demand. Execution feeds learning. Decisions connect. That is the retail brain. Not a monolithic platform. A set of boring, reliable intelligence loops embedded across the business. Tools optimise tasks. Brains remove heroics.

The uncomfortable conclusion

Retail AI does not fail because the technology is immature. It fails because retailers cling to heroics. The retailers winning with AI are doing deeply unsexy work. Removing noise. Enforcing discipline. Automating the predictable. Letting machines handle routine decisions. They are making retail boring again. And that is exactly why they win.