For decades, retail forecasting has been built on averages: national trends. Seasonal patterns, historical sales data. These tools weren’t wrong; they were the best available. But in a commerce environment where a supply chain disruption, a tariff change, or a viral product moment can invalidate last quarter’s model within days, averages aren’t just insufficient. They actively mislead.
The question isn’t whether to replace them. It’s what replaces them, and whether retailers are ready to act on the answer.
The data problem underneath the AI opportunity
Most discussions about AI in retail jump straight to the outputs: smarter pricing, better recommendations, faster fulfilment. What they skip is the precondition that determines whether any of that works: the quality and structure of the underlying product data.
This isn’t a comfortable point to make in a marketing context, but it’s the honest one. We’ve seen it directly at Rithum, working across hundreds of retailers and brands at scale. The bottleneck is almost never the algorithm. It’s the data the algorithm is trying to work with including inconsistent taxonomies, missing attributes, product descriptions that mean different things across different channels. AI applied to poor data doesn’t produce intelligent decisions. It produces confident mistakes, faster.
Before asking what AI can predict, the prior question is: are your products machine-readable? Do your SKUs have the attributes, categories, and enrichment that allow a model to reason about them accurately? If the answer is “mostly” or “we think so”, that gap is where forecasting value leaks.
What good demand sensing looks like
When the data foundation is right, the picture changes substantially. AI can move forecasting from static, historical cycles to live, multi-dimensional models that integrate real-time channel signals, return patterns, and fulfilment costs. Not just to predict volume, but to predict margin.
The distinction matters. Volume prediction tells you what will sell. Margin prediction tells you whether selling it is worth it accounting for where it’s shipping from, what the return risk looks like in that region, and what the cost of a stockout versus an overstock would be. These are different calculations, and aggregated averages make them invisible.
Hyper-local demand signals take this further. Rather than treating a region as a uniform market, AI can identify demand patterns at the level of individual store catchments or fulfilment zones and route inventory and advertising accordingly. A product trending in one postal region based on seasonality, local competition, and recent search behaviour tells a different story than the national average. Acting on that difference is where margin is won or lost.
Returns as a data signal, not a logistics problem
One of the most underused signals in retail is returns. Conventionally treated as a cost to manage, returns are a diagnostic. They tell you, at scale, where the gap between product expectation and product reality is widest.
Our own research found that 18% of consumers cite improving the accuracy of product details like pricing, availability and specifications, as the single most important area for improvement in their online shopping experience. That’s a data quality metric you can help control.
When AI analyses return patterns alongside product attributes and customer feedback, specific failure modes become visible. A product that consistently underperforms in one region due to sizing discrepancies. A category where image quality consistently creates expectation mismatches. These aren’t complex fixes. Often a revised title, an additional
attribute, an updated image is enough. But the impact at scale is material. Reducing return rates by even a percentage point or two translates directly into retained margin across thousands of SKUs.
The closed loop between prediction and execution
The real value of AI in commerce isn’t any single capability. It’s the feedback loop that connects demand prediction, merchandising decisions, retail media, and fulfilment into a system that learns.
When data flows continuously across channels, when advertising responds to predicted local demand spikes, when inventory routing reflects live fulfilment constraints, when return patterns feed back into product content, decisions compound rather than degrade. The more accurately you can predict what will sell where and at what margin, the more efficiently every downstream decision can be made.
The retailers who will benefit most from this are the ones who invest in the data foundation that makes AI useful: clean, consistent, trusted product data across every channel and every algorithm that touches it. That’s not a technology problem. It’s an operational commitment. But it’s the one that determines whether everything else delivers.
