AI: We Put It To The Test

In a few of our previous blogs – see here and here – we’ve discussed how our entire team has been evaluating how AI can be used to help our client fashion brands.

And as we’ve also discussed, we’ve been looking at both the pros and cons because it’s not all positive. Indeed, AI can hallucinate; AI models can get withdrawn; AI costs are heavily subsidised; and new open source, low costs models are coming into play. So it’s important that we move forwards carefully.

But there comes a point when you have to stop talking about AI and actually put it to the test.

And here’s what we did: We took some real sales data from a UK womenswear brand, across a predefined period, covering styles, colours, sizes and sales locations. We then instructed AI to analyse the data and tease out any sales patterns, anomalies and trends that we might otherwise have missed.

In short, could it tell us which styles were selling best, which colours were strongest, how buying patterns differed by region, and what effect the weather was having on different product categories? And was there anything we could have otherwise missed?

The Result of our Test

We took the sales data from June 2025 and compared it to June 2026, with the weather layered on top to see whether that impacted or predicted demand. The model we used? Open AI GPT-5.x.

Here’s what we found:

Firstly, AI correctly identified many of the things that were obvious in the sales figures: Dresses were the strongest category; Jumpsuits performed well; neutral colours such as Stone, Beige, Cream and Khaki were strong, etc.

But here’s where things get interesting.

• Weather didn’t always produce the obvious result: Despite favourable summer weather, shorts underperformed. AI flagged that temperature alone wasn’t enough to explain the so-so demand.
• It connected location with product demand: London and the South East had a 70% dress mix, while Scottish stores showed greater demand for Tops and Blouses. This led to an actual recommendation: don’t over-allocate occasion dresses to Glasgow and Edinburgh.
• It looked forwards, not just backwards: Using the sales data alongside weather forecasts, AI predicted stronger demand for Dresses and Jumpsuits; lower demand for layering products as temperatures increased; and recommended focusing heatwave stock and marketing in London and the South East.

Conclusion

For any seasoned fashion professional, the results were modest. But it gave us some glimpses about how AI could be used, especially over a longer time period, and when overlaid with regional data and weather forecasts.

We feel the potential is there. And that’s why we are progressing with the actual implementation of AI into our ERP product, continually refining and improving our model as we go along.

Watch this space!