Precise size optimisation is key to exceptional inventory management, driving sell-through rates and reducing waste across retail operations. In this second instalment of our series exploring data applications across a product’s critical path, Jennifer Gaskins, PhD shares her experience on how robust product data, granular analysis and machine learning can transform how retailers approach size-level buying decisions.
What are the main challenges retailers face in maintaining accurate size ratios across different product categories?
Size ratios are surprisingly complex. They’re not just a simple mathematical equation, but a dynamic landscape that shifts with product nuances and changing customer preferences. What works for one product category might completely miss the mark for another. The key insight is that looking at size ratios at a broad category level can lead to significant inventory missteps - you might end up with too much or too little stock, neither of which is good for business.
What role does detailed product data play in improving demand forecasting accuracy?
Detailed product data is an essential tool to accurately forecast product demand. Machine learning models can now learn incredibly sophisticated patterns, predicting demand for products even when there’s no historical sales data. This means retailers can accurately forecast trends for brand-new or emerging items, not just rely on the past performance of similar products.
How do you ensure that AI models are aligned with constantly changing fashion trends and size demands?
For AI to keep pace with the rapidly evolving fashion industry, the machine learning models require constant retraining with fresh data and where possible, bringing human expertise into the loop. By combining cutting-edge machine learning with real-world fashion insights, we can keep our predictive models as dynamic and adaptable as the industry itself.
What challenges do you face in training machine learning models on size ratios and demand forecasting?
There are a few key hurdles when it comes to training machine learning models.
The first is around data sparsity - imagine trying to predict demand for a size that’s barely been sold before, this is something both online and in-store sales can suffer from. Fortunately, machine learning’s strength is its ability to learn from global trends while still tailoring predictions to specific sales points.
Another significant challenge is data quality. Accurate size-level information about availability, sales, and returns isn’t always easy to come by. Retailers who invest in robust data collection unlock the true potential of machine learning forecasts.
There’s also the critical risk of underbuying certain sizes. Mitigating this can dramatically improve sales strategies and minimise lost opportunities.
How do you measure the accuracy and reliability of your AI models, and how often do you update them?
Retraining AI and machine learning models frequently on new data and using expert-in-the-loop feedback on the models is critical to meeting the challenges of forecasting for the rapidly evolving fashion industry.
We believe there needs to be an alignment between model performance metrics and meaningful business metrics. We have found that using a metric that’s refreshingly straightforward: the average full-price sell-through rate per product (FPSTR). This means calculating how many items are sold at full price across all sizes, compared to the total inventory bought. We rigorously test our models’ real-world performance, ensuring they’re not just mathematically sound but practically effective. In order to keep things up to date, the models are updated with the latest data before every buying cycle.
What advice would you give to retailers looking to implement a more data-driven approach to size ratio optimisation?
The journey toward data-driven size optimisation is best approached incrementally. The first critical step is ensuring product data is comprehensive and accurate - from detailed product attributes to consistent sizing specifications across categories. Once this foundation is established, focus on collecting robust size-level data for sales, returns, and inventory positions. Close collaboration between merchandising teams and data scientists throughout this process is crucial, as merchant expertise remains invaluable in validating and refining any data-driven approach.