A Shopify client came to us unsure whether to restock a perishable line. We ran their data through a Hidden Markov Model: a statistical method which aims to predict whether a product is in a growth or decay regime, and forecast (with statistical confidence) which phase it is about to enter.
The model detected a decay signal with high confidence, and we produced a prediction using autoregression on the data. The client had access to this, however what use would a big dashboard be? They just needed a call on whether to restock or not before Christmas.
The chart below shows the forecast made at the decision point the red dashed line and what actually happened to sales afterwards. I think there is a lot of statistical low hanging fruit out there beyond the various averages shown by dashboards, the tree just needs to be approached from a different angle.
For a decision like whether to restock a perishable product before Christmas, standard shopify dashboard often don’t provide enough insight. Metrics such as traffic, conversion rate and total sales tell you what happened, but they don’t necessarily help predict what will happen next.
In this case, using a Hidden Markov Model combined with autoregression provided a more actionable approach. The model analyzed historical sales patterns to determine whether the product line was in a growth phase or a decay phase, then forecasted future demand based on those trends.
The results showed a strong decay signal and predicted lower sales going into the holiday period. As a result, the recommendation was not to increase inventory aggressively, helping reduce the risk of excess stock and waste associated with perishable products.
What makes this approach valuable is that it focuses on business decisions rather than reporting metrics. Instead of simply showing historical performance, the analysis answered a specific question " should we restock before Christmas?" The subsequent sales data closely followed the forecast, providing additional confidence in the model’s recommendation.
For shopify merchants dealing with seasonal products, perishables, or inventory- sensitive categories, predictive modeling can often deliver more practicle guidance than traditional dashboard analytics becuase it turns historical data into forword-looking business decisions.
Good example of why a single number on a dashboard rarely tells you what to do. The dashboard shows what happened, the decision still needs context - seasonality, margin, lead time - that doesn’t fit in a standard metrics view.
The restocking question is a good one because it forces you to think about profit per unit, not just sales volume. A product can be your top seller and still be a margin drain once you factor in returns and holding costs.
