A luxury retail client asked us to take a fresh look at their promotional strategy. Every quarter, they ran seasonal discounts of around 20%. Their existing analysis showed these promotions drove significant sales uplifts, and on the face of it the numbers were convincing.
But something didn’t add up. Margins were shrinking, and the business was becoming increasingly reliant on promotional periods to hit targets. They wanted to understand whether the discounts were truly earning their keep.
Promotions or seasonality driving sales?
The existing models associated promotional periods with higher sales volumes. That correlation was real, but the causal story behind it was more complicated.
Promotions ran during peak seasonal demand. The models were picking up the seasonal effect and attributing it to the discount. This is a common and easy trap to fall into. Controlling for seasonality without accidentally controlling away the promotional effect at the same time is not trivial. It requires careful temporal modelling and a clear picture of what causes what.
The key question the analysis needed to answer was: what would sales look like during these same seasonal periods if the promotion hadn’t run?
Example chart showing how promotions and seasonality interacted with sales volume (Simulated data, not actual client data).
Disentangling the effects
Working alongside the client’s team, we rebuilt the analysis using causal modelling: explicitly mapping the cause-and-effect relationships between seasonality, promotional activity, pricing, and conversion. This fed into Bayesian structural time series (BSTS) models to capture the onset/offset of the promotional effect and disentangle it from the seasonal effect.
The picture that emerged was very different from what the correlational analysis had suggested. Promotions were actually having a slightly negative effect on conversion rates. The client was discounting their products and getting fewer conversions as a result.
The brand’s positioning played a key role here. The promotional messaging and discount signalling were working against the perception of quality and exclusivity that their customers valued, so the discounts were counterproductive.
Unlocked £5m/yr in incremental revenue
We identified a gap of approximately £5m per year between the client’s current promotional strategy and what the data suggested they should be doing. The client restructured their approach, and the numbers are already bearing it out this quarter.
Causal modelling for business decisions
This kind of hidden cost is more common than most businesses realise, and it spans every industry. Whenever a business decision coincides with other factors that affect the outcome (seasonality, market trends, competitor activity), there’s a risk that standard analytics will conflate the two.
Causal modelling is designed to untangle exactly these situations. It answers the question that actually matters when you’re deciding what to do next: what would happen if we did something different?
If you think there might be a similar question buried in your data, get in touch. We’d be happy to explore it with you.