Whether you own a sports team, run a ski resort, or sell school supplies, you’ve likely come across some seasonality in your sales and revenue. Indeed, even seemingly noncyclical businesses, like fast food or clothing, experience some sort of seasonality.
Seasonal sales are risky because, in the most extreme cases, you have only a few days to make your sales numbers. Imagine you’re a retailer preparing for Christmas — success or failure in mid-December might mean success or failure for the entire year. With that in mind, an analytic tool that gives you an idea of what to expect can be a crucial resource during the seasonal sales cycle. Here, we present an example of how one can apply predictive analytics to forecast seasonal sales months in advance and take pre-emptive action.
A brick and mortar example
We recently worked with a brick-and-mortar business that receives over 80% of its revenue in late Q3 of every year. Given such an intense seasonality, it is crucial to be able to predict fluctuations in potential sales and pre-empt specific trends or changes in revenue.
We worked with approximately 75 different customer segments at the business, generating specific characteristics about each segment. The goal of the modeling task was predicting major changes in revenue months in advance of the major seasonal revenue influx.
Specifically, we built a model using a cost-sensitive random forest algorithm to predict revenue fluctuations greater than 5% compared to the year before. A 5% growth is important as it signals a strong and healthy segment. On the other hand, those predicted to decline by 5% should be prioritized for pre-emptive marketing campaigns.
Using our random forest approach, we are able to increase the accuracy to 63%. More importantly, however, we can identify over 58% of all cases where revenue will fall. Better yet, it’s possible to identify the above more than 4 months in advance of the seasonal sales bump.As an alternative, random guessing would allow you to be correct a third of the time. In most situations, it would also mean lack of action for a specific segment, or investing resources in the wrong group of people.
Applying the findings
So given such an exercise, how do you apply this to actual marketing and sales operations? Simply put, the process allows you to prioritize segments for active and aggressive pre-emptive, targeted marketing strategies. Here are the steps we recommend:
- Choose an ideal number of campaigns to run. In the case of the business above, choosing the top 16 risky segments would be ideal. This would yield effective marketing campaigns on 11 of the 16 segments. Note that this is higher than the earlier accuracy, as higher risk scores tend to be more accurate.
- Begin targeting segments as early as possible. The 4-month advance notice in the case above allows for planning and execution of campaigns long before the revenue peak in Q3. Start marketing as soon as possible.
- Collect data and update models regularly. The random forest approach we used above can be updated easily within minutes on a regular basis. We recommend tracking segments and updating models daily to see if marketing campaigns are working and to plan accordingly.
Of course, there’s only so much we can write in a short blog post. We’re keen to tell you more! If you run a seasonal business, let us know and we would love to chat with you!
|Next Post:||How to measure email campaign purchase conversion – R Console tutorial|
|Previous Post:||When to offer percentage discounts vs. gift cards in e-commerce marketing|
Click here to subscribe to Canopy Labs Insights!