Setting the Promotional Strategy for a Bakery from Data

Let’s look at whether day of week explains some of this difference:Indeed, looking at the above two charts, we can see that weekends appear to explain the majority of the spikes in values.Finally, we’ll look at how the volume of items sold changes by time of day and day of week, as well as by hour of day.The above helps inform our remaining analyses by giving us knowledge of the most popular items — and how that varies by time of day — as well as telling us that we should craft a strategy that takes into account the difference in sales by day of week and/or time of day.Setting StrategyNow that we have a better understanding of the data and its distribution, we can move into the phase of using data to inform decision-making.There are a few guiding principles to take into consideration in our strategy development:We want to increase the number of items sold in a way that incentivizes items that otherwise would not have been purchased.We should take into account the variables that we know change the way that customers interact with us (i.e. time of day and day of week).The final output should provide clear directive on how to act, why this course of action is desirable, and it should not be too confusing for the customer nor too complex for an operator.With transaction data, we want to get a better understanding of the following question: The purchase of which item(s) lead to purchases of what other items?To make this more tangible, we want to know whether, for example, buying a coffee means that a customer is more likely to also buy a muffin, compared to the likelihood that any customer will buy a muffin.There’s a well-studied approach to these types of questions, and it’s called the Apriori algorithm, which has three main components: Support, Confidence, and Lift.Apriori helps us understand, for a pair of items, the frequency of item 2 being purchased (support), the likelihood that item 2 is purchased if item 1 is purchased (confidence), and the increase in the ratio of item 2 being purchased given that item 1 is purchased (lift).(For more reading, this is an excellent article).With these in mind, we have a few considerations for how to solve the problem:We could look for item pairs where item 1 has high support (i.e. is purchased frequently) and item 2 has low support, but the two have a high combined lift (item 2 is purchased more frequently when item 1 is purchased).We could look for an item that has low support but generates high lift (i.e. we want to incentivize the purchase of item 1, which isn’t bought frequently, but leads to other purchases).We can look for two or more products that are frequently purchased individually, but not purchased frequently together, and bundle those to create a more compelling offering to the customer.We would use the above to think about, for example, how to set up discounts (“buy item 1 and get item 2 at a 50% discount”) or to bundle product options (“get item 1 and item 2 for a price that’s less than item 1 + item 2”).Likewise, on a calendar level, we have a few questions to consider:Should we try to boost sales on days that have historically had lower sales (i.e. mid-week)?Should we try to sell more on days that already have a higher volume of transactions because our starting customer base is higher (i.e. weekends)?Should we target by time of day, knowing that our volume of transactions differs by hour?Let’s take a brief detour to see some of these concepts brought to life..Below I’ve shared a screenshot of the DDPerks screen of my Dunkin Donuts app..Let’s look at what they’re promoting and why:First, we see discounted cappuccinos or lattes, but only if I go between 2 PM — 6 PM > They’re trying to encourage visits during a specific time window that otherwise would (probably) have lower sales.Second, there’s two breakfast sandwiches for the reduced price of $5 > DD’s is employing product bundling to increase total order value.And finally, there’s 2X points on Wednesdays if I use their On-The-Go ordering function > The company is encouraging a certain type of behavior to demonstrate a product feature and, possibly, better manage order flow inside the stores.Further, these are likely personalized to location and my interactions with the app — what I purchase and how frequently..If you have the app, are your offers different?The StrategyWhen evaluating the results of the Apriori algorithm, combined with domain knowledge, we can pursue the following strategies:During weekday mornings before 12 PM, get any hot beverage of your choice (Tea, Coffee, or Hot Chocolate) and breakfast item (Toast, Cake, Muffin) for less than the combined price of the two.On weekday afternoons, buy a sandwich and get 50% off a tea.On weekends, buy any two items and get the third item at 33% off.Why?On point 1, we know that almost all of the purchases that drive the highest lift (increased ratio of sales) during weekday mornings are a combination of hot beverages and breakfast items (see table below)..Because our results vary by by beverage and food type, this option is broad enough to not restrict choice nor narrow the market for this option..Further, we know that a hot beverage is purchased in at least 70% of morning transactions, giving us a strong base of initial purchases.Apriori results for weekday morningsFor weekday afternoons, we want to incentivize purchases that otherwise would not have been made..To do this, we’ll look at the item(s) that have the highest combined lift..The results were tea, coffee, and sandwiches, in that order..In the afternoon, sandwiches and tea sell well, but they’re not frequently bought together..Tea is purchased in ~17% of transactions, sandwiches in ~12%, but together they only comprise ~2.5% of transactions..Therefore, we want to create the purchase of the sandwich (higher value item), given that the tea is popular too, but not when the sandwich is bought..We want to create this multi-product purchase that historically has been a single purchase.Finally, for weekends, we find that, relative to weekdays, there are more multi-purchase transactions that drive the purchase of a third item.50% of the top 10 drivers of purchases on weekends are 2-item antecedents compared to 20% on weekday morningsWhen someone already buys two items, we want them to buy a third item, and this seems more likely on weekends.. More details

Leave a Reply