Assume we already have a product in the market, & we’re running some promotions in order to stimulate sales. We want to know which are our most effective promotions? We have gathered data, which by itself could be of two kinds: actual sales figures of each particular customer, or choice data from a store, ie., whether a customer chose our product, over the offerings of our competitors.
Performing a response analysis of the sales or the choice data will give us an idea as to which is the most effective promotional strategy to stimulate demand. This is a simple illustration of this process using linear or logistic regression on the appropriate data set. It’s purpose is to just showcase the results of this exercise & how it can guide the discussion process with the marketing teams. In reality, there’s a considerable amount of behind the scenes technical & business complexity that we have to attend to before rolling out a promo.
For instance, besides refining our models, we have to calibrate the time period of observation, as sales are a lagged indicator of promotions. Here, in this eg., we’ve used a lag of a week. But, typically, this figure involves discussion with experienced sales folks, as it could vary depending on the type of good. We can also get as geeky as we want in this area, by incorporating seasonality & looking at auto-correlations, through time series analysis & dynamic regression etc., particularly when there’s enough available past data (not shown here).
Additionally, we also have to consider the overall business impact of promotions which extend both before & after the promo period, to incorporate leakage (when consumers delay their purchases before the promo), & hangover (when consumer delay their purchases again after the promo, in anticipation of future promos). Moreover, there’s also the need to determine the right time window between successive promotions, as that has a bearing on annual profit.
The point is that just like in other cases, the results & technicalities of analytics should never be taken in isolation, because its principal usefulness is in stimulating discussions between teams/departments, & holistically incorporate all business needs.
This data comprises of observations of 124 clients already on an online subscription plan on our app.
Each client was exposed to a particular combination of promotions & price. Here, we’ve recorded the subsequent sales corresponding to a particular price & promotion combo. Each observation corresponds to a week of sales.
The promotions were limited to either showing a display ad or giving out redeemable coupon codes.
Here, we see that a combo of display ads & coupons, as opposed to using just one mechanism, stimulated sales the most.
This was followed by just coupons & only display ads in decreasing order of effectiveness.
This was obtained by running a linear regression on the sales data.
This data comprises of 2798 observations of the buying choices of 300 households from several brick-and-mortar stores.
It records whether they bought either our product or a competing product, in the midst of multiple competing promotions & price changes, in each store.
As before, the promotions used by us were either display ads or redeemable coupons. Here, our competitor copied our promotion strategy.
Here, we see that we were most hurt by our competitor undercutting us on price. This suggests that consumers have a high degree of price sensitivity, something that we should incorporate into future offerings & messaging.
Additionally, what helped us the most in getting the consumer to choose our product, was exposing them to a combo of both display ads as well as coupons, as opposed to just one promo mechanism.
This was obtained by performing a logistic regression on the choice data.