September 11, 2023
Imagine this: you made a commitment to hit a certain level of revenue. You're a bit behind but you're not worried: surely, you'll be able to intensify sales velocity by launching a discount campaign. No so fast: it might not work as expected!
As we explained in detail in another blog post on sales velocity optimisation, despite of that fact that shopper are on a constant hunt for deals, discounts don't work any longer. Errors in price elasticity of demand estimation is one of the core factors that are responsible for this. To be prices, price elasticity of demand miscalculation is not to blame for inefficient discount. What it has profound impact on is overestimated sales forecasts.
But let's see how exactly these errors happen.
Price elasticity of demand (PED) is a concept that explains that price changes for various products will yield different outcomes. Some products are more price-sensitive than others.
A classic example of a category with low price elasticity is cigarettes. If cigarette prices rise, only a minor portion of smokers would quit. Most will either switch to cheaper brands or make no change.
The opposite is perfectly elastic demand: when the quantity of sold goods isn’t influenced by the price. Here, any price alteration results in a significant change in quantity demanded. But this scenario is more theoretical and rarely seen in real-life markets.
The popular formula to calculate PED isn't that complex:
Q =Quantity Demanded
P = is the initial price (or any given price)
β1 = Coefficient for price, representing PED. Or, in other words, it is a change in quantity demanded for a one-unit change in price.
The important “magic number” in this equation is Coefficient Estimation (β1). This is exactly what will tell you how your sales numbers will change when you change your price by X.
The challenges start to arise on a stage where you’d try to make an estimation for β1. The most common method used for estimating the coefficients is OLS or Ordinary Least Squares Method (it is generally widely used approach for estimating the unknown parameters in linear regressions). You can easily outsource this to Excel as this valuable tool allows you to do linear regression in a couple of clicks. The problem is not the formula. The problem is data!
In order to have a relevant β1 estimation you need whole bunch of data to be collected, normalized and then analysed.
You would need:·
· Historical prices
· Historical Sales (the more your observation period — the more relevant results you’ll get)
· Correlating marketing efforts you had (for example, you’d want to count separately sales that came ducting activeGoogle Ads campaigns and sales when you did nothing to attract additional traffic)
· Historical competitor prices
· External factors like inflation rate or weather data (if you sell seasonal goods)
Data preprocessing includes
· and transforming raw data into a format suitable for analysis.
If you’re not very familiar with this process, you might be surprised by the fact that 90% of data analysts time is actually spent on this pre-processing stage. Why? In simple words, you can’t through together pictures and graphs from ONS, couple of links from weather forecast websites and csv files with your historical sales. They are simply not compatible. So your data analyst has to bring all these different types of data to a single format before analysing.
Data analytics stage includes applying formula or using Excel to build multiple linear regression model. If you have several dozens lines of data it will take Excel several seconds to show you your Coefficient Estimation. If you have thousands of lines (because mind you, all the above mentioned data should be collected and normalised for 1item only!) Excel most probably won’t handle it. Your salvation can be Python or price optimisation software (like Aimondo).
Finally, getting back to our main equation
Now all the ingredients for estimating price elasticity are in place. Surely, we can handle it from here. Not so fast! There are still multiple factors that are completely unaccounted for in your Excel spreadsheet!
According to the Ehrenberg-Bass Institute for Marketing Science, there are 6 other factors that impact price elasticity of demand and that you'd better though in your calculations.
We’re talking not just about product brands but seller’s brands as well. Consumers tend to have market leaders on top of their consideration set, therefore they always compare other players with the leaders. This is why many smaller market players find adopting a Price Match policy very efficient: it is based on the idea that the lower you’re in the brands’ hierarchy, the more price elastic you are.
The “older” the product — the less price elastic it becomes. Hence, experienced category managers prefer to keep their assortment balanced by the product lifecycle stage parameter (we have an expert covering this topic in detail in our free Pricing Excellence course).
(remember, the cigarettes?)
Premium market segments are less price elastic than mass-market, so the scope of change should be larger to incentivize shoppers to make a purchase.
The quality of message distribution is very important. Fake Wilko campaigns hit an unbelievable number of impressions because they went all-in with Facebook ads and made sure pretty much everyone saw their creatives.
Customers who have already developed an affinity with a certain brand are less impacted by price change, aka their demand is less price elastic.
So, how realistic is the expectation that an average Marketing Manager would take all these into consideration to figure out how a simple discounting campaign would impact sales velocity of a certain product? Apparently, it's very unrealistic. But that's why AI-powered tools for predictive analytics like Aimondo exist.
Step-by-step guide on maintaining high profitability in retail.