Understanding the Impact of Price Changes on Consumer Behavior
Problem Context
A problem that businesses in the restaurant space deal with frequently revolves around the question: “How can we optimize prices for various menu items in order to maximize revenue?”. This question can stem from different concerns including inflation, supply chain issues, declining foot traffic, etc.
Over the past few months, a team of Columbia graduate students analyzed over three years of data regarding consumer purchasing habits at a US sports bar chain. This franchise caters to a wide demographic through a large menu consisting of all the classic bar foods: burgers, wings, sandwiches, dips, and more. The company’s historical success has come from both its competitive pricing as well as its full bar, which has allowed the business to compete in an environment with large incumbents, such as Buffalo Wild Wings and Chilis. However, given the state of the world in recent years, and in particular the impact of COVID-19, the company is looking for new ways to drive revenue across its business.
Our initial goal was to understand how sensitive customers are to price adjustments and how elastic the pricing is for various menu items. Ideally, we would be able to use the results of our analysis to revise the pricing structure of their menu and drive additional revenue. Our expectation was that we would find that minor price increases would not correspond to an equal decrease in sales volume, netting a positive increase in revenue.
When it comes to consumer behavior at restaurants, there are two effects to isolate for: 1) a substitution effect and 2) a price effect. These are important to differentiate because in the case of the former, a given customer might notice the price increase in one of their favorite menu items and opt for a cheaper, similar meal. In this case, the individual has been priced out of ordering one dish but they are still a loyal customer and will continue to dine there but order different items. The second effect can be more detrimental, which is when an individual notices a price increase and decides to not order anything from that section or decides to dine elsewhere. If the price effect is occurring, there will be a more significant decline in sales volume.
Data Analysis
To solve this problem, we isolated the different factors that contribute to customer behavior, such as seasonality, sporting events, day of week, etc. We cleaned and filtered these features to fit them into a linear regression model where the results highlighted the impact each of these variables had on average gross sales per check. For the dependent variable, we initially looked at gross sales (revenue) but as we made progress in our work, we eventually focused on average gross sales per check.
In the initial regression, we focused on one price change that occurred in 2022 and looked at a 6- month window prior to the change and post. The dependent variable The results were inconclusive — revenue increased for some affected menu items while others decreased. This initial dive into the data highlighted 2 factors we hadn’t considered:
1. Prices were not the only changes made — there had also been changes to the menu in terms of where the items were listed (i.e a menu item was moved from the left to the right side of the menu). Additionally, the restaurant highlights certain dishes as a “Signature Item” or “New”. These are deliberate tactics they engage in to drive customer attention towards specific items.
2. The menu changes occurred at the beginning of March, which coincides with March Madness, a major college basketball tournament that attracts large crowds to sports bars.
To isolate the effect of COVID, we compared the 12 month period between September 2021 — August 2022 with the 2019 calendar year. While the number of checks per month shows high volatility between periods and did not portray any clear pattern, the total spend per check is noticeably stable. Then we considered both changes of the check composition, (grouped by food and alcohol) and store contribution to total gross sales. Our key takeaway was that we should focus revenue per item per check and that our target variable should be average revenue per check.
After these observations, we introduced gross sales per check (including discounts) as our dependent variable because revenue alone doesn’t account for changes in foot traffic nor specific item demand. The results of this second iteration finally made sense and provided strong evidence that demand was inelastic for the price changes that were implemented in 2022. Progress! Our biggest takeaway from our final model was the surprising finding that with a 10% increase in price, there will be a 7.8% increase in sales per check. In other words, 78% of the price increase is transferred to an increase in sales per check. We also found no statistically significant correlation between change in price and quantity per check, which proves the demand inelasticity of the affected menu items.
However, the linear regression model fell short when it came to being able to differentiate between whether there was a price effect or a substitution effect. The optimal next step would be to conduct an A/B test in real time so as to really hone in on what behaviors the price change was causing.
A/B Test Proposal
The A/B test we proposed involved changing the price for one popular entree dish and one section of the menu (a popular burger dish and all items available under Sides). For the former, this would capture any substitution effects because only one item’s price was manipulated whereas the latter would capture any price effects because all of the prices in that section were changed. An additional layer to this A/B test is to control for right versus left digit increases. For example, if an item is listed as costing $11.29, is there a different reaction to increasing the price to $12.00 or increasing the price to $11.99?
In practice, if this was implemented, our next steps would be to conduct exploratory analysis on the data collected and also feed it into our linear regression model. A noticeable gap we found in the company’s workflow was the lack of a robust analytics tool that was able to easily read in and manipulate their data to see changes in demand / revenue for certain menu items over time. Our team set out to build a script using Python to automate these minor formatting tasks that would save their analysts hours of work in Excel.
This script outputs an Excel file that contains an itemized breakdown on the 6 key metrics: Price, Quantity per Check, Sales per Check, Check Count, Gross Sales and Quantity. The table below is a snippet of what the script outputs.
From here, a dedicated team can easily assess what price changes impacted revenue and demand on an item basis. For example, they can manipulate this table to take the YoY % Change and chart this using Excel functionality.
Takeaways
The key finding for our team was that in the short-term, consumers are not sensitive to price changes and for every $1 increase in price, we will see $0.78 flow through to sales per check. In other words, the demand for this sport chain’s products are inelastic (for this price increase). Moving forward, the company could track average revenue per check at the item level to see behavioral changes.
