A small data exploration on #spreadsheetsaturday: looking at a model for predicting a KFC menu price given a group of different menu items.

Taking a meal with different components as input (e.g. comprising a burger, fries and drink), how could we predict the retail price of the menu? We can start by making a matrix breaking down listed adult menus into their components. The column “component sum price” shows the total price if we were to buy the components separately without any meal deal.

A matrix breaking down menus into sums of individual components. Some individual prices are guesstimated (e.g. popcorn chicken L is unavailable outside a deal, and the average is taken of different dip prices). Where a quantity of 1 in unavailable, the figure is derived from the smallest purchasable quantity divided by number purchased. Prices as of 17/7/22 as listed on delivery service Thuisbezorgd.

We can plot the two columns “component sum price” (x) and “retail price” (y) and compute the regression line:

Note the clusters around, for example, y=€13.95, where different meals have a common retail price.

The equation for the regression line f(x) = 0.51x + 3.83 provides a method for predicting a menu price from the sum of individual components. This is used to calculate the “predicted meal price” column in the matrix above.

By looking at the relative prediction error, we can also explore which menus are more or less expensive than expected. This could be a interpreted as a quantification of “menu value for money”, with menus significantly cheaper then expected/predicted being a better deal.