Restaurant operators have access to more data than ever. Here’s how to work with it.
The adoption of more advanced POS systems in restaurants has led to data becoming more accessible than ever before. In today’s data-driven world, restaurants frequently want to capitalize on this by incorporating analytics into their decision-making.
When done correctly, analytics can be leveraged to generate actionable recommendations and guide strategic decisions, like deciding whether you should run a sales promotion again, or whether a certain item should stay on the menu.
The challenge faced by most operators is that their expertise lies in the complexities of daily operations, not necessarily in analytics. This confusion can be exacerbated when operators do their own research, as most analytical approaches (such as machine learning or econometrics) have typically been used by academics, and a lot of online resources can be difficult to decipher.
Understanding the common challenges that arise can be advantageous, as it allows operators to either better validate external analyses or perform simple analytics themselves. Below is a brief overview of a few key things to keep in mind when evaluating or performing analytics yourself:
Clean data is essential for successful recommendations. In my experience, one of the most common mistakes that novice researchers make is that they focus too much on algorithms, and not enough on the data.
Real-world data is often incomplete, contains errors or is in a format that is not immediately usable. To avoid these issues, the researcher must aggregate and compile the data set used in the analysis, which often requires assumptions. This is frequently referred to as “cleaning the data.” Visualizing data can help identify outliers and missing values. As an example, plotting guest counts over time can illustrate missing days and other issues with your data points.
Cleaning data can be a thankless and difficult task, but it’s instrumental to the success of the subsequent steps. If the data is poorly organized or not aggregated correctly, this will likely adversely affect the analytics.
More data is better. Make sure your data represents what’s happening in your restaurant. The more data you have, the more representative it is of your business. One question that often comes up is, “How much data do I need?” More data means more information, which means you can be confident in your results. When dealing with small data sets, an individual observation has more weight on the results, which can skew your results if there are extreme values. Increasing the amount of data can help you avoid this.
Viewing analytics as a toolbox means using the right tool for the right job. You wouldn’t hire a contractor who builds houses with only a hammer. So why answer every question with the same analytical approach? Creating insightful analytics requires using the right tool for the right job, and, in part, the right tool depends on the question you are trying to answer.
If you are looking to answer cause-and-effect type questions, test-and-control experiments or regression modeling are commonly used. For example, a regression model would be advantageous when evaluating the effect of renovating on guest counts, while allowing you to control for other factors affecting guest counts, such as weather.
On the other hand, machine learning approaches, such as tree-based approaches or step-wise regression, are common when attempting to predict an outcome. As an example, these approaches to making predictions can be used to assess which site characteristics are the best predictors of a location’s annual unit volume, helping you choose your next location.
Models should accurately reflect customer behavior. It’s not necessarily what you’re asking, but how you ask it. When creating a model, keep in mind that the model should be a simple representation of the way your business operates. It’s important to include explanatory variables in a way that reflects how you believe your customers make decisions.
For example, if your business is affected by rain, do guest counts fall by an increasing amount as it rains more? Or do guest counts fall if any amount of rain falls at all? In this case, how you would account for rain depends on which assumption you believe to be the most representative.
The result should be actions that help your business. It is important to keep in mind that for many, the point of analytics is to generate insights that allow you to improve the performance of your business. Be careful not to waste time by pursuing approaches that cannot be leveraged to answer the business question at hand. Always ask yourself if the output of a model can be used to generate recommended actions that could improve your business.
For example, telling an ice cream shop that sales are higher in the summer than in the winter is not useful, but identifying why the sales were higher this summer than last summer is. When setting up your analysis or choosing potential questions, always ask yourself if the results will be interesting and more importantly, applicable.
Experts are your allies. For more complex analyses, it’s important to work with an expert who understands that quality analysis isn’t necessarily easy or simple. We know that some companies shy away from a detailed, flexible, analytical approach because it isn’t convenient, which can be a mistake.
Often, the people doing the analysis make key mistakes early in the analytical process, either by using incomplete data or by using the wrong analytical method to answer the question at hand. Such a mistake can be costly, because recommendations obtained from poorly done analytics can lead to misinformation, and could potentially lead you to make a disastrous decision in your business.
As you think about how analysis can help you get to the answers of your questions, know that there is no one-size-fits-all way to use analytics. Keeping these points in mind can help you ask the right questions and ensure that the analytics presented to you are representative of your business and useful.
Make sure you have enough data, that your approach matches the question and that the output of the analysis can be applied, and is actionable.
Alexander Cairns, econometrician in research and development for Revenue Management Solutions, holds a MSc. in food, agricultural and resource economics from the University of Guelph. Since joining RMS in 2013, he has developed a variety of business models and performed advanced analytics for restaurant clients.
Article originally published on April 18, 2017 by Nation’s Restaurant News