4 minute read

Why Most System Forecasts Fall Short of Your Needs

This is Part 5 of a 5 Part series on Forecasting by Dan Bursik. To read Part 4, click here.

One of the things that a store operator learns very quickly once they manage a variety of stores is that some stores are easier to forecast than others. Any chain that operates a variety of stores in different neighborhoods, different demographics or operates in different format variations gets to know very quickly that business does not flow the same in every store. And the real difference isn’t necessarily about the big things: Thanksgiving is still Thanksgiving and other big calendar events don’t vary that much. The real difference is in the flow of business through the days of the month, the pattern of the week and the times of the day. That is where every store can be different from its peers.

So, why then, do many organizations use a forecast methodology that is one size fits all? Many do this and they embrace a “close enough for the stores to edit” mentality about what they want out of forecasting. While that is one way of doing it, relying on store personnel to fix what may be a misinformed forecast can be a solution that causes more problems than it solves.

Many systems have limited approaches for forecasting. Some abdicate the function altogether and force the system users to select the historical data set and identify the math operations (e.g., straight average, weighted average, trend over LY or selected week, etc.) to generate the forecast. Even then, the operation you select might apply to all forecast elements (e.g., Front End same as Deli, Meat, Bakery, etc.). And the one-approach-fits-all might even mean that the process you select to forecast your suburban stores is summarily imposed on your inner city stores whether it fits or not. Does this make sense?

A better approach to forecasting is one that allows different methods to be used for different stores. An even better approach is one that allows different approaches to different departments within the same store, as well as different approaches for different metrics within a department.

It should be flexible enough to get it right, but useable enough to support a reasonable workflow each week that arrives at highly reliable outcomes. There is a variety of reasons that certain algorithms perform better in some situations than others. It is also true that certain algorithms may work well in normal weeks but may fall short in holiday weeks that require transposition of history across weeks, or as the day of a holiday lands in one year versus the next.

Ideally, you want to be able to leverage the best of the best algorithms. The trick is to know which algorithm to use for any given week to get your reliable outcome. Like a skilled carpenter with a full workbench of tools, you don’t want to use a mallet when a chisel will do a better job.

Learning algorithms are a breakthrough opportunity for improving the quality and reliability of your forecasts. I use that term to mean two things:

1. That the algorithm is tested on a regular basis to understand its performance relative to other available algorithms so that the system can automate the process of selecting the right algorithm for each forecast metric.

2. That certain elements within the algorithm can be fine-tuned based on that experience of forecasting prior data sets for a given store-specific metric.

While limitless options are very powerful, the system must provide usable options to select. It must also use the right algorithms for each metric and make that assessment for each metric at each store within your enterprise. That’s a very powerful leap beyond manual forecasting.

Keep in mind that the whole objective is to make the system more accurate – more helpful in guiding your stores through best practices to put the right people in the right place at the right time doing the right things. It may not be rocket science, but it is good, solid planning.

So why is reliability just as important as accuracy? It makes all the difference in your people’s ability to trust the system and stick to your best practices. If that’s not your goal, then be satisfied with every store trying to wing it as best they know how. But, in the absence of best practices, don’t be surprised if you also see associate preferences, or managerial blind spots creep into your operation as your people, with the best of intentions or otherwise, do what they feel they need to do in order to post next week’s schedule – whether there is a plan behind it or not.

If you agree that the best road to success is to understand your best practices, to use them to plan your work, and to create systems and support to enable your manages to successfully execute your plan with the least amount of on the fly adjustments that you can manage, then forecasting will eventually become a point of focus for your Labor Management systems and process. It’s not the only thing that can make your organization more successful in helping your stores run better, but it can make a big difference in your overall success. Make sure your systems do it correctly, do it well and do so reliably so that your people can trust the plan you create and follow your best practices as closely as possible.

Focus on what can make a difference for your organization. Hopefully, I’ve given you some fresh ideas and the benefit of my experience. Now it’s up to you to plan your work and work your plan.

Continue reading

Let’s Connect