A Single Demand Forecast (Part 4): What Should Retailers Ask For?

After reading the preceding posts, you probably agree that a single forecast for all retail business processes is not the right thing to ask for.

But then how do you address the original issues that prompted the discussion:

Each business function is working with a different forecast in mind. Can’t all of us work off a single number?

The technology and personnel investments required to address these differing forecasting needs are mind boggling. Can we consolidate them?

These are valid issues that should be addressed. Our recommendations:

Establish a Sales & Operations Planning process to achieve alignment

Create a central forecasting methodology group that serves different organizations

Seek a data and technology infrastructure that serves multiple forecasting needs

Forecasting Insights Retail Supply Chain

Bottom-Up Forecasting: Be Wary of the Sophists

It is common for some retail guru’s/ practitioners to speak to the virtues of forecasting bottom-up as the best way to get a good forecast as a universal truth . Let’s examine the mathematical fallacies of this argument with an example.

Scenario: You are interested in creating a forecast for how much a store is going to sell in a given month. You want to use this as one of the inputs for the district management team. Your arch-nemesis (Dr.Z) and you are given the task of creating the forecast, the person with the best forecast is going to be promoted. The store sells only three items. You have 12 months of data (shown below). The race is on.

Dr. Z is a believer in bottom-up forecasting and is going ahead and creating forecast models using the finest software he has. Although the item sales figures are rather volatile, he confident that he can find enough causal factors to explain a lot of this variation. So, he sends his ace data folks to collect all the causal data he can to figure out what is causing the ups and downs in the forecast. He’s off to a flying start. Dr. Z successfully develops a model, where he has a forecast error of 10% for each of the three items and is very thrilled with the accuracy he’s able to get with the bottom-up forecasts (results below).

You just came back from vacation, and you learn about the fancy work that Dr. Z and his team have been doing. You have only one day left before you are expected to make the presentation. What do you do? You add up the sales for the items to look at the pattern of sales at the store level, as that is what you have been asked to forecast (shown below):

It’s evident that forecasting at total store level for this store is relatively simple and you may not need to do a lot of work. So, you submit that the forecast for this store is $100/ month.

From the example above, it’s evident that forecasting at the lowest level of the data (you have) is not always going to yield the best result. It may yield a good result, but is always a good idea to try multiple methods (top-down, bottom-up or a middle-out approach) to determine which method (or combination of methods) will yield the best result.

This is not just a clever example, but the nature of most processes in business (and nature). The lower level is always more noisy and harder to predict than the higher level (as noise across different time series cancels out like in the example above).

Forecasting Insights Retail Supply Chain

A Single Demand Forecast (Part 3): Be Careful What You Ask For!

A point I made earlier is worth repeating. The premise that I am challenging is a single forecast for all business processes, NOT the premise of a single forecast for a single purpose across the supply chain (which I think is a good idea, worth pursuing). The lack of clarity on this issue is what I am seeking to clarify here.

There are two problems with a single forecast that supports all business processes:
a)      A single demand forecast for all business processes is sub-optimal
b)      It is not possible to create such a single demand forecast (note that (a) trumps (b))

A single forecast for all business processes sub-optimal
When a business process requires a forecast, it has a certain expectation of forecast granularity and forecast horizon.

Forecast granularity: At what level of the business do you need the forecasts (to take the operational decision)? For most purposes in retail, the granularity is presented in terms of Merchandise – Location – Time.
Forecast Horizon: How far out do you want the forecast for, and how long do you need the forecast for. This is determined by the time it takes to act on a forecast, e.g., if you need to order an item from a supplier who has to make it, you generally need to send them a forecast way in advance in comparison to an item that the supplier keeps in stock.

Some examples in the table below:

Let’s consider the extreme examples here: Strategic Planning and Labor Scheduling. It is intuitive for managers to recognize that it is better to forecast at a higher level for strategic planning purposes rather than to roll-up hourly-item-store level forecasts to arrive at the forecast for the next four years. Intuitively we know that, while forecasting for the next 4 years, you need to consider the trends in your overall business, competition and macro-economic factors as opposed to your current merchandise mix, or last month’s hourly sales patterns. On the other hand, when you are doing labor scheduling for the next two weeks, it is critical for you to understand hourly sales patterns not only of similar time periods from past years but also of last week.

Beyond the fact that it intuitively does not make sense, it is mathematically sub-optimal. For more on this topic, read my previous post on this topic: Bottom-Up Forecasting: Be Wary of the Sophists.

It is not possible to create such a single demand forecast
Once again in our strategic planning and labor scheduling example, if the approach to a single demand forecast is roll-up the bottom up forecast. One needs to know the merchandise mix for the next four years at the store level. It is not possible. In many cases, you don’t know the precise merchandise mix that will be in the stores six months from now. So, it is not possible to create a strategic plan based on a roll-up of hourly forecasts. As a more practical example, it is not possible to create a merchandise financial forecast for the next 6 months based on hourly forecasts for the next six months.

In summary, the idea that a single forecast at a store-sku-week level will serve all business planning purposes is mis-conception. It intuitively does not make sense, mathematically incorrect and ultimately impractical.

Now that I have described what you should not ask for, I am sure you are thirsting to read what the solution is. We cover this in Part 4: What should retailers really ask for?

Forecasting Insights Retail Supply Chain

A Single Demand Forecast (Part 2): Why do Retailers ask for it?

Forecasting is a necessary evil in the retail industry (or for that manner in any industry). In the past two decades, retailers have realized that there is significant value in making science an integral part of their decision making.

Starting with Computer Assisted Ordering(CAO), retailers have continued to gain benefits from applying scientific approaches to processes such as: Replenishment, Allocation, Pricing, Promotions, Markdowns, Assortments, Size mix, Labor scheduling and Financial Planning. Forecasting is a critical component of all these areas.

Without exception, in all the areas mentioned above, a good forecast is atleast 60% of the answer. Hence, all these solutions come up with a forecast, and recommend how to control the operational levers (e.g., staffing) to meet such a demand forecast.

In many cases, making individual processes smarter, has provided rewards and the money is in the bank. Now comes the next series of questions:

Why do we need so many forecasts? If all of these processes are creating forecasts, is it not correct to assume that there’s one answer that’s better than all others?
Can we consolidate the efforts into something more meaningful (like a single forecast)? Forecasting requires skilled individuals who are quantitatively adept and have a solid understanding of the business. So, the inevitable question:
Should we be buying a forecasting system and put all these business processes on top of that technology? Good forecasting requires a lot of infrastructure – good granular sales and inventory data, capture of causal information, good master data for products and stores etc., In addition, large volumes of data need to be crunched in short weekend windows. All these  require large (and on-going) investments in hardware.

It is easy to lead to the conclusion, of course, we need a single forecasting system that produces one answer and the entire business runs on it. We all wish for a simple world, but Retail is not meant to be that easy, it wouldn’t be fun, would it?

Read my point of view on why these are the right questions, but a single forecast for all processes is not the answer in the next installment of this blog: A single demand forecast (Part 3) – Are you really sure you want one?

Forecasting Insights Retail Supply Chain

A Single Demand Forecast (Part 1) – Debunking the Myths

Retail executives are struck by the number of seemingly conflicting forecasting processes that their organizations are investing in. This series of 4 blogs examines the problem and recommends a solution, and it is not what you think.

Retail executives are struck by the number of seemingly conflicting forecasting processes that their organizations are investing in. In response to concern, there’s a whole industry that has emerged that touts the benefits of a “single demand forecast” and how they either provide a methodology or a software solution that delivers exactly to this expectation.

Many retail executives have asked me the same question. Unfortunately, the answer is not that simple.

Pause for a minute, and think – A single demand forecast – for what?
(a) For a single business process (e.g., replenishment) across the entire supply chain (to avoid the bull-whip effect); or
(b) For all processes (e.g., labor scheduling, replenishment, strategic planning etc)

If you answered, (a), you are an enlightened soul, hope you have a lot of power to make it happen in your organization (as multiple forecasts across the supply chain for the same thing does cause a lot of inefficiency). But you should note that you still have the problem of multiple forecasts for different purposes and may want advice on what to do about it.

If you answered (b), my writing here is specifically intended for you – to clarify the issues and hopefully convince you that you should be more specific in what you really want, otherwise, you might regret the outcome.

I don’t think there is a single demand forecast that will serve all the needs of a particular retailer (or for that matter any organization). If you really want it, I am sure someone can provide you one. Will it help you better sense and respond to customer demand in a profitable manner, most likely not. It’s not a technological weakness, it’s the wrong thing to ask for.

In this series of blogs, I will provide my point of view on this issue. The series will cover:

Part 2: Why do retailers ask for this? i.e., what problem are they seeking to solve
Part 3: Be careful what you ask for? i.e., how a retailer could actually lose their ability to respond to the market by seeking a single demand forecast
Part 4: What should retailers really ask for? Alternative thinking that actually addresses the problem and makes your organization more responsive to consumer demand

Forecasting Insights Retail Supply Chain
Our site uses cookies to improve your experience. You can control cookies by adjusting your browser or device settings. By using our site and closing this box, you consent to our use of cookies.