2

Why You Need a Forecasting Plan Before You Buy Forecasting Software

Many companies rely on forecasting software for planning and inventory management, but an algorithm rarely solves all forecasting needs. Coming up with an accurate forecast requires a mix of software, expertise in data analysis, and most of all, an intelligent planning framework.

Forecasting systems rely on the assumption that the past will repeat itself. In the modern, fast-paced business environment, this assumption is often wrong. Increased competition, new product launches, and shorter product life cycles are just a few elements fueling instability and making it difficult to obtain accurate forecasts no matter what algorithm is used.

So what is it exactly that forecasting systems find hard to do?

Forecasting software struggles to assess new trends and handle abrupt pattern changes. This is because every forecasting system aims to produce a forecast with the fewest number of errors possible based on

Even the best algorithms can’t challenge input from the sales team or question qualitative market data,such as the impact of new customers or markets.

​Vendors of forecasting software will say “you don’t need to know how an engine works to drive a car.” True, but forecasting is much more involved than driving a car. Roads don’t constantly change, and they even have helpful signs to indicate where you should go.

In the world of forecasting road signs are hard to spot. It’s you against a huge, constantly-shifting stream of complex data. Unless you want to end up stranded, you need a mechanic riding with you who does understand how the engine works and knows when to tinker with it.

The truth is that the more sophisticated your forecasting system, the greater your odds are of making mistakes with it. Highly sophisticated models are very reactive, increasing the need for an expert to control them.

Here are 5 common situations with which statistical forecasting models struggle:

1

New trends or sudden changes in sales patterns or sales levels

Let’s suppose your competitor stocks out of a product similar to one you carry, which leads to increased sales of your product. Forecasting software might misinterpret the short-lived increase in sales as the start of a new trend and forecast increased sales over the long run. However, there are many potential causes for a change in sales, so it’s important to question the data before creating a sales forecast.

Now, you may already have automated alerts set up to inform you of changes in sales. However, like your forecasting software, these alerts can only act based on pre-set rules. So, for example, you may blow up with alerts once a certain product hits 1,000 sales, but never be the wiser after months of sales hovering at or just below 999. A human forecast analyst would be able to catch this.

Furthermore, setting alerts that cover many different products is a sensitive process. Small adjustments can make the difference between receiving 20 alerts per week or 200. Too many alerts can lead to alarm fatigue and subsequent desensitization to important notices. So, relying on automated alerts is not the solution.

2

Complex sales patterns requiring business logic to decipher.

For most companies, 70% to 80% of items tend to be slow movers with complex sales patterns that look random. Seasonal patterns are often hidden within this seemingly-random data.

someone unfamiliar with your business. Your forecasting software will treat these patterns as random, leading to a flat forecast.

Now, suppose you have 1,000 slow-moving products with an incorrectly-flattened forecast in a period during which you will be experiencing decreased sales. Your forecast will end up being too high, harming your capacity planning. This is a situation in which predictions made by forecasting software must be coupled with business logic.

3

Slow-growing trends

​Imagine that sales for one of your products have been increasing so gradually that you have not noticed it. Since your forecasting system relies on a sales history several years long to make future predictions, it could take 6 months or more for your software to recognize the new pattern as significant. As a consequence, you will run the risk of stocking out, because your software will have underestimated sales over a long period. Analyzing slow-growing trends is especially critical for products with long lead times, as mistakes can be very expensive to fix.

4

Interference from the sales team

​​Say your sales team has been overriding your forecasts because they see your forecasting software as a spooky black box. They don’t trust it, especially if they have experienced stock-outs before. Sometimes this strange software does counterintuitive things, like increasing the forecast during a period of seemingly low sales. This may be perfectly rational if the software is looking at a longer period of time than the sales staff, but could seem stupid. Since they trust their own brains more, the sales team will likely just add more inventory.

​The brain is biased, however. As a consultant, I have seen companies across sectors consistently overestimate stock needed by as much as 50%, indicating a natural bias toward overestimation. It seems fair to say that sales teams fear stock-outs more than excessive stock accumulation. Since stock-outs directly affect their bonuses, and could even lead to job loss if they miss too many sales targets, this makes sense.

​A software program can’t stop your salesforce from changing forecasts, explain how its algorithm works, or take their input into account. This is a management issue that will take time to resolve. The sales team needs to see accurate forecasts over time to believe that the system works. However, a forecasting expert can help communicate with the sales team and start building trust with them.

5

Missing market data

​​​Now, imagine your sales team has only provided input covering the next few weeks, but you need to work with a much longer forecasting period. Just as I have seen a tendency to over-forecast in the short-term, I have also often seen a lack of sales input in the medium to long-term range.

​The problem is that your software can’t challenge missing data, so it will do the best it can with a short horizon, providing an inaccurate forecast. From experience, I can tell you that the sales team needs to be prompted often to chip in. Otherwise, their input will arrive in bulk, making stable planning difficult. Only a human can manage this relationship.

​Now, imagine your sales team has only provided input covering the next few weeks, but you need to work with a much longer forecasting period. Just as I have seen a tendency to over-forecast in the short-term, I have also often seen a lack of sales input in the medium to long-term range.

​The problem is that your software can’t challenge missing data, so it will do the best it can with a short horizon, providing an inaccurate forecast. From experience, I can tell you that the sales team needs to be prompted often to chip in. Otherwise, their input will arrive in bulk, making stable planning difficult. Only a human can manage this relationship.

If you would like to improve your forecast, never start with software

​A forecasting system is a tool, like a screwdriver. Before you take out the screwdriver, you need to assess whether the item you are trying to fix contains screws, nails, or both. Likewise, when you are trying to fix your forecast, you need to start by assessing the type of sales patterns you’re dealing with. The type of sales patterns you see will tell you to what extent your products are forecastable, and your forecasting requirements will become easy to formulate.

The key to overcoming forecasting problems is having a robust planning framework that assigns an optimal forecasting approach to each item.

Since many sales patterns are not suited to a purely statistical approach, the type of sales pattern that a product demonstrates should determine how it is forecasted. There’s no point in turning to a forecasting system for products that require manual control, just like it’s a waste of time to pound nails in with a screwdriver.

​Segmenting products according to sales patterns is the first step. Depending on how many items you have in each group, you need to make sure you have the resources available to create accurate forecasts. In 90% of cases, a combination of human assessment and statistical analysis must be used. The application and amount of weight to be given to either requires a closer look.

Consider the 3 types of forecasting tools: statistical analysis, market intelligence, and business logic.

To establish a robust planning framework, you need to decide which items should be managed using statistical analysis, which require market intelligence, and which require the application of business logic.

    • Statistical analysis. Predicting future sales based on an algorithm works when sales patterns for a product are stationary or stable, when a product shows consistent seasonality, or when sales demonstrate a clear trend. This mathematical approach still requires more than plugging in numbers. A forecast analyst must check over the forecast to ensure it makes sense. This expert must have an in-depth knowledge of statistical forecast modeling and know how and when to adjust statistical models.
    • Market intelligence.​ Market intelligence comes from the sales team. You need input from them regarding anything unusual that impacts your products. This type of input is needed to forecast products that are affected by changing sales patterns and promotion-induced variations. For example, if a product will be affected by a promotional campaign, you need the sales team to let you know so that you can add the variation you expect from the promotion to the baseline forecast.
    • Business logic.​ Market intelligence comes from the sales team. You need input from them regarding anything unusual that impacts your products. This type of input is needed to forecast products that are affected by changing sales patterns and promotion-induced variations. For example, if a product will be affected by a promotional campaign, you need the sales team to let you know so that you can add the variation you expect from the promotion to the baseline forecast. ​Business logic is the key to distinguishing randomness from patterns that are decipherable to experts, such as subtle seasonality, when looking at many products. This is a situation in which a forecast analyst will work with the sales team to understand the nature of product sales.

    By providing the right software, the right experts, and the right processes, Perito Consulting can help you solve your forecasting problems.

When you form a partnership with us, our team of forecasting experts will work with your sales team and our advanced software to ensure that you receive a world-class forecast every week. Having an accurate, actionable forecast each week will allow you to improve your customer service, boost efficiency, enjoy greater profitability, and spend your valuable time actually running your business.

Thanks for your time - Stephan