What is Demand Forecasting? A Comprehensive Guide

What does a good demand forecast look like?

The harsh truth is that forecasts, by their very nature, are at best, guesses. Naturally, the more data you have, the better those guesses are.

But there’s still an abundance of factors that can change the landscape.

Fail to highlight these factors and they’ll quickly thwart your forecast in front of your very eyes.

So how do the best businesses in the land forecast demand? How do they continually assess what their customers want and align their efforts to that end?

Here are a few pointers:

1. Continuous validation unpins the demand forecasting process

Market insights can take statistical forecasts based on historic sales data, and give them the gift of foresight.

Say you sell office equipment and sold out of laptop stands continuously over lockdown, does that mean you’ll continue to now?

Or does the fact people are returning to office life mean sales will slow down?

Historic sales data can give you a snapshot, but more often than not, it’s not the full picture. The most successful companies substantiate their forecasts with insights from the market.

By doing so, they instil a validation process. Done right, this will involve various teams and be a consensus of best practices.

2. It’s understandable and easily explainable

A demand forecast shouldn’t be so complex that no one understands it. Quite the opposite. It should be easy to understand by any of the stakeholders.

Further to this, it needs to be explainable. Easily explainable. Both the insights within it and the desired actions arising from it.

3. It should be recordable and measurable

Fail to record the accuracy (or rather the error) of your demand forecast and you’ll never know whether you were out by 0.5% or 50%.

And that’s important: 0.5% is a wild success. 50% is a total failure.

Knowing which one you were closer to will help the next time you have to create one. Which should be soon.

The more you measure your forecast, the better you’ll be able to tweak and make changes to it.

Errors you don’t know you’re making will be repeated again and again.

4. Every demand forecast should have an owner

Every forecast you create must have an owner. Someone who is accountable for its success and accuracy. Without an owner, it’s more likely to fall flat on its face than create a competitive advantage.

Remember, the decisions you make on the back of this forecast could be the difference between success and failure.

If you’re the one who’s making those decisions, you’ll take ultimate responsibility if they’re wrong.

How to build your own demand forecast

There are many different models you can contemplate when thinking about designing your own demand forecast. Naturally, there are pros and cons to each one. In some cases, combining them can give you the best results.

In the next section, we will explore some of the most common demand forecasting approaches.

Which demand forecasting models should you apply?

1. Time-series analysis

What can sales from the last 12 months predict for the upcoming year? That’s the premise of the time-series analysis.

As one of the most common methods of statistical demand forecasting, this is a handy one to get your head around.

It looks for patterns in data which might be replicated moving forward. It also allows you to see what might’ve influenced these patterns and make forecasts more accurate.

There’s more than one type of model in this analysis function too. There are moving averages, exponential smoothing, and autoregressive integrated moving average (ARIMA) models.

Each model uses a different method to analyse its data. Which one you need should be paid close attention to, to increase its potential success.

2. Judgement-based demand forecasting

How likely are the people in your company to make great predictions about future demand? The answer to that will depend on the strength of their capability and differ by company, team or even person.

One of the biggest challenges in effective demand forecasting is that the data can only tell you so much.

If you’re a new business, or releasing a new product in a new market, there’s almost certainly a deficiency of data you can rely on.

That’d make judgement based forecasting worth considering.

The same can be said about markets with high levels of disruption, given the limited ability of machine learning and statistical modelling in unknown areas.

If your demand forecasting approach ensures that market insights are actively gathered, you will be better informed and therefore have a higher chance of creating an accurate forecast.

To give yourself the best chance of succeeding in judgement-based modelling, you’ll need to research a huge number of areas.

That’d probably include market research, competitor analyses, speaking with customers, your sales team and even exterior market experts.

Considering this approach might help you analyse factors which aren’t picked up by other models.

But as with every other model, there are drawbacks to consider.

Are the people making these judgements totally impartial? Are they making decisions solely based on data or with prejudice from their own feelings about products or markets?

Equally, can you repeat this process? It’s often very difficult to produce the same numbers with different stakeholders, making judgement-based forecasting hard to replicate on a long-term basis.

3. Causal Analysis

Causal analysis identifies the cause-and-effect relationships that impact demand. It involves analysing data to show how changes such as price, promotions, or economic indicators, can change demand.

Causal analysis can be used to make both short-term and long-term forecasts and often involves the use of experiments to test the causal relationships between variables.

4. Regression analysis

Regression analysis explores the link between demand and multiple other variables. Whether that’s price, market share, environmental or market conditions.

This type of analysis allows you to assess how changing one of the variables might impact demand. And therefore optimise them. Which will help your supply chain planning.

What’s the difference between causal analysis and regression analysis?

One key difference between causal analysis and regression analysis is that causal analysis tends to focus on specific variables, while regression analysis takes a more comprehensive approach and considers multiple.

Another difference is that causal analysis often involves experimentation for the benefit of testing. Regression analysis is typically based on historical data and statistical modelling.

5. Machine learning

The capabilities of machine learning and AI is rather impressive.

Using these advanced techniques to analyse past data and make informed decisions about future demand could save you time and money.

Giving historical data to a machine learning platform can draw out patterns and make estimating future demand a lot easier.

Especially where more traditional methods might not cut the mustard.

If you’re in a business where the industry changes sharply and without warning, underpinning your forecast with machine learning could be incredibly useful. AI is defined as such because it continues to learn.

And so in theory, the accuracy of your forecast with machine learning should improve over time.

There are pitfalls of course. Not having enough data to analyse can result in a limited forecast. Relying too heavily on data which is incomplete or not accurate could result in failure.

How can you improve the accuracy of your demand forecasts?

Choosing the best forecasting model for you, will help make them more accurate, and therefore help you make solid, informed business decisions.

But even then, there are steps you can take to unlock rich insights and bring greater clarity to your business.

Our next section will focus on how you can further elevate your demand forecasts.

Aggregation vs disaggregation

The rise of data and its use in business is monumental. But as any Data Scientist will tell you, having the data and using it are two very different things.

We often talk about the importance of having specific stocking strategies for every SKU. It’s understandable then, you might think the same thinking could help you forecast, too.

Only that’s not always the case.

The below graph explains why aggregation can help in your forecast.

The reason aggregation can help improve your demand forecasting model is because of the power of larger data sets. The law of large numbers dictates the larger your sample size, the more representative it is of the wider population.

Taking sales data from one village will never relate to an entire country. But take that same data from 5 cities within the country and it’s far more likely to correlate.

Use find more data points and you will cut out the noise from anomalies. That means you’ll find more patterns and a greater ability to account for variables.

Likewise, imagine you are forecasting demand for a SKU that has recently been launched. You are unlikely to have a complete data set. However, if you aggregate the forecast from the category level, you are likely to attain a better idea of how demand will evolve for the SKU in question.

Here are a few common examples of how you can aggregate your demand plans:

A huge positive of using aggregation is that your data is less likely to be influenced by singular irregularities.

On the flip side to that, however, not paying attention to outliers which may ordinarily stand out, you may miss important data that could impact your business.

Decide to use aggregation on a case-by-case basis, and you won’t go far wrong.

How can disaggregation help you?

Sure, aggregating data can be very helpful. However, you will still need granular-level insights to understand the demand patterns for each product individually.

Typically the lowest level of granularity in demand forecasting refers to the disaggregating the demand forecast by the SKU location combination or even by customer.

The former can help you to optimise your allocation and replenishment to ensure you meet forecasted demand with the minimal level of stock.

The latter can help you identify which customers are most profitable, which have growth potential and which only account for a small proportion of forecasted demand.

And why should you care about this? Well, this invaluable insight will help ensure you invest your time, money and energy in the areas of your business or the customers that add the most value.

How can ‘time buckets’ create more reliable and robust demand forecasts?

Another method of aggregating your demand forecasts is by time.

Monthly forecasts tend to be more reliable than daily or weekly forecasts. The reason for that is they tend to be far less volatile. There may well be differences week to week or even day to day in how many products are sold, but over the course of a month, these are ironed out.

Here are a few reasons why demand forecasting in monthly buckets is more reliable:

1. Monthly forecasting’s larger bucket better absorbs customer volatility.

If someone normally orders from you in the first week of the month, but then orders in the second week, that’d massively disrupt your data if modelled by the week.

On a monthly basis, that deviance would be absorbed and your data just as useful.

2. Monthly forecasting can be more efficient

Creating weekly forecasts can be far more time-consuming and resource-intensive than creating monthly forecasts.

With monthly forecasts, you may need to gather and analyze data once a month, whereas with weekly forecasts, you may need to repeat the process four or five times a month, depending on the number of weeks in the month.

3. Monthly timeframes handle seasonality better.

Every company has slow weeks. Imagine you’re in a market that slows down for Christmas, the Chinese New Year period or perhaps even during the summer season.

In weekly forecasting models, these quiet weeks can produce gaps which have a greater influence over your demand forecast. Consequently, you end up with a far more nervous forecast. This in turn, is far more difficult to plan from.

Is there any use for weekly demand forecasting buckets?

Yes. There is. But you need to know when and how to use it effectively.

Weekly forecasting needs more effort than monthly, but where patterns arise that dictate irregular, but repetitive, sales activity depending on the week within the month, it’s worth thinking about.

Here’s a nice example:

As you can see in this image, 60% of sales come in the first week of the month.

By using data showing outcomes like this in a weekly forecasting model, you are likely to be more responsive and can take more proactive steps to align supply and demand more closely.. This means you can optimise inventory levels, improve revenue management and therefore profits.

Equally important uses for weekly forecasting models might be if you have a customer who also uses a weekly model. Increasing the link between you and them can be priceless in increasing accuracy and aligning both company’s efforts.

Demand forecasting with Slim4

There’s a lot of information in this article which will hopefully really help in your demand forecasting. But there’s a lot to consider.

Which is where Slimstock, and Slim4 can really help.

Do you find spreadsheets limiting?

Are you trying to anticipate the demand of hundreds or thousands of SKUs?

Are you attempting to plan multiple demand streams and only have a small team to help?

The Slim4 engine can help you out.

Find out how, by clicking here.