Understanding Demand Forecasting

How much sales will my operation make next month? Is it enough to cover the payroll costs? How much should I order before my inventory runs out? All of these questions have a common denominator that we will talk about in this post: demand forecasting.

Demand forecasting is defined as the exercise of predicting future sale quantities. Of course, this is easier said than done, and as a matter of fact, you are doing it already (whether knowingly or not). Either by hunch (millions of years of evolution have endowed the human brain with the ability to make educated, and often wrong guesses about the future), or by some system (including computational algorithms that process historical orders and other data). Next, we will talk in detail about the different systems to do demand forecasting and explaining their pros and cons.

Method 1: Gut feeling

The easiest method, as the name indicates there is no other mechanism to guess the demand other than by feeling. This method is especially practical for a new business, as there is no really any historical data to make guesses on. So, this is the best method for beginners.

Pros:

-Easy to do (no extra setup needed).

-Cheapest method (but there is no free lunch)

Cons:

-Not sustainable over the long term or with a large number of items.

-Increased risk of missed sales or clients when the demand exceeds inventory stock.

-Operational uncertainty when resources (such as staff, supplies, materials) have to match the demand.

‍

Method 2: Historical Trend Analysis

As the name indicates, this method takes into account the historical trends of sales, and uses simple models to forecast the demand. Normally this can be done in excel or other spreadsheet software by aggregating the sales data daily or monthly and using the built-in functions to project a trend line for the sales. This method is more time consuming as it requires processing the sales data into a spreadsheet-friendly format.

Pros:

-More accurate than gut feeling.

-Available in most spreadsheet software (Excel, Google Sheets, etc).

-Easy to do once the data is in the right format.

-Allow to better allocate resources to the projected demand.

Cons:

-Potentially time consuming to set up (if data is not readily available/processed).

-Not scalable to hundreds/thousands of different products and larger time horizons.

-The method is insensitive to other side data (such as weather or other drivers of the demand).

‍

Method 3: Data-driven Models (Including Machine Learning and AI models)

In this method, all the available data is leveraged to produce estimates for the demand for different time horizons and thousands of products. This is done by feeding raw data to data-driven models and algorithms that learn the patterns that best fit the data, and then interpolate the likely trajectories of sales. Prediction can be as granular as desired (for example, hourly) and the models should be able to incorporate all kinds of side data that affects the demand (such as seasonality, product characteristics, consumer patterns and economic drivers).

Pros:

-When implemented right is the most accurate method.

-Can incorporate useful side-data (seasonality, product characteristics, economic drivers and so on).

-Allows to match resources (staff, supplies, materials) to the demand proactively.

-Scalable to larger prediction time horizons and number of products or different supply chains.

Cons:

-Potentially expensive (requires both a combination of expertise and resources to do it right).

-Difficult to distinguish top performing solutions from "snake-oil" ones (for example, AI models such as Large Language Models are bad at doing quantitative analysis).

-Needs a data infrastructure to feed the models (less of an issue nowadays as many point of sale systems allow API access to historical transaction data).

‍

Conclusion:

All in all, there is not a clear winner between the methods presented. The choice depends on the scale of your business operation and how sensitive is the profit to demand changes (for example, in retail businesses matching the supply of workers to the demand heavily affects the operational profit and can be the difference between a healthy and profitable business or a direct path to bankruptcy). If the resources allow a middle ground between sophistication and a practical approach is what's best for most businesses. That's one of the reasons why we developed Predictheus, a platform that gives your business the best of both worlds without having to hire a team of data scientist in a scalable, cost-efficient and no nonsense way.