Forecasting Methods for Product Managers
Elements of Forecasting and Its Applications in Product Management
The word "forecasting" is often associated with the weather. However, it goes beyond just weather. Knowing what will happen in the future help us prevent disasters or prepare to reduce the impacts. An interesting thing about forecasting is that the more you know about the past better is your prediction about the future. The history of forecasting goes way back in time. The Egyptians used forecasting techniques to predict harvests based on the water level of the Nile River during the flooding season. Since then, forecasting techniques have slowly evolved and become an integral part of the business. They are used almost in every area of business. Finance uses forecasting in making investing and budget decisions. Operations use it for managing inventory, staffing, and business expansion. Sales and Marketing use it to analyze the market and opportunities. Similarly, product managers must master forecasting techniques to understand product dynamics, coordinate with various departments, and understand market trends.
With effective forecasting methods, product managers will be in a great position to understand customers, plan production, and deliver the right products to the market. In this article, I discuss popular forecasting methods available for product managers. Let's take a look.
Categories of Forecasting
Broadly, forecasting can be classified into three main categories based on the duration:
- Short-term forecasting: to take immediate actions, running experiments, quick improvements/fixes, delivering, or purchasing
- Medium-term forecasting: for warehouse, inventory, and material supply planning
- Long-term forecasting: strategic product planning, pivoting directions, maximizing returns, etc.
For instance, weather forecasting is done hourly (short-term), 7-days (medium-term), and 14-day or monthly (long-term).
To achieve greater accuracy in forecasting, the amount and quality of historical data play a vital role. Variability in forecasts increases when there is no or limited historical data.
Based on the availability of historical data, you can adopt one of the following methods:
- Qualitative Methods
- Time-series Methods
- Causal Methods
- Indirect Methods
Let's look at the above methods in detail below.
This method is generally used when there is not enough data for forecasting. Especially startups, when introducing a new product to a new market, it isn't easy to gather data initially. Customers' opinions and professional judgments become a primary source for forecasting. In addition to these sources, some product managers experiment with the following techniques.
1. Delphi Technique
Using Delphi Technique, you can establish a group opinion by surveying a panel of experts individually in a structured manner. It is based on the principle that a group's opinion is generally more accurate than those of individuals. This technique is generally administered by a facilitator who establishes several rounds of interactions with the experts by receiving their opinion and providing them with group opinions to refine further. The facilitator usually takes a mean or median score in the final round to converge toward a more accurate forecast. For product managers, the Delphi technique can be a good tool for understanding new markets.
2. Jury of Experts
This technique relies on the opinions of a selected group of experts or even individuals. Everyone in the group reaches a common consensus by discussing each other's views and reviewing various strengths and weaknesses. This method allows companies with great leadership to quickly set new directions or make key decisions. A very good example is an executive meeting convening top executives from various departments (Sales, Marketing, Finance, Product Management, Operations, etc.) to decide to pivot a business strategy. These meetings commonly forecast something quickly when there is no data. It suits well for small businesses. However, caution should be taken to avoid any biased opinions.
3. Sales-force Composite
The revenue forecast is the key activity for any business to operate. Without that, it is hard to establish an annual budget and understand the company's prospects. It is generally the starting point of forecasting in the business. The revenue-driven budget will help drive key decisions. Sales-force Composite is another forecasting technique that develops an overall company sales forecast by consolidating forecasts from each sales agent from their respective sales region or market segment. It's a bottom-up technique wherein the sales force gives their best opinion on sales trends so that top management can predict the revenue estimates for the future. As salespeople are closer to the market, this technique can help in accurate sales prediction. However, it requires salespeople to be skilled at forecasting. Using this technique, product managers can get a quick rough estimate on revenue forecasts.
4. Consumer Surveys
Consumer Surveys reveal future demand. They are simply direct interviews with potential customers. An interviewer will typically ask how many products consumers would purchase given several alternative price levels. Surveys can take place in any of the following methods:
- Complete Enumeration Method: This method works well when the consumer population is concentrated in one place. In this method, the interviewer will contact almost all potential product users to get probable individual demand (PD) and aggregates them to estimate the total probable demand (TPD). See below for a simple equation.
TPD = PD1 + PD2 + PD3 + PD4 + …… + PDn
- Sample Survey Method: When the target population under study is large, this method can be helpful. The total probable demand is estimated by surveying a sample size of consumers and using the below equation.
TPD = (CR * TCP * AEC) / CS
- CR is the Number of Reporting Consumers, TCP is the Total Consumer Population, AEC is the Average Expected Consumption from reporting consumers, and CS is the Number of Consumers Surveyed.
- End-use Method: This method helps to estimate the demand for inputs by consuming industries. The forecaster usually builds a schedule of probable aggregate future demand for inputs by consuming industries keeping the desirable norms of product consumption fixed. The buyer takes the burden of demand forecasting. This method works well for industries that supply goods in bulk to industries.
As the name implies, time-series methods help us forecast events over a period of time using simple numerical progression techniques. The best example is a 10-day weather forecast displaying a change in temperatures. Time-series methods work only on quantitative data and rely heavily on historical data for better estimations.
Following are different ways to implement time-series methods:
It is a non-statistical approach to forecasting out of eyeballing. Forecasting is based on the current period's sales or revenue plus x%. It is useful for sanity checks. An example is setting up a sales quota for next year, the current year's sales plus 20%.
2. Moving Averages
The moving average technique is a great way to smoothen spikes in temporal data and evaluate trends. Typically, it is performed by taking an average of N number of time periods and moving one time period ahead.
For example, to find out the moving average of monthly sales of year Y, you will estimate:
Avg(Jan, Feb, March), Avg(Feb, March, Apr), Avg(March, Apr, May).. and so on.
A major drawback to this technique is that it smoothens the seasonal variations too. However, it is useful for near-term sales forecasts.
3. Exponential Smoothing
This is similar to the above moving average technique. However, higher weight is given to the time periods where the forecast was closer to the actual numbers observed. It is useful to eliminate any unusual factors that affect forecasts.
4. Trend Projections
This is another popular method of time-series forecasting in which the "trend line or curve" is used as a guide. Most people refer to the trend line as a "best fit," which is very popular in linear regression analysis. Best fit can be represented in many forms of equations starting linear to multiple degrees of a polynomial equation. This is a go-to technique if the data present a good relationship with time.
Causal methods are a bit complex. Product managers will have to seek professional help to use these methods in forecasting. Appropriate statistical techniques are used, considering various factors such as social conditions, location, economy, environment, and other market dynamics. Below, let's discuss two common categories of causal methods.
1. Simple Regression Methods
This is similar to the trend projections. However, forecasting is driven by specific factors or variations of another entity. For example,
- Variation in product sales in correlation with advertising expenditure, pricing, regional economy, etc
- Sales of baby products depend on the number of births
- Sales of home appliances based on new home constructions
- Increase in vehicle traffic on roads leading to several jobs or an increase in income
The efficiency of forecasting using simple regression methods depends on the quality of the data, which is believed to influence the subject parameter.
2. Multiple Regression Methods
These methods are too complex as they use econometric models, meaning they are built on interdependent regression equations (typically more than one) that describes some sector of sales and profit activities. Forecasting relies on many different input factors. Companies with a very large product market may use these methods with the help of BIG data analytics.
Unlike direct forecasting methods, indirect methods use the forecasted or existing data of other similar products or services or competitors' data. The following techniques are commonly used in this category.
Suppose there are similar products or services with good sales and market history. In that case, the new product may have similar growth provided the product has similar target customers, benefits, and price points. This method is called "Forecasting by Analogy." A good example is forecasting the sales of a new Samsung smartphone, looking at the growth of Apple's new iPhone. Although forecasting is not accurate, it can provide a good rough estimate.
The best way to implement this approach is by using Boss Model. The Boss model is used in new product forecasting. This model is used when there is no historical demand for new products. It uses the demand for an existing product and applies it to a new product. The following equation is used in the calculation of the demand.
- F(t) is the probability of adoption at time t
- f(t) is the rate at which adoption is changing with respect to t
- N(t) is the number of adopters at time t
- m is the total number of consumers who will eventually adopt
- p is the coefficient of innovation
- q is the coefficient of imitation
Input-Output is an excellent forecasting method when the outputs of one industry produce the inputs for another industry. This type of forecasting is very helpful for industrial suppliers, Original Equipment Manufacturers (OEM), etc. For example, forecasting the sales of tires based on recent car sales, forecasting the sales of batteries based on the new sales of electric vehicles, etc.
Every business has a total addressable market (TAM). Most business forecasts are done considering the growth in market share from 0 to x%. Knowing how big the market is, how much can be penetrated, and how fast we can grow would be very helpful in revenue or business performance forecasts. Using the Top-down approach, forecasting starts with assessing the market as a whole and then factoring in all the parameters that determine a serviceable obtainable market (SOM) to forecast the revenue eventually. On the other hand, the Bottom-up approach builds up forecasts based on the data and assumptions for productivity and capacity constraints in sales, production, and distribution.
Well, that's a lot of information on forecasting. I hope it was worth discussing everything in one post. In predictive analytics, forecasting techniques play an instrumental role. As a product manager, understanding the concepts and experimenting with different techniques will surely help you a lot in predicting and planning things ahead of the curve. When data is available or can be gathered, my recommendation is to rely on quantitative techniques such as time series and causal to achieve higher accuracy in forecasting. Use indirect or qualitative methods for a new product or service to get a closer estimate.
For further reading, I highly recommend visiting the below link to get more information about this topic.
- Product Forecasting — Wikipedia
- "How to Choose the Right Forecasting Technique" by John C. Chambers
- "Forecasting Best Practices For Product Managers" by Kate Kurzawska
Thanks for reading! I hope you enjoyed reading this post. For any questions or concerns, please connect with me on Twitter or arjunken.com, or on LinkedIn.