How to forecast new product sales?

New product forecasting is a critical process that helps businesses estimate the future demand for a new product or service. Accurate forecasting is essential for marketers and various business activities, including production planning, inventory management, and financial planning.

In many industries responsibility for these estimates lies on the shoulders of the brand or marketing manager. Many times, especially for innovative products, this can be an overwhelming task for a brand manager as a lot depends on him to get as close to the actual demand as possible. Gross underestimation leads you to a situation with significant sales loss while overestimation will land you with a pile of inventory to write-off.

Thanks to current computing technologies like AI & machine learning, forecasting has become a relatively easy task as computer codes can crunch vast amounts of data at great speed.

Although the task of calculation is simplified the logic behind these analytical models comes from core mathematical equations, thus lack of understanding of the concepts might render all these models useless.

One of the simple and powerful equations that is very commonly used across industries is The Bass Diffusion Model, developed by Frank M. Bass in 1969, is a mathematical model used to predict the adoption and diffusion of new products or innovations within a market.

The Bass Diffusion Model is based on two types of adopters within a population:

1. Innovators (Pioneers): These are the first individuals or organizations to adopt a new product or innovation. They are typically risk-takers and are motivated by the desire to be among the first to try something new.

2. Imitators (Followers): These adopters are influenced by the actions of the innovators. They may adopt the new product because they see others doing so or because they perceive value in it.

Let’s try to put this with a simple example, imagine you want to predict how many kids in your school will start playing with a brand new toy. The Bass Diffusion Model can help with that, and it has two important numbers you need to know:

Innovators (p): Think of innovators as the very first kids who get the new toy. They're like the trendsetters in your school. So, "p" tells us how many of these trendsetter kids will start playing with the toy because they just love trying new things. It's like the cool kids who always have the latest gadgets.

Imitators (q): Now, think of imitators as the kids who start playing with the toy because they see other kids having fun with it. They're not the first to try it, but they follow the trend. "q" tells us how many of these kids will start playing with the toy because they see their friends enjoying it. It's like when your friends get excited about a new game, and you want to try it too.

So, when we use the Bass model, we use these two numbers, "p" and "q," to guess how many kids will play with the new toy. We start with the innovators, and as more kids see them having fun, they become imitators. Together, they make up the total number of kids who will play with the toy over time.

By knowing "p" and "q," we can make a good guess about how many kids in your school will end up playing with the new toy, even before it becomes really popular. It's like making predictions about how fun things will catch on with your friends!

In the above examples imagine you want to predict how many kids will start playing with a cool new toy at your school over time. The Bass Model helps us do that with this simple equation:

N(t) = p (1 - e^(- (p+q) t)) / (1 + (q/p) e^(- (p+q) t))

Now, let's break it down:

N(t) tells us the number of kids who have the toy at a specific time, like after a certain number of days or months.

"p" represents the kids who are super excited to try new things. They're the first to get the toy, and they start playing with it right away.

"q" represents the kids who see their friends having fun with the toy and decide to get it too. They follow the trend and join in.

The "t" in the equation represents time. It's like counting the days or months after the toy was introduced.

Here's what the equation does:

It calculates how many kids, the innovators (the "p" group), start playing with the toy immediately when it's introduced. This part of the equation is p.

It calculates how many kids, the imitators (the "q" group), start playing with the toy because they see their friends having fun. This part of the equation grows over time as more kids adopt the toy. It's the second part of the equation with q, t, e^(- (p+q) * t), and some math magic to make it work.

So, by using this equation with "p" and "q," we can make predictions about how many kids in your school will have that new toy as time goes by. It helps us understand how things become popular over time!

The term e^(- (p+q) * t) represents how the adoption curve changes over time, showing how the number of patients using the medication grows.

The letter "e" represents the mathematical constant called Euler's number, which is approximately equal to 2.71828. Euler's number is an important mathematical constant that appears in many areas of mathematics, including calculus and exponential growth

By using this equation with "p" and "q," we can make predictions about how many patients will be using the new medication as time goes by. It helps us understand how a new medication becomes popular and widely adopted in the market.

I hope this explains the concept to some degree and I agree a lot can go wrong before arriving at your numbers as many assumptions need consideration that vary from industry to industry.

Being aware of certain common mistakes can help improve the accuracy and reliability of the modeling results. Here are some common mistakes made when conducting Bass modeling:

Assuming Linearity: One common mistake is assuming that adoption follows a linear trend. In reality, adoption often follows an S-shaped curve, as reflected in the Bass model. Failing to account for this curve can lead to inaccurate forecasts.

Overfitting: Overfitting occurs when the model is too complex and fits the historical data points very closely. While this may seem like a good fit, it can result in poor out-of-sample performance. It's important to strike a balance between model complexity and predictive accuracy.

Ignoring External Factors: The Bass model assumes that adoption is driven solely by internal factors (innovators and imitators) and doesn't consider external factors like marketing campaigns, competitor actions, or regulatory changes. Ignoring these external factors can lead to inaccuracies in the model.

Inadequate Data Quality: Bass models require historical adoption data to estimate parameters accurately. Using incomplete, inaccurate, or biased data can lead to unreliable forecasts.

Inconsistent Time Periods: Ensure that the time intervals in your data are consistent. Mixing monthly, quarterly, or annual data can introduce errors into the model.

Inappropriate Parameter Estimation: Estimating the parameters (p and q) of the Bass model incorrectly can lead to unreliable forecasts. Common methods of estimation include nonlinear regression, but choosing the wrong method or not considering initial parameter estimates can be problematic.

Incorrect Segmentation: If your market can be segmented based on different adoption rates or patterns, failing to do so can lead to poor forecasting. Segmentation can be based on demographic factors, geographic regions, or other relevant criteria.

Not Validating the Model: Failing to validate the Bass model's performance with out-of-sample data or historical data not used for parameter estimation can lead to overconfidence in the model's accuracy.

Assuming Constant Parameters: Some practitioners make the mistake of assuming that the parameters (p and q) of the Bass model remain constant over time. In reality, these parameters can change due to various factors, and models should be updated accordingly.

Ignoring Competitive Dynamics: In competitive markets, the presence of competing products can influence adoption rates. Ignoring these competitive dynamics can lead to inaccurate forecasts.

Lack of Sensitivity Analysis: Failing to conduct sensitivity analysis to assess how changes in parameter values or assumptions affect the forecasts can result in overly optimistic or pessimistic predictions.

Not Considering Saturation: The Bass model assumes that the market eventually saturates, meaning that almost all potential adopters will have adopted the product. Failing to account for saturation can lead to overly optimistic long-term forecasts.

To avoid these mistakes, it's essential to carefully consider the assumptions and limitations of the Bass model, validate the model's performance, and incorporate real-world context and external factors when appropriate. Additionally, consulting with domain experts and conducting sensitivity analyses can enhance the reliability of the forecasts.