New Product Forecasting

Today, most companies involve themselves with a wide range of products. Competitive compulsions force companies to keep the product range fresh and relevant in the marketplace and hence introduce new products in high frequency than ever before. However, from the planning perspective, forecasting demand for the new products is one of the biggest challenges for the planning team. Accurate forecasting is absolutely critical to optimizes the resources invested in new products.

A lack of historical data is the biggest obstacle when it comes to forecasting demand for a new product. The good news is that most new products are truly a variation of an existing product. This gives planner a big clue on the expected behavior of the new version and which is likely to be same as the product it is replacing or a similar another product. This is true for almost all consumer products. When we see ‘new’ models of cars being launched or a ‘new’ soap or a ‘new’ laptop model. None of these are truly a new product but a new version of an existing product.

The method used to manage such new product introductions is called ‘Forecasting By Analogy’. As the name suggest in this method we pick an existing analog product (or a ‘alike’ product) to use its history for forecasting demand of the new product. Planamind Forecasting System has an Analogy Model feature which enables new product forecasting. Lets discuss in more detail on how to work with this approach.

Step 1 : Picking an analog product

The first and the most important thing is to pick the right existing product whose sales behaviour we expect the new product to emulate. For a company which has a large number of SKUs, this is a tough choice to make because if this goes wrong, everything else will too. We shall discuss about how to make use of other statistical techniques to help us with this choice.

Step 2 : Selecting the Date Range

We need to select the historical period range of the analog product that we would like the new product to emulate in its behavior. This is important as too long a history or too short a history may provide a different consumer behavior. For a technology products it is better to use shorter history.

Step 3 : Aligning Seasonality

Forecasting for seasonal products is more complex as the launch month is more significant than the actual launch. The launch month in this case needs to align with the season.  As opposed to  with the month in which the existing product was launched. The Align Seasonality option helps align the launch month to that of the similar product, thus overcoming this complexity.

Step 4 : Specified Launch Total

The Marketing team usually conducts a survey before product launch to predict product acceptance. The alternative is to use their judgement to arrive at initial launch quantity. Planamind helps demand planners incorporate the launch amount using “specified launch total” option. Once the demand planner keys in the launch amount, it is  distributed over the launch horizon. This is  based on the historical trend of similar item. In absence of a specified launch total the system will derive the launch total basis the historical data of the analog product within the selected date range.

Planamind screen for Forecasting By Analogy making it easy for Demand Planners

Selecting the right analog product

As discussed earlier, for a large product base organization it is often times tough to pick the right analog product from its wide range with minimal differentiation.  We can make use of other statistical techniques to help ourselves solve this dilemma. At Anamind we have proprietary Clustering Algorithm that we use to find answers to such problems.

In Clustering, we basically cluster similar products to pick the best fit or the closest relative. Let’s take an example of a new handbag (Red Sling Bag) being launched by a company to understand this approach.

The first step  is to Identify the attributes of the new product that we want to use in choosing the analog product. These parameters can be qualitative or quantitative in nature. For this example, we in case of a handbag the attributes could be price, material, color, size etc as parameters to find the similar product.

It is important to assign weightage to these parameters based on the business need and product. For example, price is an important factor for consumers so it may be given more weightage than other parameters

Identify the optimum number of clusters to be used. Too many clusters may fragment the options and too less may not make the cluster truly homogeneous. The good number is usually a relative judgement call. In the this example we identified 12 clusters as a good number from the total underlying data set of about 150 possible choices.

Identify the cluster for each product and then identify the closest existing product within the same cluster. In this example, Red Sling bag is a new product worth INR 25,000. The proxy identified for this product is Pink Slingbag, made of the same material, which costs around 20,000 . Other two options are Black Handbag and Brown Handbag.


BI screen of recommended proxy products

We know the ideal analogous product, what next ?

The new product is assumed to behave like the analogous product, thus it is important to monitor product behavior. Check forecast vs actual reports on a periodic basis. If the deviation is very high opt for a different analogous item from the list. The benefit of this approach is that you will get multiple best options to choose from.

By following the above steps the demand planner ends up with an accurate and reliable forecast for his new product.

You may write to us for understanding this approach better or if you need help in identifying the right analog product for your new products. Our email id is hello@anamind.com.

T-20 Offer – To get 20% off on New Product Forecasting Using Machine Learning, a certificate course click here (*offer valid for first 20 Users)

 

 

 

Comments

  1. Steven Miller says:

    This is a very well-written article I particularly am impressed with the advanced analytics you are using, i.e. k-means clustering, to help assist in finding products that are ‘close’ to the new product in n-dimensional Euclidean space.

    Thank you for writing this article and placing it into the literary discussion on such a difficult topic for demand planners

    1. rishi says:

      Thank you so much for the appreciation,Steve. Glad you enjoyed it.

Leave a Comment

Your email address will not be published. Required fields are marked *