In a perfect world, your business forecasts would always be based on historical data. But even when there’s no such data available, forecasting is still possible — and valuable.

The challenges of forecasting with limited or no data are numerous. On the qualitative side, you might produce overly-optimistic forecasts that are influenced by personal agendas or a herd mentality at your company, while a dearth of data on the quantitative side could leave you with misleading mathematical models.

Currently, there are multiple approaches that will allow you to forecast with limited data, but they often fall short. Those include, but are not limited to:

  • Historical review
  • Test markets
  • Executive judgement
  • Diffusion modeling
  • Before-after trials
  • Simulation
  • Statistical/probability-based modeling

To make more accurate forecasts, you need to combine historical review with statistical/probability based modeling — a method we call forecasting by historical analogy. For instance, let’s say you want to forecast the demand for a brand-new product. Because it’s a new item, you haven’t collected any historical data. However, conducting a historical review of data that you’ve accumulated for a similar product will give you what you need for efficient forecasting.

So, what’s a similar product? Products that are analogous to one another have similar time-series patterns because they appeal to similar customer tastes, have comparable levels of competition in the market, or are part of the same local economic cycle. That means you can use one to forecast the other. Here’s the data you need to do so:

  • Product attributes for a prior product, as well as the new product
  • Predictions from your top executives
  • Historical data from prior product releases
  • New data as it becomes available

Collecting this data will help you decide which products should be part of your equivalence group, which is a group of analogous products. Once you’ve validated the data related to the items in the group, you can use that data to build a model to simulate the product or forecast how it will perform in the market, among other things.

For more on forecasting with limited historical data, below are slides from my talk at IBM Insight 2015.

Want to keep learning? Download our new study from Forrester about the tools and practices keeping companies on the forefront of data science.

Photo by FutUndBeidl / CC BY 2.0

Saba Dowlatshahi
Author
Saba Dowlatshahi

I am a data analyst at DataScience with a background in quantitative finance and statistics. I am intrigued by the complexity of forecasting and predictive analytics and my passion is to create a better understanding of cutting edge tools to improve business and marketing forecasting.