Understanding Forecasting Methods in an AI World

In this special guest feature, Geoff Birnes, SVP of Customer Engagement at Atrium, discusses how the CRM space is being completely altered by the introduction of machine learning, data analytics and advanced AI.

As a co-founder and SVP of Customer Engagement, Geoff is responsible for Atrium’s customer outcomes.

Geoff brings extensive experience in large scale business transformation programs across sales, marketing, service and middle office.

Geoff attended Penn State University where he earned a B.


in Engineering.

The CRM space is being completely altered by the introduction of machine learning, data analytics and advanced AI.

Traditionally, the world of CRM has been focused on data storage and minimal engagement, but now there is an opportunity to create full-scale systems of intelligence.

Organizations can take advantage of their data and gain valuable insights leading too precise, actionable forecasts.

There are two extremely powerful forecasting methods that businesses can use: top-down and bottom-up.

These approaches can be used individually or combined.

Both answer questions about the future but require different types of data and help answer different types of questions.



push(function() { googletag.

display(div-gpt-ad-1439400881943-0); }); Let’s look at the two forecasting methodologies and how to use them.

Top-Down vs.

Bottom-Up A top-down approach takes historical data to predict further into the future.

In other words, it leverages aggregate-level information to make top-level market forecasts.

It is not as helpful in predicting what will happen tomorrow, but it can be helpful for anticipating customer demand, allocating resources or being proactive about potential future difficulties.

The bottom-up method is more focused on specific instances.

It is sometimes referred to as a “rollup” forecast because each individual event is predicted and then combined.

Companies track their opportunities and gather tons of data along the way.

Through a CRM, this data can be leveraged to estimate predictions of winning deals and converting leads for each dataset.

This can help sales reps at the ground level to identify key factors and highlight actions that might help convert a deal.

More power comes when these insights are rolled-up to generate forecasts for larger groups or even the whole company.

Consider the differences in the two approaches through the lens of sports.

A top-down approach can predict how likely a sports team is to make the playoffs, while a bottom-up forecast can predict each individual game.

Implementing the Top-Down Approach  Popular machine learning models, such as neural networks, do not work well for top-down forecasting.

In fact, many traditional statistical methods can outperform high-tech algorithms.

This is because most machine learning models do not take into account autocorrelation, a phenomenon that says observations made at close points in time are more closely related than those made at more distant points in time.

For example, data from sales for the last year will better indicate future sales than the data from two years ago.

Therefore, when building a forecasting model you must take into account autocorrelation.

Unfortunately, too many statistical tools operate under the assumption that observations in time are independent.

As a result, a standard model on data with a time-based correlation can generate misleading results.

However, there are tools that can handle this type of data.

A few examples include exponential smoothing models, moving average models and smoothing splines.

Implementing the Bottom-Up Approach When implementing a bottom-up forecast, it starts with logistic regression models on historical data that estimate the probability of success for each record in the dataset.

To estimate the expected value for each record, combine the predicted propensity of success with the size of each deal, lead, and opportunity in the data.

Then, all the records can be aggregated in a combination of meaningful ways (i.


by team, region, etc.


The Benefits Not only can you generate forecasts, but you can also identify key factors that will help improve the forecasted outcome.

Through the forecasting work, you might discover that a seven percent discount in a specific region for a specific product increases the probability of a close much more than a six percent discount.

Simple insights like this can have a massive impact on the win rate for sales reps.

By learning these factors, organizations are able to invest time into increasing the likelihood of lower-probability deals to convert while still making precise forecasts around each deal’s expected value.

  Companies need to turn to well-developed statistical tools to power their forecasting models.

While forecasting doesn’t solve all problems, it does help businesses become data-centric and therefore make smarter choices about the future.

  Sign up for the free insideBIGDATA newsletter.


Leave a Reply