How to Improve Your Sales Forecast with Ensemble Forecasting

By Pete Furseth

At the start of each quarter, the scramble begins. Sales reps make their commits. Sales ops is neck-deep in Excel spreadsheets. Your CFO expected the number before the end of last quarter.

As sales leaders, forecasting plays an important role in our lives, yet most of us do not take the time to measure our accuracy. If we did, we would quickly realize there is room to improve. The solution is not to find the one perfect method. The solution is to combine multiple techniques into one ensemble forecast.

An ensemble forecast is a group of forecasts viewed as one, with the goal of improving accuracy. It works like diversifying a stock portfolio. By holding multiple stocks, you reduce unsystematic risk. Sales forecasting is no different. If you make multiple forecasts and blend them together, you improve accuracy and reduce risk.

Five Forecasting Techniques Worth Using

Each of these techniques approaches the forecast from a different angle. That is the point. You want independent inputs that capture different signals in your business.

1. Resource Forecast

This technique is based on the number of sales reps you have, their assigned quota, their tenure, and seasonality in your business. You calculate it based on how much quota you have assigned and how much you expect to be attained this quarter.

The key inputs are team experience and expected sales ramp rates. A team with five new reps in their first six months will produce very different results than a team of tenured veterans.

Resource forecasts are useful because they are independent of pipeline data. They give you a top-down view of what your team should produce based on capacity alone.

2. Roll-Up Forecast

This is the traditional method found in most sales organizations. Each sales rep makes a commitment, each manager rolls up those commitments, and the numbers get combined into an overall forecast.

Roll-up forecasts are easy to understand and create ownership. The downside is they carry every bias your reps have: sandbagging, over-optimism, wishful thinking about that one big deal. They are a necessary input but should never be your only input.

3. Pipeline Forecast

This approach is based on the number of opportunities in your funnel and their characteristics. It might involve calculations using average deal size, expected win rate, average days to close, and other deal-level metrics. Or it might be a simple multiple, like a 3x pipeline coverage ratio.

Pipeline forecasts are useful because they are math-based. The weakness is that they assume your pipeline data is accurate and up to date, which is rarely the case without good CRM hygiene.

4. Time-Series Forecast

This forecast is based entirely on data you observe through time. It considers business trends, cyclical changes, seasonality, and random noise. It does not look at individual opportunities in your pipeline.

Time-series forecasts work surprisingly well for businesses with a high volume of deals. If you are not dependent on one or two large deals per quarter, the historical pattern of your business often provides a strong baseline estimate.

5. Opportunity Scoring Forecast

This approach is similar to roll-up forecasting, but instead of asking your sales team to make the commitments, you use an algorithm to determine which opportunities will win and which will lose. The appeal is its ability to remove human bias. No more sandbagging. No more over-commitments.

Opportunity scoring models typically use factors like sales stage, deal amount, time since last stage transition, customer type, and time until expected close date. For more on the specific attributes that matter, see our post on sales forecasting and predictive analytics.

How to Build Your Ensemble Forecast

Once you have two or more individual forecasts, you combine them.

Step 1: Start simple. Average your individual forecasts. Give each one equal weight. This alone will get you closer to the predicted outcome than any single method. Step 2: Measure each forecast independently. At the end of the quarter, compare each forecast to actual results. Which technique was closest? Which consistently over-forecasts or under-forecasts? Step 3: Assign weights based on accuracy. If your time-series forecast has been within 5% for three quarters and your roll-up is consistently off by 12%, give the time-series more weight. You can use a simple regression or machine learning models to determine the optimal weights. Step 4: Recalibrate quarterly. As your business changes, so do the relative strengths of each technique. A method that worked well during growth may lose accuracy during a market contraction. Keep tuning.

Why Ensemble Forecasting Works

The mathematical principle behind ensemble forecasting is error cancellation. Each individual forecast carries its own biases and errors. When you combine independent forecasts, the errors tend to offset each other.

A roll-up forecast might be optimistic because reps are chasing a SPIF. The time-series might be conservative because last quarter had a seasonal dip. The pipeline forecast might be right on for existing deals but miss intra-quarter arrivals. When you average them, the result lands closer to reality than any single input.

This is not theory. It is well-established in weather forecasting, financial modeling, and every field where predictions matter. Sales should be no different.

The Measurement Requirement

None of this works without measurement. You must:

1. Record each individual forecast at the start of the quarter 2. Track weekly revisions to spot drift 3. Compare each method to actuals at quarter end 4. Calculate the optimal weights for the next quarter 5. Repeat the process every quarter to continuously improve

For a framework on how to measure forecast accuracy and track it through time, start with our guide on measuring sales forecast accuracy.

Getting Started This Quarter

You do not need five techniques on day one. Start with two. Most organizations already have a roll-up forecast and can quickly add a pipeline-based calculation. That gives you two independent inputs to blend.

Over time, add a resource forecast based on quota capacity and a time-series model based on historical results. The fifth technique, opportunity scoring, requires predictive analytics tooling but delivers the highest marginal improvement because it eliminates the human bias that contaminates every other method.

The key is to get a process in place and measure the result of each forecast. Your accuracy will improve as the ensemble matures. The best forecasting organizations did not arrive there overnight. They built their process one technique at a time, measured relentlessly, and let the data tell them which inputs to trust.