The 6 practical challenges of revenue forecasting (and 8 solutions)

Revenue forecasting can get complicated. Fast.

But as a revenue operations professional, an accurate revenue forecast is one of the most strategically valuable deliverables you can produce for your organization.

So how do you hand-off complexity while still delivering something that’s accurate enough to be useful?

It all starts with being aware of the practical challenges involved with forecasting, and having a toolkit of solutions at-the-ready to overcome them.

Below are six of the most insidious challenges, followed by eight practical solutions you can implement today to keep your forecasts fresh and accurate.

1) Complexity creep

If, like us, you love data, you’ll be tempted to crunch all the numbers available to you in one go.

Don’t do this.

Complexity creep is real, and it’s dangerous.

Sometimes, it’s unavoidable. Those at the enterprise level, building bespoke contracts for big clients, will have to overcome the challenges of tracking the many unique elements of multiple deals.

But complexity invites errors (human or otherwise). When the stakes are high (and with revenue forecasting, they always are), a high-level, directionally accurate forecast is better than one that’s fancy, granular, and wrong.

2) Granular data

Data at the weekly level and below can be difficult to use in your forecasts.

Why?

First, the seasonal period. The number of weeks in a year is not only relatively large, it’s also a noninteger (precisely 52.1775 weeks).

Daily and sub-daily data are even more complex.

“Seasonal” patterns can start to show up in odd places. Sometimes, they’re fairly self-explanatory. A drop in sales around lunchtime, for example. But it’s easy to think you’re seeing signal in the noise, and granular data exacerbates this tendency.

3) Time series of counts with small values

Forecasting methods that use counts (e.g. X number of sales in a month), are prone to inaccuracies when those counts are low.

A small business selling a few high-ticket items in a month, for example, is likely to see wild swings in revenue. A month that misses sales target by just one or two sales, might see actual revenue that’s way off what was forecast.

We see a variation on this problem in businesses that sell big enterprise deals. One massive deal that’s forecast for the end of Q2, for example, can make or break revenue for the entire quarter. Should you include it, or should you not? 

Answering questions like these is a real practical challenge for businesses looking to produce accurate revenue forecasts.

4) Very long and very short time series

Most models typically don’t work well with either very long or very short time series.

In very long time series, there’s enough data that underlying mechanisms affecting price or sales behavior can change over time. This makes the model less predictive.

At the same time, for very short time series, you become limited on the number of things you can estimate with your forecast. If you only have a month’s worth of data, you need loads of data points throughout that month. (This becomes even more true if there’s a lot of noise in the data.)

5) Data scarcity, outliers, and missing values

A new sales rep forgets to log a field in the CRM… A data entry error creeps in… Mistakes happen, but accounting for them is difficult.

Missing values can introduce bias to your forecast, while outliers can throw your forecast off. But you can’t just remove or replace outlying data without considering why it occurred.

Data scarcity is a particular problem in startups. How do you forecast success (or failure) if you have no past sales data at all?

6) Manual processes & disconnected tools

Depending on your business’ maturity, and the maturity of your revenue operations function, you may or may not have a streamlined tech stack.

If you (or your sales people) are stuck with manual processes, human error in data entry is a “not ‘if’, but ‘when’” scenario.

Similarly, disconnected tools and sticky processes disincentivize reps from logging information well. They also make it harder for you to get your hands on the data when you need it.

This has knock-on consequences, because the best forecasts are the ones you update frequently.

Wondering how to do your own revenue forecast?

Check out John Lorenc's step-by-step guide to doing a revenue forecast.

8 Solutions to these challenges

1) Combine forecasting methods

It’s well-established that combining multiple forecasting methods makes for a more accurate forecast overall.

“The results have been virtually unanimous: combining multiple forecasts leads to increased forecast accuracy. In many cases, one can make dramatic performance improvements by simply averaging the forecasts.” - R.T. Clemen, ​​Combining forecasts: A review and annotated bibliography - International Journal of Forecasting (1989)

Using a bottom-up approach? Combine it with the top-down method. Incorporate macro market data. Sense-check your numbers against that data. This keeps you grounded in reality, and guards against getting lost in the weeds.

Plus, as Clemen suggests, simply averaging the results of multiple quantitative forecasts can result in much greater forecasting accuracy.

2) Predictive analytics

Don’t just look backward. Look forwards, too.

If performing a historical forecast, combine it with forward-looking, predictive forecasting methods. This gives you a fresh perspective on your forecast, and acts as a useful sense check.

Predictive analytics helps you do just that, and is handily built into many modern tools.

3) Prioritize your granularity

Don’t try to model every data point. Instead, ask which metrics affect decision-making the most in your business. Model these well, and frequently. The rest can be dealt with in a more light-touch manner.

4) Rolling re-forecasting

Break an overall forecasting effort down into smaller, more frequent cycles.

This makes each forecast a lighter lift. Rather than relying on a single, static forecast, you can also incorporate new information as soon as it comes in.

As an added benefit, by retouching your forecast more frequently, you can work-in improvements each time you revisit it. This gives you more iterations in the same unit of time, resulting in a much more useful forecast after, say, one full year.

5) Use decomposition techniques

If you’re working with a lot of data at an org that demands a highly technical forecasting approach, you might benefit from using decomposition techniques.

Time series decomposition breaks down data into its most basic parts: trend, seasonal patterns, and residual randomness. This will give you the confidence that you’re dealing with signal, not noise.

6) Aggregate data to higher levels

Forecast accuracy tends to increase as granularity decreases. So an easy way to increase the accuracy of your forecast when you don’t have much data available is to aggregate the data you do have. 

Apply forecasting methods to this aggregated data instead.

For example, if you don’t have much historical data at the monthly level, aggregate it to work at the quarterly level. This will allow you to create a sensible quarterly forecast.

7) Don’t ignore qualitative data

If you really don’t have much quantifiable data, don’t forget that you can still get a good forecast using qualitative data. Expert opinions, surveys, and scenario-based forecasts can offer strategic direction, even in the absence of quantifiable data.

8) Set up good data collection practices early

Decide early what data you need for a useful forecast. Then start snapshotting that data immediately. The sooner you do both these things, the larger the dataset you’ll be able to use in every forecast from that point onwards.

Even small amounts of data, collected consistently, will build you a usable and useful set of data over time. Invest time in building and rewarding a data-driven culture, where reps are bought-in to documenting data in the CRM.

Tabular summary of forecasting challenges and their solutions

Next step: 10 Metrics for better forecast accuracy