Before we get into the details of different sales forecasting methods, I want to tell you a story!
But first… for those of you who are visual/auditory learners, you can check out our youtube video as another option:
My Sales Forecasting Story
10+ years ago, I got the request from our divisional president (he’s now the CEO of a major public company) to put together an advanced multivariate sales forecasting predictive model which would help predict our volume and sales.
The task fell to me (as the divisional FP&A director) and the vice president of the commercial team. We were fortunate to work with a brilliant data scientist (he had 2 PhDs from top-tier institutions) with serious computing support and a full team of resources behind him.
We worked HARD for about 3 months (on top of our day jobs!) with this data science team to create a sophisticated multi variable predictive forecasting model.
Gathering the data was the first step. It had to be clean, they needed the background, we had to normalize for some one-off items in the data.
We gathered and cleaned data back several years at very granular levels (this was a $2B revenue business unit!) daily sales, by customer (by product type, product category, by part, and on and on). This itself took about a month! After all, if you start with bad data, you won’t get a good model!
Now a lot of the heavy lifting shifted to the data science team. They would run these simulations that would test the correlation and interlinkages between our internal data, data we paid for from a 3rd party service that collected industry data, and over 250 different publicly available data points (fed funds rate, oil price, mortgage rates, T-bills).
All of this data was tested at different lags (i.e. does a change in cost of oil per barrel indicate a change in our sales to a given customer on a one month lag? Two months? Three months? And so on. Some of these simulations would take their supercomputers 24-48 hours to run!
We iterated, we worked tirelessly, the data science team taught us about these simulations and models and how they worked so we could better inform them.
We taught their team about our businesses, our products, and our customers so they could ask us better questions. We finally arrived at our model!
We brought the model to the leadership team. We gave them the background, taught them how it works and FINALLY showed them what the model predicted unconstrained demand would be for our products and what we would be able to satisfy. Effectively, the volume / revenue forecast for the next 4 quarters.
We waited eagerly for his response… “Wow, this is a lot of great work! This makes a ton of sense. Thank you for putting all the time into it. But … that number is too low, maybe we need another model”. He was (half) joking of course.
His job was always to push us harder and increase the business results, but he still set our target higher than what the model indicated!
I shared this because I think it is a funny example. But in reality, no single forecasting methodology is the best!
Which model that is used depends on your business model & product, the resources (and skill sets/abilities of those resources) available to forecast, the frequency of forecasting, what software or CRM your company uses, and many more factors!
Sales Forecasting Methods
I’m going to take you through the different forecasting methods, give you examples along the way. And quickly touch on some of the Pros & Cons of each.
It’s important to remember that with sales forecasts there is no overall ‘best’ forecasting method. The key is finding a forecasting method that fits your business model and gives you an accurate sales forecast to use to run your business.
Make sure to check out our article on why we revenue forecast to learn more about why accurate sales forecasts are so important.
Here is what we will cover today
Historical forecast (trending)
Bottoms-up / Input-based
Multiple Variables (Algorithm) … predictive analytics
Combination Historical/Trending + Inputs
Probability / Pipeline / Backlog Conversion
Historical Forecasting (Trend Forecast)
Historical sales forecasting is going to use your historical sales data to forecast what the future will look like.
A historical forecast will have some assumptions built into it and could be relatively simple or could get more complex. Here are a couple scenarios you might see…
Sales History plus growth
Company sold an average of $100 per month this year, they expect a 5% growth rate, so now they expect $105 per month this year
Do a linear regression on the last 18 months of historical data to predict the trend line of future sales.
Pros of historical forecasting:
Easy to establish a process using past sales data (historical data)
In a stable environment, it can have relatively good accuracy for fairly low effort
Easy to understand for sales managers & sales leaders
Cons of historical forecasting
Doesn’t take into account customer information / sales projections from sales
Puts a lot of weight on assumptions (i.e. how do you know there will be a 5% increase)
My two cents..
There is merit to using a historical sales data forecasting method, but not in isolation. Rather, use the historical as one input into your forecast!
You use this forecasting model when you are very limited on time and resources or when forecasting isn’t that important to your business (i.e. it doesn’t have a big downstream effect).
If you do choose a historical process for your sales forecasting model, make sure you are adjusting out one-time items and have sound assumptions.
Bottoms Up / Input-based / Intuitive Forecasting
A bottoms up forecast (or input based forecast) takes inputs from the sales (this could also be commercial / customer / marketing depending on your business) team and consolidates that to predict future sales.
In this model you would create a template that each sales person has to complete with their sales forecast, the FP&A team aggregates this together and, ‘voila’ a sales forecast!
Pros of Input Based Forecasts
Uses intelligence from the people in your company who are closest to the customer
When done correctly, they can better align on a customer or product basis than using history along
Cons of Input Based Forecasts
Customers have agendas. For instance, it might be in their best interest to over forecast their demand to make sure your company has plenty of inventory to ship to them
Sales people have bias! Some people want to beat their forecast, some people are overly pessimistic or optimistic. You have a big human element with a bottoms up sales forecasting method.
You must fully understand the length of the sales cycle, the timing of when sales pipeline opportunities close, etc. in order to have an accurate sales forecast.
My two cents…
Similar to historical forecasts, inputs can be valuable to a forecast, but not when unchecked … and not when the only input into a forecast.
It is great to have market intel (which is essentially what sales reps and sales leaders provide) should often be included in some form!
However, this shouldn’t be used unchecked, but in a business with a heavy customized product it might make sense to rely on the inputs from the customer.
Multivariable Analysis Forecasting
Let us step into the world of data analytics and data science! A multiple variable forecast which is also called a predictive model or people may reference it as an algorithm, or some other name.
This is the type of sales forecasting method I mentioned in the introduction!
What this is, is a sophisticated forecasting method which incorporates several different inputs into a model / algorithm which generates a sales forecasts regularly (it usually also offers upper & lower bands on that forecast).
The inputs can be wide ranging from historical data, past performance data, customer/company information, various industry data (both public and 3rd part aggregated), different indices (inflation metrics like CPI, housing starts, oil or other commodity prices, mortgage rates, 10-year T-Bill … this list goes on and on), length of sales cycle, market research data, trend analysis, and more.
Since these multivariate forecasts require some type of software, they run iterations over different time lags, incorporate correlations between the different inputs, and MUCH more!
Pros of Multivariable analysis forecasting
These are sophisticated models and can often result in the most accurate forecast
Usually multivariate forecasts do a good job limiting human bias and discretionary inputs
Because they are ran on software, they typically give data at a granular level of data … and can be helpful at pointing out anomalies by customer/product to drive action in the commercial team
Cons of Multivariable analysis forecasting
Time intensive to create and require data science / data analytics skill set
Maintenance is required, these models need to be continually monitored for accuracy, inputs need updated, models need to be tweaked and changed as the world changes (think about what happened to many businesses when Covid-19 hit!)
Because the sales team doesn’t have much in the way of input into these models, they often don’t feel as ‘bought in’ to the forecast. It is a hurdle to making them feel like they need to hit the sales target.
My two cents
These models can be quite costly to create and to maintain. You also need to make sure you have the information at the right levels so you are able to talk to variances from the forecast.
This is something that a company needs to think through and understand the costs and benefits (and skill sets within the company).
Combination Sales Forecasting – Historical (Trending) + Bottoms Up Forecast
Spoiler here, this is the method I’ve most often used in my career (especially when we don’t have the need or resources to do a multi variable forecast).
This combines historical data, trending information, adjusts for one-offs, and adds in some intelligence from the sales team to come to a number everyone agrees to (or more likely, everyone is mutually upset about!)
The reality is, this is kind of a melting pot of methods, even beyond historical and bottoms-up. Effectively, you are combining the methodologies to come up with something that works for your business.
The way this works is YOU the fearless commercial FP&A person is going to have a process when you are gathering data, cleaning it up (removing/adjusting for one-offs), layering on assumption(s) and creating a view at what volume / revenue should be.
This view of the forecast goes to the sales team and they then make (or request to make) any adjustments they have based on their customer/product specific knowledge.
You may also have to bring in the S&OP, Ops, or Procurement team to make sure you will be able to support the demand). We as finance / FP&A gatekeepers need to make sure these changes make sense, we often have to push back on these or at least get a better understanding!
Pros to a Combination Historical + Bottoms Up Forecast
Combining these methods can capture historical data and market trends and intelligence in a fairly time and cost effective manner
The key people all have a hand in this forecast which means the organization will be more aligned, and given the sales team involvement, they should feel more accountable to it.
Cons of a Combination Historical + Bottoms Up Forecast
If you are doing this method, you are likely doing it in excel or a simplified software and it won’t have the sophistication of a multiple variable forecast (and you will likely have more discretionary inputs)
This method requires some back and forth with the sales team to come to an agreement.
These can also get a little complicated if you have to layer on new business wins, new product roll-outs, new store openings or store closings, and so on.
My Two Cents…
As I mentioned above, in the absence of resources for something more sophisticated, I have defaulted to this methodology many times in my career.
I think it does a good job getting buy-in, driving accountability, and has a decent balance of speed and accuracy.
Probability / Pipeline Forecasting (Backlog Conversion)
For businesses that have larger dollar products, custom products, or sometimes service based businesses, you might work off of a pipeline or backlog type forecast. (You may also hear this called opportunity stage forecasting.
This method takes a population of completed deals, opportunities at various stages, likelihood for deals to close and timing from a completed deal to actual revenue, and turns it into a forecast!
Using a CRM of some kind to facilitate this forecast makes a probability forecast a lot easier. When doing this manually it is either very subjective or tricky to come up with a real process for calculating the likelihood of a deal to close.
I also am lumping in ‘backlog conversion’ forecasting here. Some businesses are very long cycle and the order book is complete months (sometimes even years) in advance.
One example I have seen would be a company that sells buses. They have an order book of 6 months, so what they sell in the month would rely more on what they are able to make and ship vs. what demand is!
Pros of a Probability / Pipeline Forecasting
Once you develop the process of tracking and updating the pipeline, this method can be quick and have less resource requirement
If you have a large history of good data and a CRM it can be an accurate way of forecasting.
If backlog based, it removes subjectivity and puts execution in your own control
Cons of a Probability / Pipeline Forecasting
These are difficult to run without a CRM
The CRM needs to have a very strong process to update regularly, the sales team needs to be engaged
This only works for certain business models
Bias still exists … Sales people come up with dates and probabilities!
My two cents …
This can work well if you have a business model that makes sense! You have to develop a good process and it’s best to do only if you have a CRM
SAAS Sales forecasting
Software as a Service businesses typically operate on a monthly recurring revenue basis. Their revenue is largely driven by recurring sales to their customer base.
In a given month they will have recurring revenue, some new subscribers, and some ‘churn’ of existing subscribers.
There are multiple methods (many similar to the list above) that can be used to forecast SAAS revenue (and often a combination of methods is used).
Historical / Trending
MRR … new customers, churn (lost) customers
… the new customers may be identified from the pipeline OR pipeline forecasts
Sales input / Bottoms-up
I wanted to call it out separate as the models you will see for SAAS are usually a bit different than many other business models!
Sales Forecasting Wrap-up
This is NOT an exhaustive list. You can combine multiple methods and create new ones. The right method is going to be highly dependent on your business, your resources, skill sets, and more.
You also don’t set a method and forget it, you have to maintain it, monitor accuracy, refine and iterate!
Spend some time thinking about what would work best for your business, start the discussion with the different stakeholders of your business about how you might create or refine your process.
Also, start playing with your data a bit and thinking about process steps.
Next, we will cover the steps to actually build the revenue forecast, how to build a forecasting culture, and how to monitor and continuously improve your forecast.