Optimal Sales Forecasting Method for B2B

Summary

• 16 different methods of sales forecasting and empirical research studies related to B2B sales forecast accuracy and sales improvement are reviewed and referenced.

• The most useful method for B2B is Pipeline Forecasting provided individual deals can be analysed and their win probability calculated methodically. 

• Most CRMs use the Opportunity Stage Method to do this but this has significant limitations with no coaching benefits. 

• Proven methods of improving deal win probability accuracy are discussed and the design of an intuitive, proactive and highly accurate universal Pipeline Forecasting  system is shown.

---

How to improve B2B Sales Performance

Our own experience of facilitating hundreds of B2B account planning workshops worldwide (for IBM, HP, Microsoft, Cisco and others)  has given us a wealth of data to understand the vagaries of B2B sales forecasting and the levers to outperformance. 

Statistical analysis of that data shows that if the sales rep / account team is made aware of gaps in their knowledge of the prospect’s reality (challenges, priorities, projects, people) and takes actions to address the knowledge gaps, then that alone is the biggest contributor to an average 5x revenue outperformance and reduced sales cycle¹ 

Why? 

There are many possible reasons, for example; 

• showing an interest in the prospect’s situation;
• getting the prospect to talk about their concerns and priorities rather than selling to them;
• automatically presenting your solution in terms of the prospect’s priorities and projects;
• defining a collaborative solution; 
• prioritising opportunities with a better fit;

Whatever the reason in each case, the boost to Return On Sales Time (ROST) is undeniable.

This insight led us to investigate how simple qualification questions can be used to reveal these knowledge gaps and simultaneously predict deal win probability with a high accuracy. 

The questions are designed by the sales team, based on consolidated experience and weighted using a learning algorithm. 

By making this process simple and fast with instant feedback, such a deal qualification & forecasting tool is no longer the preserve of large companies and large accounts but can be applied to B2B sales of any size.

Sixteen Methods of Sales Forecasting

Top-down estimates of the expected future revenues of a company can be made by analysing historical data and projecting this forward using various techniques (quantitative methods). Alternatively you can use judgement or opinions (qualitative methods). 

These estimates can be made by region or business unit and then summed to form the total.

16 Different methods of Sales forecasting are described below;

Qualitative: 

1. Market Research

This method analyzes, competitor activities, market trends and customer behaviors to predict sales. Complex and time-consuming but can provides valuable insights that help align new product offerings with market demand.

2. Expert opinion

This method requires the inputs of industry professionals or experts within the company for their insights into future sales. They consider market conditions, trends, and other relevant factors to generate a forecast. It is useful when launching new products or entering new markets where historical data is not available.

3. Delphi method

A group of experts anonymously answer a series of questionnaires, and the responses are aggregated and shared with the group after each round. The process repeats until a consensus is reached. It reduces the bias of any single opinion and leverages the collective wisdom of the group.

4. Salesforce composite method

Here, individual salespeople estimate their own sales based on their understanding of their territories and customers. These forecasts are then aggregated to form a company-wide sales forecast. Generally the quantitative version of this – pipeline forecasting – is preferred.

5. Buyers expectations

This method directly involves the customers, asking them about their purchase intentions through surveys or interviews. It provides first-hand insights into customer thinking and plans, but it can also be time-consuming and depends on customers’ willingness to participate.

Quantitative:

6. Time series analysis

Time series analysis uses historical sales data to identify trends, cycles, and seasonal patterns that are likely to repeat in the future. It is a popular and often accurate method, provided that market conditions remain relatively constant.

This method analyzes the sequential order and timing of past sales data to uncover underlying patterns and relationships. It employs statistical techniques such as smoothing, decomposition, and forecasting models to capture and extrapolate trends. By understanding the historical behavior of sales over time, you can make informed forecasts for future periods. 

7. Causal models

Causal models predict sales based on cause-and-effect relationships between sales and independent variables. Also known as explanatory models, they’re useful for understanding the impact of specific factors on sales. 

These models use techniques like regression analysis to identify how changes in variables like marketing spend, pricing, or economic indicators impact sales. They provide insights into the drivers of sales performance to assess the impact of different strategies or external factors. 

These models forecast sales based on the relationships between sales and one or more independent variables, such as advertising spend or price changes. The challenge with these models is identifying and accurately measuring the variables that impact sales.

8. Regression analysis

Regression analysis is a statistical technique that identifies the relationship between a dependent variable (sales) and one or more independent variables (e.g., price, marketing spends). It helps you understand how changes in the independent variables affect the dependent variable. 

By analyzing historical data and applying regression analysis, you can develop regression models that can be used to predict future sales based on changes in the independent variables. This method can account for multiple factors simultaneously and provides insight into their relative impacts on sales.

9. Length of sales cycle

This approach forecasts sales based on the average time to close a deal. The length of the sales cycle refers to the time it takes for a sales opportunity to move through the entire sales process, from the initial contact with a potential customer to closing the deal. 

The length of the sales cycle varies significantly depending on the complexity of the product or service, the industry, the target market, and the sales approach employed. It may range from days to months or even longer for larger enterprise sales.

Understanding the length of the sales cycle helps you estimate the timing of revenue inflow, plan resource allocation, and track the progress of opportunities in the pipeline. 

10. Pipeline forecasting

Pipeline forecasting uses current opportunities at various stages in the sales pipeline to predict future sales. It looks at each deal in a company’s sales pipeline and estimates the likelihood of its closure. By combining these individual predictions, you can generate a forecast for total sales. 

This method considers factors such as the number of deals in each stage, historical conversion rates, and the average deal size. It provides visibility into the potential revenue that can be generated from the existing opportunities in the pipeline. However, the accuracy of this method depends heavily on the accuracy of estimating closure probabilities.

11. Opportunity stages

Opportunity stages are key milestones within the sales process that represent the progression of potential customers toward a successful sale. It assigns a probability of closure to each stage of the sales pipeline and then multiplies these probabilities by the value of the opportunities at each stage to forecast sales. 

Here, each stage signifies a specific level of qualification, engagement, and advancement in the buyer’s journey. Common stages include prospecting, qualification, needs analysis, proposal/presentation, negotiation, and closed/won. The main challenge with this method is accurately determining the probabilities.

12. Multi-variable analysis

Multivariable analysis uses multiple variables or factors, both independent and dependent in forecasting. It uses software to analyze multiple variables simultaneously and predict future sales. This method can capture complex relationships among variables and is often more accurate than methods that consider variables in isolation.

This method considers various independent variables, such as marketing spend, pricing, customer demographics, and competitive factors, to determine their combined impact on the dependent variable, which is typically sales.  It allows you to understand how changes in multiple factors simultaneously influence sales.

13. Historical forecasting

Historical forecasting uses historical sales data to predict future sales performance. Meaning, it simply assumes that future sales will follow the same pattern as past sales. But, it does so by analyzing patterns, trends, and seasonality in past sales data, to extrapolate historical trends to forecast future sales. 

This method is straightforward to implement and useful when historical data is abundant or when market conditions remain relatively stable. However, it assumes that the future will closely resemble the past, disregarding potential changes or disruptions. 

While historical forecasting provides a baseline estimate, it is better to complement it with other forecasting techniques and consider external factors for a more comprehensive and accurate sales forecast.

14. Lead-driven forecasting

Lead-driven forecasting uses the quantity (number) and quality of leads to predict future sales. It analyzes the historical data on lead generation and conversion rates to estimate future sales based on the number and quality of leads in the pipeline. 

This method tracks metrics like lead volume, lead sources, lead quality, and conversion rates at each stage of the sales process. By understanding conversion rates and average deal sizes of different lead types, you can forecast future sales by projecting lead quantities and their expected conversion rates. 

15. Intuitive forecasting

Sometimes, experienced salespeople or executives rely on their gut feeling to predict sales. Yes, you read that right. 

Intuitive forecasting method relies on the experience, judgment, and intuition of individuals to predict future sales. Instead of relying on data and statistical models, this approach involves tapping into the expertise and insights of seasoned sales professionals or executives who have a deep understanding of the market and customer behavior.

You can use this method when there’s a lack of historical data or when external factors are too uncertain for other methods. While it may lack the precision of data-driven approaches, it can provide valuable insights and quick estimations in situations where relying solely on data is not feasible or practical.

16. Test-market analysis forecasting

Test-market analysis involves introducing a product in a test market before a full-scale rollout. It conducts a controlled trial of a product or service in a specific market segment to gather data which is later used to predict sales in the broader market. 

Businesses select a representative sample of the target market, implement marketing strategies, and monitor key metrics, such as sales volume and customer feedback. All of this data is used to make informed forecasts about the potential success of the product.

Test-market analysis provides accurate forecasting and reduces risk before a full-scale launch. However, it can be expensive and time-consuming.

The primary requirement of a forecasting method is accuracy and a simple, robust method. Whilst multivariable analysis is often considered to be the most accurate, this requires a large amount of data concerning many variables and expensive tools and resources. 

Where sales are the sum of individual deals which are the responsibility of individual sales reps, then bottom-up methods of forecasting are used; wherever possible, pipeline forecasting is the preferred method for B2B sales teams.

Deal Analysis in Pipeline Forecasting

In pipeline forecasting each individual deal is analysed and its size, close date and win probability is estimated. This has the added advantage of enabling manual or automated coaching and feedback to identify issues blocking success for a specific deal. 

For these reasons, Pipeline Forecasting is used by most successful companies for high value, complex sales. The accuracy and utility of this approach is in turn dependent on the accuracy of each deal win probability estimate. 

If you have a large number of low value deals then analysis of each one is not tenable, so less accurate methods of aggregate forecasting are usually used. Our proposed optimal B2B forecasting method overcomes this issue.

Measurement of B2B forecast accuracy

In statistical analysis, a set of probabilistic predictions for binary outcomes (e.g. win/loss) is generally evaluated using the Brier Skill Score (BSS);

“A BSS value of zero means that the score for the predictions is merely as good as that of a set of baseline predictions, while a skill score value of one (100%) represents the best possible score. A skill score value less than zero means that the performance is even worse than that of the baseline predictions.” (Wikipedia)

Comparing forecast systems using BSS is not without its limitations but is the generally accepted method.² 

Opportunity Stage Method of Sales Forecasting

Each deal can be mapped to the company’s standard sales process, expressed as a sequential series of deal stages. Ideally this sales process is aligned with the customer ‘s buying process. The sales rep is able to judge what steps have or have not been completed. 

The deal win probability increases as the deal progresses through the stages, so this progress can be used as a crude proxy for win probability.

Most CRMs (Hubspot, Salesforce, Pipedrive, Zoho, etc) calculate deal win probability based on the current deal stage (Opportunity Stage Method). 

The accuracy of this method is limited (BSS 30-50%), since it cannot distinguish between “time-wasters” and truly engaged prospects.

Focusing solely on this measure of progress of any deal also ignores the opportunity to express a qualitative judgement on the depth of understanding of the prospect’s reality, our fit to the prospect’s buying process and the level of engagement. The valuable deal-specific coaching opportunity is thus lost.

Objective vs Subjective Forecasting

In practice, most B2B companies rely on management judgement to drive the sales forecast. These subjective estimates whether at the aggregate or deal level are subject to compounding error and bias. They have been shown to be generally no more accurate than simply copy and pasting the prior actual results (implying a BSS of 0).³

Various tools can be used to inject objective rather than subjective judgement.

Lead Qualification and Forecast Accuracy

Lead Qualification Models have been developed to provide an objective score for a lead as it moves through the sales funnel.  The score is primarily used to remove poorly qualified leads from the sales funnel.

At the top of the funnel there is a wide & shallow data set available to calculate the parameters of the model. Many companies will use such data-driven models to score their “Marketing Qualified Leads” (MQL) for example.⁴

Provided there is enough data to learn from, machine learning can fine tune these models and significantly improve sales performance by qualifying out poorer prospects.  Such models do not influence forecast accuracy however.⁵

Deal Qualification and Forecast Accuracy

Qualification in the later deal stages can have a significant effect on forecast accuracy (Duncan et al). Here the scope of the qualification model is to determine directly the likelihood of a win. Deal qualification can therefore be used to estimate the deal win probability.

Coaching and learning is also enabled if the qualification scoring method is transparent and known to be both objective and an accurate predictor of deal win probability. If I know that; 

1. By verifying X or doing Y, I will increase the deal win probability and 
2. this calculated probability is an accurate prediction of my sales success

…then I will do it, whether I have attended a training course about it or not.

EBQ: the Gold Standard in Deal Forecasting

The gold standard in deal forecasting is to use a consistent set of deal qualification criteria each scored on a scale of 1-5, ideally using Evidence Based Qualification (EBQ)⁶ to inject objectivity. 

Qualification methods used by world-leading sales organisations include BANT, MEDDICC, SCOTSMAN, FACT, but any sales organisation can and should define their own simple set of qualification criteria based on the experience of their own sales teams. 

The qualification criteria become a powerful coaching tool as they indicate gaps in the sales rep’s knowledge about the prospect and thus indicate a course of action to fill these gaps.

Continuous Improvement of Sales Forecasting Accuracy

A common mistake is to assume that each qualification question is of equal importance. The best approach is to use machine learning to calculate the weighting to give to each answer.  

This can be done iteratively after each and every win or loss, recalculating the weighting to give the most accurate predictive value to the overall deal score. The more win and loss data there is, the better the forecast accuracy. The accuracy automatically improves over time.

Surprisingly little research has been done in this area, but various mathematical models have been tested for this iterative learning in relation to sales qualification (neural networks, regression tree, etc). Bayesian networks have been shown to have “distinct advantages over the other methods in accuracy of prediction, transparency of procedures, interpretability of results, and explanatory insight.”⁷

A further advantage of Bayesian methods is that they yield good results even with small data sets if a naive Bayesian classifier is used.

The author contributed to the development of the Qualitative Sales Predictor (QSP) used originally by Digital Equipment Corporation and then Hewlett Packard for complex sales, which yields consistently accurate forecasts (70% Brier Skill Score). As expected, applying Bayesian machine learning to modify the QSP weightings increased this to 90% accuracy.⁸

Design of a Robust B2B Deal Forecasting System

We set out to build a simple, robust and highly accurate pipeline forecasting system that could be used by any B2B sales organisation, for deals both large and small.

System Specification

Produce a sales forecast that is over 90% accurate (Brier Skill Score).
Be more accurate and useful than other CRM deal forecasting methods.
Be suitable for small businesses (SMEs).
Be suitable for deals of any size.
Assume no existing data for system training.
Assume the user is not necessarily a full-time sales resource or experienced.
Provide instant, valuable feedback as soon as data is entered.
Coach the user on what to do to win the deal.
Allow the user to formulate their own qualification criteria.
Allow the user to include bad, irrelevant, redundant qualification criteria.
Allow the user to add or remove qualification criteria over time.
Enable evolution from simple scoring to evidence based qualification over time.
Have no upper or lower limits on the number of criteria.
Calculate personal weightings that compensate for bias, optimism etc.
Work from the first prospect and continuously improve.
Improve ROST by both reducing sales time AND increasing revenues per sales rep.

No existing CRM or sales enablement tool could do this. 

To achieve the flexibility, robustness and small data set goals, we used a naive Bayes classifier algorithm to maintain both company and personal weightings for a customisable set of qualification criteria. Crucially, separate weightings are calculated for the contribution of each of these criteria to the likelihood of a win vs its inverse correlation to a loss. This made the algorithm more responsive to “deal breaker” type criteria that may be introduced. 

The requirement to include bad, irrelevant or redundant criteria led us to include further refinements. 

We found that by systematically eliminating the weakest criteria in each set (i.e. the least important for determining a win or loss) the BSS score increased by about 12%.

The separate estimates of win and loss probabilities for each deal allows for faster learning, typically achieving over 85% forecast accuracy after only 10 win/loss records. The system learns to ignore redundant or irrelevant questions at the same time. After 20 win/loss data points a BSS of 90% is achieved.

By accommodating unanswered, irrelevant and modified questions, the system is not merely robust but antifragile. A naive user can define a poor set of qualification questions and the system shows which ones to remove. New questions can be added and the system automatically recalibrates with no lost data.

Personal Sales Coach 24/7

To test the feedback and coaching requirements set out above we incorporated the forecasting system in the Sales Coach element of the ProspectSafari CRM. After each contact with a prospect the user is invited to adjust their answers to a set of (user defined) qualification questions and the deal win probability is instantly recalculated.

An intelligent checklist shows progress through the deal stages, identifies issues, sets alerts and creates and completes tasks according to the established sales process.

Videos and expert guides on how to perform these tasks and answer the qualification questions are pushed to the user as appropriate, to provide automated coaching.

The qualification questions highlight what must be done to improve the win probability. Given the high accuracy of the probability estimate, the user pays attention.

Sales Improvement through Automatic Best Practice

If your manager requires you to answer qualification questions but you see no benefit then this becomes yet another time-wasting, box-ticking exercise to be avoided.

If however the answers to these questions instantly yield a deal win probability that is known to be accurate, then the dynamic changes. You pay attention to the questions and focus on getting the answers, as that will lead to your personal success. Effectively you self-coach and automatically work to best practice.

Keep it simple

By clicking a few buttons to adjust your answers to 5-10 questions after each contact with a prospect, this iterative system automatically generates a weighted qualification model unique to your specific business and produces a sales forecast that is far more reliable. 

It also provides valuable and valued guidance to the sales rep, revealing gaps in their knowledge or specific issues to address to progress the deal.

Conclusion

Our proposed B2B Deal Forecasting system is derived from considerable experience coaching strategic account teams in complex sales. It leverages best practice in deal qualification and multiple empirical studies in statistical analysis of deal win probability and uses the most effective machine learning methods from academic research.

Incorporated in ProspectSafari it provides deal-level feedback and coaching and better sales forecast accuracy than any other CRM. 

It can be applied to any deal size or type from micro-business to enterprise sales.

---

FAQ for B2B Sales Forecasting

---

References

1. Statistical analysis of the correlation of account planning workshop activity to revenue results, Christopher Neil-Jones, Talent Capital Partners, 2008  ↩︎
2. Interpreting the skill score form of forecast performance metrics, Edward Wheatcroft, London School of Economics, International Journal of Forecasting Issue 35, 2019 ↩︎
3.  A field study of sales forecasting accuracy and processes, Michael Lawrence, Marcus O’Connor, Bob Edmundson, European Journal of Operational Research, Volume 122, Issue 1, 2000 ↩︎
4. Implementing Lead Qualification Model Using ICP for Saas Products. Priya V, Laxmi and K, Hariharanath,  . International Journal of Management, 11(10), 2020  ↩︎
5. Probabilistic Modelling of a Sales Funnel to Prioritise Leads. Duncan, Brendan Andrew and Elkan, Charles Peter} (2015) Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ↩︎
6. Sales Process Excellence. Michael Webb 2014 ISBN 0977107221, 9780977107223 ↩︎
7. Machine Learning for Direct Marketing Response Models: Bayesian Networks with Evolutionary Programming. Geng Cui, Man Leung Wong, Hon-Kwong Lui, (2006). Management Science 52(4):597-612. ↩︎
8. A Bayesian Classification Approach to Improving Performance for a Real-World Sales Forecasting Application, C. Gallagher, M. G. Madden and B. D’Arcy  2015 IEEE 14th International Conference on Machine Learning and Applications (ICMLA), Miami, FL, USA, 2015 ↩︎