AI Sales Forecasting

A small and medium business primer on AI forecasting

I remember when a Vice President of Sales made it clear to me the importance of accurate sales forecasting. I was Director of Sales for the Midwest region. While we were on a conference call with all of the district managers the Vice Resident of Sales asked me the following question” MARCUS, whydid you miss your sales forecast” my reply was that I had made my quota. His reply was I didn’t ask you about your f******* quota. I asked you about your f****** forecast. You did not make your forecast. So here’s my question to you are you a liar or are you stupid? My answer was I’m a liar. His response was a least we have something to work with. That is not the best way to manage, but in the Glengarry Glen Ross sales environment of the 80s that’s the way it was.

AI offers new opportunities to improve your sales forecasting. However, for it to be effective, you need to truly understand your sales cycle. This blog is to give small and medium business owners some ideas on how to use AI in forecasting

Objective Standards

Salespeople are notoriously inconsistent sales forecasters. I had salespeople who were great forecasters, and I had others who were terrible. Salespeople forecasts varied based upon human factors. This was particularly true when they were struggling with their quota. It is difficult to create a consistent forecast if you are dependent entirely on subjective measurements. The larger, the salesforce, the more difficult it becomes.

AI is not magic. The more high-quality data you feed it the better response you are going to receive. It is a predictive analytic engine that becomes more accurate as it is fed more accurate and relevant data. This requires very clear and defined sales cycle in which data points can be recorded and fed into the system.

The need for objective standards was important before the introduction of artificial intelligence. However, with the introduction of artificial intelligence, the stakes have become much higher. The consequences of feeding AI bad data can be catastrophic. Therefore, it is extremely important to plan prior to implementing AI into your forecasting system.

My company sold complicated networking solutions for small and medium businesses. The proposal cycle varied but was usually 4 to 6 months on the average. The proposals were complicated and the customers needs often very sophisticated. Therefore, I had to set out some extremely specific guidelines so that I could understand the sales funnel.

Initially, I had decided to identify important milestones in the sales process. I must give credit to my old boss Tom Putnam for teaching me the outlines of this system. The system works as follows:

1. I placed all pre-proposal opportunities as a list at the bottom of the funnel. I further divide these opportunities into two more categories. The first category was there was no pricing or proposal development activity going on and the second was the salesperson was developing a proposal for the client.

2. Submitted proposals was the second category. I assigned these proposals a 30% probability of sale. I assigned the probability of sale based upon prior experience.

3. If the client asked for a formal presentation. The probability of sale was incresed to 40%.

4. If the client told us we were a finalist for the job, the probability of sale increased to 50%

5. If the client told us we were a finalist, the probability went to 60%

6. If the client told us we were the leading vendor the probability went to 70%

7. If the client gave us a verbal, then the probability went to 80%

8. When we entered contract negotiations, the probability went to 95%

9. We considered the deal closed we we received a final contract and the payment.

This gave me an idea of where every opportunity was in the sales cycle, according to these objective and easily identifiable milestones. I considered more variables to affect the forecast probability, however, these were the most objective. The system was not always perfect, but it gave me an excellent idea of where everything stood.

The role of AI

AI analyzes your data and makes predictions. The structure I used is a start. It sets up baseline data. There are more factors that go into determining the probability of a sale. This is where I would use AI to help me set up a structure that would enable me to provide more data to increase accuracy.

Major manufacturers of CRM’s are adding AI to their systems. Establishing the data structure is critical no matter whose CRM you are using. This sets up what data to be recorded and how to monitor the recording. The following is Claude Sonnets method to incorporate my basic forecasting structure into an AI predictive model:

Core Implementation in CRM

Data Structure Setup:

• Create custom fields for each milestone with boolean (completed/not completed) status

• Add timestamp fields to track when each milestone was achieved

• Include deal value, sales rep, industry, deal size category, and other relevant attributes

• Set up automated probability updates when milestones are marked complete

Workflow Automation:

• Configure triggers that automatically update the deal probability when a milestone is completed

• Set up validation rules to ensure milestones follow logical sequence

• Create reports and dashboards showing pipeline by milestone stage

• Implement alerts for deals that have been stuck at certain milestones beyond expected timeframes

AI Enhancement Strategy

Historical Performance Analysis:

The AI system should analyze your historical data to identify patterns and adjust base probabilities. Key factors to track:

• Conversion rates from each milestone to deal closure by sales rep, industry, deal size, and time period

• Average time spent at each milestone for won vs. lost deals

• Seasonal patterns and market condition impacts

• Client behavior patterns (e.g., deals that skip certain milestones)

Dynamic Probability Adjustment:

Instead of fixed percentages, implement dynamic scoring:

Adjusted Probability = Base Milestone % × Performance Multiplier × Context Multiplier

Performance Multipliers:

• Sales rep historical conversion rate from this milestone

• Industry-specific conversion patterns

• Deal size impact on conversion likelihood

• Seasonal adjustments based on historical patterns

Context Multipliers:

• Time elapsed at current milestone (deals stuck too long may have lower probability)

• Competitive landscape indicators

• Client engagement metrics (email responses, meeting attendance)

• Economic indicators relevant to the client’s industry

Implementation Steps

Phase 1: Data Collection

• Ensure all historical deals have milestone completion dates

• Clean and standardize existing data

• Begin tracking additional context variables (meeting attendance, response times, etc.)

Phase 2: AI Model Development

• Build machine learning models using historical data to predict conversion likelihood

• Start with simple logistic regression models for each milestone transition

• Gradually implement more sophisticated models (random forest, gradient boosting) as data volume increases

Phase 3: Real-time Adjustment

• Implement scoring algorithms that update probabilities based on deal characteristics

• Create feedback loops where actual outcomes train the model continuously

• Set up A/B testing to compare AI-adjusted vs. fixed probabilities

Key Metrics to Track

Model Performance:

• Forecast accuracy vs. actual results by time period

• Calibration (do 70% probability deals actually close 70% of the time?)

• Early warning indicators for deals likely to stall

Business Impact:

• Improvement in forecast accuracy over baseline system

• Sales team adoption and workflow efficiency

• Pipeline management effectiveness

Technical Considerations

CRM Integration:

Most major CRMs (Salesforce, HubSpot, Microsoft Dynamics) support custom probability calculations through:

• Custom fields and workflow rules

• API integrations for AI model outputs

• Third-party apps specifically designed for sales forecasting

AI Model Deployment:

• Cloud-based ML platforms (AWS SageMaker, Google AI Platform) for model training

• Real-time scoring via API calls

• Batch processing for periodic model updates

The key to success is starting with your solid milestone foundation and gradually layering on AI enhancements as you collect more data and identify the factors that most significantly impact your conversion rates. This approach maintains the objectivity you’re seeking while adding predictive power based on your specific sales patterns

Human Factors

Throughout my career, I worked for companies that implemented magic new systems that were going to solve all of our problems. Inevitably, the implementation was horrible, and it took years to make them work appropriately. This made monitoring and management of the system difficult to impossible. There was just too much friction built into the system to make it easily adaptable.

This is why I would have the salespeople give their input into what they think is important. I would also ask for their opinion about how certain requirements were going to affect their job and how difficult it would be for them to be compliant. System designers need to consider behavioral design principles in the development and implementation of any system. In my experience, if you screw up the first round, it is difficult to come back. In a future blog, I will explain some of the behavioral design principles that are important to effective implementation of a system.

Summary

You can use AI to improve your operations and management. It can certainly improve the accuracy and consistency of your sales forecasting. However, there are risks with this approach. Inaccurate and faulty data will make these risk come true. The old IT principle of garbage and garbage out is still as true today as it always as. . I hope this blog stimulates some ideas and helps you use AI to increase the accuracy of your forecast