Forecasting from CRM data

Forecasting from CRM data means deriving expected revenue from new deals from what is in your CRM: the value, stage, expected close date and history of every open deal. The standard method is the weighted pipeline, deal value times win probability. It only works if the probabilities and timing are measured on your own closed deals, and if you correct for deals that stall or keep slipping. A CRM forecast also only predicts new revenue: running contracts and recurring customers have to be added separately.

What does a CRM know, and what does it not know?

A CRM is the best source for one question: what new revenue is on its way? It knows the open opportunities, their stage, their history and the activities of salespeople. No other system has that information.

What a CRM usually does not know:

• What existing customers will buy without a new deal. Recurring revenue is often not in the CRM.
• What happens to a deal after signing. Whether it became an order, when invoicing starts and whether everything is invoiced.
• How the revenue from a won deal is spread over time.

A forecast based only on the CRM is therefore a forecast of new revenue, at the moment of closing. For a revenue forecast, it has to be supplemented with contracted and recurring revenue, and converted to the moment of invoicing. How those layers come together is described in what is revenue forecasting.

Which CRM fields do you need for a forecast?

For a usable CRM forecast, you need at least the following per deal:

Most CRMs store some of this automatically, such as Salesforce with field history, and HubSpot and Pipedrive with activity and stage history. Check whether the history of close date and value is kept. If a field is overwritten without history, you cannot see afterwards how often a deal was pushed back.

If fields are missing or poorly filled, you can still build a forecast, with more uncertainty. The article forecasting with incomplete CRM data shows how.

Step 1: calibrate the win probability

The default stage probabilities in your CRM are a starting point, not a measurement. Replace them with what your own history shows.

1. Take all deals closed in the past 12 to 24 months, won or lost.
2. Determine per stage: of all deals that ever reached this stage, what share was won?
3. Do the same per segment if you suspect it differs: new versus existing customers, small versus large deals, per product or per region.

Watch out for one trap: also count the deals that skipped a stage or went straight from a stage to lost. If you only look at deals that moved neatly through every stage, you overestimate the probability.

Step 2: calibrate the timing

A win probability says nothing about when a deal closes. Determine from your history:

• The average and median time from each stage to closing.
• How often the expected close date was met, and how far deals overran on average.

Use this to spread expected revenue across periods. A deal with a close date next month, in a stage where deals on average still need four months, does not belong in next month's forecast at full value. The concept of pipeline velocity helps you assess the speed of your pipeline as a whole.

Step 3: calibrate the value

For won deals, compare the CRM value at the time they were in the "quote" or "negotiation" stage with the value eventually signed. The average difference is your value correction. If the signed value is on average 8 percent lower, reduce the value of open deals in those stages by 8 percent.

Step 4: correct for stalling and slipping

Two corrections make a big difference:

Stalled deals. Deals that have been in their current stage longer than, for example, twice the median time for that stage have a much lower probability than the stage suggests. Determine from your history how high that probability still is, and apply it.

Slipping deals. Deals whose close date has already moved two or more times. Look in your history at how many such deals were eventually won, and in which period.

These corrections are the difference between a pipeline that looks healthy on paper and a forecast that comes true.

Step 5: supplement and convert

The CRM forecast is now an expectation of new revenue at the moment of closing. Convert it into revenue per period by assuming a billing schedule for each deal: one-off, monthly over the term, or in instalments. Add contracted and recurring revenue from the accounts and the contract register.

How do you keep CRM data quality under control?

A CRM forecast is only as good as the CRM. The checks that do most for your forecast:

• Deals without a close date, or with a close date in the past.
• Deals with no activity for longer than your normal cycle.
• Deals with a value of zero or an unrealistically high value.
• Won deals with no follow-through in invoicing.
• Large differences between the forecast category the salesperson chooses and the calibrated probability.

How to run such checks automatically is described in how do you check CRM data automatically.

Worked example

Worked example: suppose a B2B company has EUR 1,500,000 of open deals with an expected close date in the coming quarter. The default weighting in the CRM gives a forecast of EUR 620,000.

After calibration on 18 months of history:

• The measured win rates are on average lower than the default probabilities. That brings the weighted value to EUR 480,000.
• Of these deals, EUR 400,000 in value has been stalled for longer than twice the normal cycle time. History shows such deals rarely close. After correction: EUR 410,000.
• Signed value is on average 7 percent below CRM value. After correction: about EUR 381,000.
• According to history, some deals close a quarter later than expected. After correcting for timing: about EUR 320,000 in this quarter, with the rest moving to the next.

The CRM forecast for the quarter goes from EUR 620,000 to about EUR 320,000. That is not pessimism, it is what the company's own history says. The figures in this example were chosen to show the steps, not as a benchmark.

Where does AI help?

The steps above are manual work with fixed rules. A statistical or AI model can combine them and apply them per deal, while weighing more signals: number of touchpoints, involvement of several contacts, customer response time, previous purchases. The advantage is consistency and scale. The limitation is the same as with manual work: the model learns from your history, and if that history is incomplete or distorted, it learns the wrong patterns. See AI revenue forecasting explained.

Frequently asked questions

Which CRM is best for forecasting?

The CRM your team uses consistently and that keeps the history of its fields. Salesforce, HubSpot and Pipedrive can all provide a usable basis. The difference lies far more in how it is filled than in which package it is.

Are the forecasting features in my CRM enough?

For a quick look at the pipeline, yes. For a revenue forecast, no, because they usually work with default probabilities, only see new deals and count at the moment of closing instead of the moment of invoicing.

How often should I recalculate the win rates?

Every quarter or every six months. Win rates change with new products, different prices, different salespeople and a different market. A calibration from two years ago is quickly out of date.

What do I do with deals the salesperson calls certain but the model rates low?

Discuss them. Sometimes the salesperson knows something that is not in the CRM. Record both estimates and measure afterwards who was right. That makes the next forecast better, and the CRM too, because the missing information finally gets a place.