Most RevOps teams inherit a weighted pipeline model. Someone assigned probabilities to stages years ago, the spreadsheet survived two CRM migrations, and the number it produces is now read aloud on the forecast call every week. The question is whether a machine learning forecast is worth the switch, and the honest answer depends on how far into the quarter you need to be right.
What is weighted pipeline forecasting?
Weighted pipeline multiplies every open opportunity by a fixed probability tied to its stage, then sums the results. Stage four gets 60 percent, stage five gets 80 percent, and a $100,000 deal in stage five contributes $80,000 to the forecast.The appeal is that it takes an afternoon to build and anyone can follow the arithmetic. The mechanics and the standard weight tables are covered in our guide to weighted pipeline. The weakness is buried in the word "fixed." Those percentages came from a historical average someone calculated once. They do not know that your average deal size fell 12 percent last quarter or that your cycle stretched by three weeks.
What does a machine learning forecast do differently?
It groups opportunities by behavior and attributes, then predicts a close probability and a close timing curve for each group from your own outcomes. Stage is one input among many rather than the whole model.At ORM every opportunity is grouped by a machine learning model, and each group gets a predicted curve for how long it takes to close. Those curves run from 1 to 80 weeks, with most of the expectation landing before week 12 and very few groups carrying expectation past 52 weeks. That is a different kind of output. A weighted model tells you a dollar figure. A grouped model tells you a dollar figure and when it arrives, which is what makes in-quarter planning possible.
Which one is more accurate?
The model wins, and the margin is largest on day one of the quarter. Weighted pipeline converges toward correct as the quarter closes because the remaining deals resolve on their own. That convergence is worthless. Getting the forecast right in the last week does not help anyone, since the quarter has already happened by then.Across the market, forecast accuracy on new and expansion revenue usually sits around 90 percent, and teams get there through significant manual effort that stops working as conditions change. ORM targets 95 percent without manual adjustment and holds it from day one through day ninety, updating as the quarter progresses. The stability across the quarter matters more than the headline number. Definitions for how to measure any of this are in forecast accuracy.
How do the two compare on what RevOps cares about?
Six dimensions decide it.| Dimension | Weighted pipeline | Machine learning forecast |
|---|---|---|
| Setup time | An afternoon | Four to six weeks of model training |
| Day-one accuracy | Weak | Strong |
| Reacts to market change | Only when someone edits the weights | Automatically as new outcomes arrive |
| Timing prediction | None. Close date is taken as given | Predicted close curves by deal group |
| Explains a miss | Points at stages | Points at groups, aging, and slippage |
| Handles deal size variance | Poorly | Handles it as a modeled attribute |
When is weighted pipeline still the right call?
When you have too little closed history for a model to learn from, a short cycle, and consistent deal sizes. Under those conditions the arithmetic is defensible.It is also the right call when the forecast is a communication device rather than a planning tool. If your team needs a shared number to talk about in a weekly meeting and nobody makes resource decisions from it, the extra precision buys little. Be honest about which situation you are in. Most teams describe the second and behave like the first.
What breaks both models at the same time?
A change in the business or the market that neither one was built to see. The most common reason a forecast fails is that something shifted and the forecast still runs on old assumptions.Four examples show up repeatedly. A new competitor enters and creates pricing pressure, so average deal size falls. Interest rates rise, private equity firms slow capital deployment, portfolio companies cut costs, and win rates drop. Broad uncertainty makes buyers hesitate and cycles stretch from qualified to closed. Territories get reshuffled and reps lose momentum while pipeline still looks healthy on paper.
In every case the visible signals look fine for weeks. Pipeline stagnates, deals close for less than their recorded value, and win rates slide. A weighted model reports the old rates until someone notices. A trained model picks the shift up in the close rates and adjusts, which is the entire argument for running one. The related mistake is treating coverage as the answer, covered in why the 3x pipeline coverage rule is wrong.
Can you run both at once?
Yes, and you should for at least one full quarter. Publish the model output as the official forecast, keep the weighted number as a variance check, and review the gap every month.Two rules keep this honest. Name the official number before the quarter starts and do not change which one you cite based on which looks better in week eight. Second, when the two diverge by more than a few points, treat the gap as a research task rather than an argument. The divergence usually points at something real: a segment where deal sizes moved, a batch of aged pipeline nobody cleared, or a set of close dates that all shifted in the same week.
Once the model has beaten the spreadsheet on day-one accuracy for two consecutive quarters, retire the weights. Keeping both indefinitely gives every stakeholder a number to pick from, and a forecast that can be chosen is not a forecast.
Frequently Asked Questions
What is the difference between weighted pipeline and a machine learning forecast?
Weighted pipeline multiplies each open deal by a fixed probability attached to its stage and sums the result. A machine learning forecast assigns each opportunity to a group based on its attributes and behavior, then applies a close probability and a timing curve learned from how similar deals actually resolved. One uses a static rate you chose. The other uses a rate your history produced.
Is a machine learning forecast more accurate than weighted pipeline?
Usually, and the gap widens the further you are from quarter end. Weighted pipeline is at its worst on day one of the quarter, which is exactly when a forecast has value. It also cannot react to a change in market conditions, because the stage weights sit in a spreadsheet and update only when someone edits them.
Do stage probabilities ever work?
They work as a rough sanity check in a stable market with a short cycle and consistent deal sizes. They fail when deal size varies widely, when cycles stretch, or when a single quarter depends on a few large opportunities. In those conditions the weighted number lands close to correct by accident and is wrong about which deals produced it.
Can we run both forecasts at the same time?
Yes, and running both for a quarter or two is the cleanest way to build trust in a model. Publish one number as the official forecast and keep the second as a variance check. What you cannot do is switch between them mid-quarter depending on which looks better.
What replaces stage probabilities once the model is live?
Group-level close rates and timing curves. Instead of asking what percentage of stage four closes, you ask what percentage of this kind of deal closes and how many weeks it takes. That reframing is what makes the output usable for planning rather than just for reporting.
See how ORM turns these insights into action
ORM builds custom revenue forecast models for B2B SaaS companies. Not dashboards. Prescriptive analytics that tell you what to do next.
Schedule a Demo