Sales forecasting methods: which one fits your pipeline.

How this comparison was made

Every method description below is drawn from the vendor's own documentation, fetched and read on 29 September 2026: Pipedrive's published weighted-value formula, Salesforce's object reference for the ForecastCategoryName field, HubSpot's forecast-tool setup guide, and HubSpot's own catalogue of forecasting models. The small-pipeline section is a direct calculation from the binomial distribution, not a vendor claim, and it's checkable with a calculator rather than a citation. Every quote below is sourced at the bottom of this page. Nothing here is instructed by, or paid for by, any vendor named.

The four methods, in one table

Before the detail, the shape of the comparison. "Inputs" is what the method actually needs to run; "blind spot" is the thing it structurally cannot see, not a bug you could fix with better data entry.

This isn't the full list of forecasting models in use. HubSpot's own guide catalogues twelve, including seasonal, regression-based, lead-driven and multivariable models that layer on top of these four. The four here are the ones that show up as an actual setting inside a CRM's forecast tool, which is what most sales teams are choosing between when they pick "how we forecast."

Stage-weighted forecasting: how the calculation actually runs

Stage-weighted forecasting multiplies the value of every open deal by a probability, then adds the results. The part most people get wrong is which probability wins when a deal has more than one available.

Pipedrive publishes the exact logic. Its support documentation lists four cases:

Two things follow directly from that. First, a deal-level probability always beats the stage default, even if a rep set it once and forgot about it three stages ago. Second, if nobody has touched either setting, the "weighted" forecast is just the raw deal value at 100%, which quietly turns a weighted forecast into an unweighted one for any deal a rep didn't bother to adjust.

Worked example, using Pipedrive's own numbers: a stage with a default 50% probability holding deals worth $4,245.74 in total produces a weighted value of $2,122.87 for that stage, before any single deal's own probability is applied on top. Multiply that same logic across every stage in the pipeline and you get the one number that ends up on a forecast slide. The number is only as good as how recently someone reviewed the probabilities feeding it.

This is also the mechanism behind pipeline velocity reporting: velocity measures how fast deals move through the stages this method weights, so a slowdown in velocity and a drop in weighted forecast are usually the same underlying problem showing up in two reports.

Rep-commit forecasting: the category every major CRM ships, under a different name

Rep-commit forecasting isn't a calculation. It's a manual sort: a person looks at a deal and places it into a category that represents how confident they are it closes, independent of whatever the stage-weighted number says.

Salesforce's own object reference documents the standard set directly: "The standard forecast categories are Pipeline, Best Case, Commit, Omitted (not included in forecasts), and Closed." HubSpot's forecast-tool documentation describes what is structurally the same five-bucket system under its own labels: "Not forecasted: deals that are in the pipeline for the current time period, but are not included in the forecast. Pipeline: deals that have a low likelihood of closing. Best case: deals that will close in the best case scenario, which have a moderate likelihood of closing. Commit: deals that have a high likelihood of closing and have been committed to the forecast. Closed won: deals that have closed within the forecasted time period."

Two vendors, effectively the same five buckets. That convergence is the evidence this method is genuinely standard, not a Salesforce quirk or a HubSpot quirk. What neither vendor's documentation addresses, because it's a management question rather than a software feature, is who checks whether a rep's Commit category has actually meant "90%+ likely to close" historically, or whether it's meant "I want my manager to stop asking about this deal." A category system is exactly as reliable as the discipline of whoever is filling it in, and unlike stage-weighted forecasting, there's no formula to audit.

Historical run-rate forecasting: projecting from what already happened

Run-rate forecasting doesn't look at the pipeline at all. It looks at what already closed and assumes the pattern continues.

HubSpot names two versions of this. The first is time-based: "The time series forecasting model predicts future revenue by analyzing historical sales data over consistent time intervals. By identifying patterns, trends, and recurring cycles in past performance, this model helps sales organizations anticipate future outcomes... Time series forecasting focuses on when sales happen rather than why they happen... this model assumes that past patterns will continue under similar conditions." The second is a simpler trend extrapolation: "The historical growth rate forecasting model projects future revenue by applying past growth trends to current performance. This model assumes that historical growth patterns will continue under similar market conditions."

Both versions share the same structural feature: neither one has any input for "we hired three new reps last month," "our main competitor just cut their price," or "the product had an outage in Q2." A run-rate forecast is a straight-line continuation of the past, and it will keep confidently projecting the old trend for a full reporting cycle after something real has changed underneath it, simply because closed-deal data lags the change by however long the sales cycle takes to complete. HubSpot's own framing of it, "used as a baseline model for annual planning," reflects that limit: it's a starting assumption to adjust from, not a live read of what's actually happening in the pipeline right now.

Deal-by-deal forecasting: one person's judgment, one deal at a time

Deal-by-deal (HubSpot calls it opportunity-based) forecasting skips both the formula and the category. A rep or manager looks at a specific, named deal and estimates whether and when it closes, based on everything they know about it that a CRM field can't capture: how the champion sounded on the last call, whether procurement has actually started, whether the competitor is still in the room.

HubSpot's guidance is specific about where this fits: "The opportunity-based forecasting model estimates revenue by evaluating individual deals rather than aggregated pipeline stages. Each opportunity is assessed based on factors such as deal size, close date, probability, and known risks... Best for: Forecasting revenue when a small number of large or strategic deals materially impact results."

That's a narrower brief than most teams give this method. It's genuinely strong when a handful of large deals determine the quarter, because a human who knows the deal can weigh context a probability field can't hold. It falls apart the moment it's applied to an entire pipeline of forty or fifty deals per rep, because at that volume "judgment on each one" becomes a rushed gut-check done once a week before a forecast call, which is worse than a stale stage probability, not better. The method doesn't scale down to zero attention per deal gracefully. It just gets less honest about how much attention it's actually getting.

Where probability math breaks on a small pipeline

This is the part none of the four methods, or their vendors, will tell you, because it isn't a feature gap. It's a property of small numbers.

Stage-weighted forecasting works by treating a 40% probability as "this deal is worth 0.4 of its value, on average, across many deals like it." That's a completely reasonable way to describe a large set of similar deals. It's a much shakier way to describe five.

Take 5 open deals, each independently weighted at 40%. The weighted forecast says 2.0 deals' worth of value will close. But the actual number of deals that close follows a binomial distribution, and at n=5, p=0.4, the spread around that "2" is wide: a calculation of the exact probabilities shows roughly a one-in-three chance (33.7%) of closing 1 deal or fewer, and roughly a one-in-three chance (31.7%) of closing 3 deals or more. The forecast isn't wrong to say "2." It's just describing the center of a range that's nearly as likely to land well below it as well above it.

The pattern holds regardless of the exact probability you use: the swing scales with the square root of the number of open deals, not with the number itself. Going from 5 deals to 20 cuts the relative swing by half. Going from 20 to 100 (five times the deals) only cuts it by a little over half again, because the gains from more data shrink the more of it you already have.

The practical takeaway isn't "don't use stage-weighted forecasting." It's that the method needs volume to do what it's designed to do, and a team with a genuinely small number of open deals (a new territory, an enterprise motion with five deals a quarter, a founder-led sales process) is better served leaning harder on rep-commit or deal-by-deal judgment for the current period, and treating the weighted number as a rough planning input rather than a number to defend to the board. This is the same reason win rate gets unreliable on a small sample: a percentage calculated from a handful of outcomes is mostly noise wearing the clothes of a metric.

Where each method is weaker

No method here is the right answer on its own. Equal weight, in order:

• Stage-weighted is only as current as the last time someone reviewed stage and deal probabilities. Left alone for a quarter, it drifts toward describing last quarter's pipeline, not this one.
• Rep-commit has no calculation to audit. If a manager never checks whether a rep's historical Commit deals actually closed at a Commit-level rate, the category is a vibe with a label on it.
• Historical run-rate cannot see change. New hires, a price increase, a lost competitor, a product launch, none of it exists in the model until it has already shown up in several closed periods.
• Deal-by-deal doesn't scale. It's strong on the five deals a VP is personally tracking and gets progressively less honest as the deal count per person climbs past what anyone can actually hold in their head.

Which method fits your pipeline

Most CRM forecast tools, including HubSpot's and Pipedrive's, let you run more than one of these side by side rather than forcing a single choice, and most mature RevOps setups actually do: a stage-weighted number for the rolling view, a rep-commit number for the number a VP defends to the board, and a run-rate check as a sanity floor underneath both. The pipeline stages feeding the weighted number and the pipeline itself are worth getting right first; a clean formula run against messy stage definitions still produces a confident, wrong number.

FAQ

If you want a second pair of eyes on your own setup, Salestruct runs a free diagnostic.