It is the last week of the quarter. Your CRO opens the board deck to the forecast slide. One number sits there in bold: $4.2M. The chair asks the only question that matters. "How confident are you in that?" And the room goes quiet, because everyone knows the honest answer is "somewhere between $3.6M and $4.8M, depending on three deals we cannot read."
I have watched that scene play out at Series A companies and at businesses doing $40M in ARR. The problem is almost never that the sales team is lazy about updating the CRM. The problem is that the forecast is a single number produced by a single method, usually weighted pipeline, and a single number cannot carry the uncertainty that actually lives in the business.
Revenue forecasting is not one calculation. It is three or four different models run in parallel, then reconciled into a range you can defend. Most B2B teams run one of them, badly, and then act surprised when the number is wrong. Here is how to build a forecast your board can trust, and why the model you pick matters more than the tool you buy.
Why your one-number forecast lies
Start with the uncomfortable data. Only 7% of companies forecast within 90% accuracy, according to Gartner. Gong found that more than 80% of companies missed their revenue forecast in at least one quarter over a two-year stretch. Clari's 2026 data put enterprise miss rates even higher, with 87% of enterprises missing revenue targets in 2025.
A forecast is a probability distribution, and you keep reporting it as a point.
The board does not need one number that turns out wrong. It needs a range with a confidence level and a clear story about what moves it.
The single-number habit comes from the CRM. You set a probability on each deal stage, the CRM multiplies deal value by that probability, sums the column, and hands you a weighted pipeline total. It looks precise. It is not. Stage probabilities are averages baked in years ago that have nothing to do with the specific deals in front of you. A deal at 60% "proposal" stage is not 60% likely to close. It is either going to close or it is not, and the rep usually has a real read that the 60% erases.
So the number that lands in the board deck is an average of averages, presented as certainty. When it misses, nobody can say why, because there was no model underneath it, just a formula.
The four models you should actually run
A good forecast triangulates. You run several independent methods, and where they agree you gain confidence, where they diverge you dig. Think of it the way a pilot cross-checks instruments. No single gauge is trusted alone.
Each model answers a different question. The run-rate tells you the floor. Weighted pipeline tells you the statistical expectation on new business. Capacity tells you what is physically possible. The commit tells you what the people closest to the deals actually believe. When all four land within a tight band, you have a number. When they scatter, you have a conversation to have before the quarter, not after.
For a B2B SaaS business, the order matters. Build the recurring baseline first. That is the part of next quarter's revenue that already exists in signed contracts, and it is the part most teams skip because it is not exciting. Then layer new business on top. A lot of RevOps work here is just getting the recurring baseline clean, which is why CRM data quality sits underneath every forecast that works.
Model one: the recurring revenue baseline
If you sell subscriptions, most of next quarter is already decided. Existing customers will renew or they will not. Some will expand. Some will contract or churn. None of that lives in your new-business pipeline, and yet it is the biggest single input to the number.
Build it as a bridge. Start with current ARR. Add the expansion you can already see, the seats a customer committed to add, the upsell in a signed order form. Subtract the renewals at risk and the contractions you know about. What is left is your baseline, and for a healthy SaaS company that baseline should cover a large share of the quarter before a single new deal closes.
The teams that miss badly are almost always the teams that forecast new business obsessively and treat the base as automatic. Then two enterprise accounts churn, net revenue retention drops below 100%, and the new-business number that looked fine cannot dig out of the hole. If your net revenue retention is soft, your forecast is fragile no matter how good your pipeline looks.
Model two: weighted pipeline, done honestly
Weighted pipeline is the model everyone runs and almost nobody runs well. The math is simple. The inputs are where it breaks.
The first fix is probability. Do not use the stage defaults your CRM shipped with. Pull your own historical data and calculate the actual close rate from each stage. If deals that reach "proposal sent" close 34% of the time over the last four quarters, use 34%, not the 60% someone typed in during setup. This one change usually cuts a forecast that was running 20% too high back to reality, because the default probabilities are almost always optimistic.
The second fix is time. A weighted pipeline number is meaningless without a close date you trust. A $200K deal weighted at 40% is worth $80K to this quarter only if it is actually going to close this quarter. Deals slip. If your average sales cycle length is 90 days and a deal entered pipeline three weeks ago, it is not closing this quarter no matter what the close date field says. Filter your weighted pipeline to deals whose realistic close date falls inside the period.
Weighted pipeline is strongest when you have enough deal volume for the averages to mean something. A team pushing 200 deals a quarter can trust the statistics. A team closing 12 large deals a quarter cannot, because the law of large numbers does not apply to twelve. That is why enterprise teams lean harder on the commit model and mid-market teams lean harder on weighted pipeline. Match the model to your deal shape.
Model three: bottom-up capacity as a sanity check
This one takes ten minutes and catches forecasts that have floated off into fantasy. Count your ramped reps. Multiply by the average number of deals a ramped rep closes in a quarter. Multiply by average deal size. That is your capacity ceiling, and it ignores the pipeline completely.
If your weighted pipeline says $6M but your capacity math says the team has never produced more than $4M in a quarter and you did not add headcount, one of those numbers is wrong. Usually it is the pipeline, inflated by stale deals nobody has the heart to close out. Capacity is the reality check that keeps optimism honest, and it doubles as an input to quota setting and capacity planning when you build next year's plan.
Model four: the commit, and forecast categories
The most accurate forecast input in a mature B2B org is not a formula. It is the rep and manager judgment, deal by deal, sorted into categories. This is the model revenue intelligence tools were built around, and when the culture supports it, it beats weighted pipeline on accuracy.
The standard categories are commit, best case, and pipeline. Commit means the rep is willing to stake their credibility that this deal closes this quarter. Best case means it could close if things break right. Pipeline means it is real but not this period. The power is that these are human reads, not stage math, so they capture the thing a probability column cannot: the champion who just went quiet, the legal review that is actually moving, the budget that got frozen last week.
Categories only work if commit means something. On teams where reps sandbag to beat a low number, or inflate to look busy, the categories are noise. Building that trust is a management job, not a tooling job, and it is the same discipline that makes pipeline stages mean anything.
Here is where the 2026 benchmark data gets useful. When you measure how often each category actually converts, the healthy numbers look like this.
Read those numbers as a health check. If your commit category is closing at 60% instead of 85%, your reps are calling deals commit that do not deserve it, and your forecast will run hot every quarter. If commit closes at 98%, your reps are sandbagging and hiding upside. The gap between category and actual conversion is one of the cleanest diagnostics in RevOps.
Reconciling four models into one number
You now have four estimates. They will not agree, and that is the point. The job is to reconcile them into a forecast with a range, not to average them into mush.
I build it as three numbers, not one. The floor is the recurring baseline plus the commit deals, the revenue I would bet the company on. The likely case adds a realistic slice of best-case deals and the weighted-pipeline expectation. The ceiling is the likely case plus the upside that only happens if the quarter goes right. Report all three to the board with the assumptions attached. "Floor $3.6M, likely $4.2M, ceiling $4.8M, and the swing is three enterprise deals in procurement" is a forecast a board can plan around. "$4.2M" is a coin flip wearing a suit.
The variance between those numbers is itself a metric. Investors start to worry when forecast variance runs above 25%, and best-in-class teams hold it inside 5%. If your floor and ceiling are 40% apart every quarter, the problem is not the forecast, it is that too much of your revenue depends on a handful of unpredictable deals, which is a pipeline coverage problem. Thin coverage forces you to count on longshots, so read this alongside your pipeline coverage ratio.
What quietly breaks every forecast
None of these models survive dirty data. The failure is almost always upstream of the math.
Close dates that nobody maintains are the first killer. If reps set a close date once and never move it, your period filter is worthless and stale deals pile into the current quarter. Deal amounts that do not match the eventual contract are the second, common on teams without a deal desk or clean pricing. And missing stage history is the third, because you cannot calculate real stage-conversion probabilities if the CRM did not record when deals moved.
Fixing this is not glamorous and it is most of the actual work. A forecast is only as good as the CRM feeding it, which is why we treat forecasting as a CRM and RevOps build, not a spreadsheet exercise. Automating the hygiene, the close-date nudges, the stage-slippage alerts, the amount validation, is where AI and automation actually earns its keep, long before anyone needs a fancy prediction model.
Spreadsheet, CRM, or a forecasting tool
The tooling question comes last, not first, because the model matters more than the software. A team under 20 reps can run all four models in a spreadsheet pulled from CRM exports, and I have seen those forecasts beat six-figure platforms because the humans understood the inputs.
Native CRM forecasting, the Salesforce or HubSpot forecast module, handles categories and weighted pipeline reasonably once you configure the probabilities yourself. It is the right first step for most mid-market teams. Dedicated revenue intelligence platforms like Clari, Gong, or Aviso add AI-driven predictions, automated deal-slippage detection, and a clean forecast-call workflow. They are worth it once you have enough deal volume and enough reps that manual reconciliation eats real hours, and once your data is clean enough to trust. If your CRM is a mess, a $90K forecasting tool just predicts garbage faster. I wrote more about that trade-off in the revenue intelligence platform breakdown.
The sequence is always the same. Get the models right in a spreadsheet, prove they beat your current forecast, then buy tooling to automate what you already understand. Buying the tool first, hoping it will teach you forecasting, is how teams end up with an expensive dashboard nobody trusts.
Your forecast keeps missing and nobody can say why?
Book a free 30-minute audit. We will look at your pipeline data, your stage probabilities, and your commit accuracy, then show you the three fixes we would make first.
Book an audit →The forecast call is the real product
One last thing that has nothing to do with models. The best forecasting teams run a weekly forecast call with a fixed structure. Every deal in commit gets a hard look. What changed since last week. What is the next step and the date. Who on the buying side is actually driving. The number that comes out of that call is worth more than any algorithm, because it is a number people have defended out loud.
A model gives you a starting point. The forecast call is where judgment corrects the model, where a rep says "I know the math says 40% but I lost the champion, drop it," and where a manager says "you have called this commit for three weeks, close it or move it." That conversation is the actual forecasting engine. The models just give it something honest to argue about. For more on making the number itself trustworthy, the sales forecasting accuracy guide goes deeper on the diagnostics.
FAQ
What is the most accurate revenue forecasting method for B2B SaaS?
There is no single best method. The most accurate forecasts triangulate several models: a recurring revenue baseline, weighted pipeline for new business, a bottom-up capacity check, and rep-manager commit judgment. Where they agree, you have confidence. Where they diverge, you have a problem to investigate before the quarter closes. Teams with high deal volume lean on weighted pipeline; enterprise teams with few large deals lean on commit.
How accurate should my revenue forecast be?
A healthy B2B SaaS forecast lands within roughly 10% of actual. Only about 7% of companies hit 90%-plus accuracy, so most teams have real room to improve. Measure it as variance: median teams run around 15% variance, top-quartile teams hold 8%, and best-in-class teams stay inside 5%. Variance above 25% is a red flag investors notice.
Why is weighted pipeline forecasting so often wrong?
Two reasons. First, most teams use the stage probabilities their CRM shipped with instead of calculating real close rates from their own history, so the numbers run optimistic. Second, they count every open deal without filtering for a realistic close date, so deals that cannot close this quarter still inflate the number. Fix the probabilities with your own data and filter by realistic close date, and weighted pipeline gets a lot more honest.
Should I forecast recurring revenue and new business separately?
Yes. For a subscription business, most of next quarter is already decided by renewals, expansion, and churn in the existing base. Build that recurring baseline first as a bridge from current ARR, then layer new business on top. Teams that forecast new business obsessively and treat the base as automatic get blindsided when a couple of accounts churn and net revenue retention drops.
Do I need a forecasting tool like Clari or Gong?
Not at first. A team under 20 reps can run every model in a spreadsheet built from CRM exports. Native CRM forecasting in Salesforce or HubSpot handles most mid-market needs once you configure real probabilities. Dedicated platforms earn their cost when deal volume and rep count make manual reconciliation expensive, and only after your CRM data is clean. A forecasting tool on dirty data just predicts the wrong number faster.
Build a forecast the board can trust
The teams that forecast well are not smarter about deals. They run several models instead of one, they reconcile them into a range instead of a point, and they keep the CRM clean enough that the math means something. That is a RevOps build, not a spreadsheet formula.
If your forecast keeps missing and nobody in the room can explain why, that is the signal that you have one model where you need four. Book a free audit and we will show you where your number is breaking and the fastest way to make it predictable.