It is the last Thursday of the quarter. Your CRO is on a call with the board in 40 minutes. She asks you one question: "What are we going to close?" You open the CRM, you stare at the weighted pipeline number, and you know in your gut it is wrong. Half the deals in commit have not had a real conversation in two weeks. Two of them are sitting at 80% probability and the champion just left the company. You give her a number anyway. You pad it down by 15% because that is what you always do. She presents it. Three weeks later you miss it by 12%.
I have lived this exact Thursday more times than I want to admit. Across ten years in RevOps and now as a founding GTM engineer, the single most common request I get from CEOs and CSOs is some version of "make the forecast believable." Not better. Believable. They have stopped trusting their own numbers, and once that trust is gone, every planning conversation gets harder.
Here is the uncomfortable truth backed by data: 79% of sales organizations miss their forecast by more than 10%, and only 7% hit 90%+ accuracy. So if your forecast is off by double digits, you are not broken. You are normal. But normal is expensive when a board is pricing your next round on the number you just missed.
This post is the forecasting system I build for B2B teams. Not a tool pitch. A method.
of sales organizations miss their forecast by more than 10%. Only 7% reach 90% accuracy. If you are guessing, you are in the majority.
Why your forecast is always wrong
There is rarely one reason. There are usually four, and they stack.
The first is that probability lives in the wrong place. Most CRMs assign a close probability to a deal stage. Proposal sent equals 50%. Contract out equals 80%. So your weighted pipeline math multiplies deal value by stage probability and spits out a number. The problem is that stage is a proxy for activity, not for the actual odds of closing. A deal can sit in "proposal sent" because the rep emailed a PDF and never heard back. Same stage, same 50%, wildly different reality. Weighted pipeline ignores deal quality, rep behavior, and what is happening in the buyer's world.
The second reason is rep optimism. Salespeople are paid to believe. That is a feature when they are working a deal and a bug when they are calling a forecast. A "commit" from a senior enterprise rep who has closed 40 deals means something completely different from a "commit" from someone in month three. Until you account for who is calling the number, every commit gets treated as equal, and they are not.
The third reason is dirty data underneath the math. Close dates that have been pushed four times. Amounts that were never updated after the discount. Deals marked open that died a month ago. No forecasting model survives bad inputs. This is why CRM data quality is the foundation, not a side project. Companies with disciplined CRM hygiene see forecasting accuracy improve 15 to 25% over raw weighted pipeline, before they add any fancy tooling.
The fourth reason is that nobody inspects the forecast. The number gets rolled up, presented, and then everyone moves on until the next quarter. There is no feedback loop. You never go back and ask why the deals you called confidently did not close. So you make the same error next quarter, with confidence.
The metric you should actually track
Before you fix anything, you need to know how wrong you are and where. Most teams cannot answer this because they only track one thing: did we hit the total number. That is too coarse to act on.
Track accuracy by forecast category instead. Three numbers matter.
Commit accuracy is closed-won as a percentage of what you committed. Healthy is around 85% at the median, with the top quartile above 95%. If your commit accuracy is 60%, your reps are calling deals "commit" that have no business being there. That is a coaching and definition problem.
Best case accuracy is closed-won as a percentage of best case. Median here is about 38%, top quartile above 55%. Best case is meant to be the optimistic scenario, so lower conversion is expected. But if your best case converts at 5%, that category is a junk drawer and tells you nothing.
Forecast variance is the gap between what you called at the start of the period and what you closed. Most B2B teams run at plus or minus 15 to 25%. High performers get to plus or minus 5 to 10%. That spread is the entire game. A board can plan around plus or minus 7%. It cannot plan around plus or minus 25%.
When I start with a new client, the first thing I do is calculate these three numbers for the last four quarters. It takes an afternoon. It always tells the same story: the forecast is not wrong randomly, it is wrong in a pattern. One rep over-commits every quarter. One stage converts at half the assumed rate. Renewals get forecast as new business. The pattern is the fix.
Forecast categories beat weighted pipeline
Here is my actual opinion, and it is not the popular one: stop using stage-weighted pipeline as your primary forecast. Use forecast categories that the rep sets deliberately, separate from the deal stage.
The difference matters. Deal stage describes where the buyer is in the process. Forecast category describes how confident the rep is that the deal closes this period. Those are two different questions, and collapsing them into one number is why weighted pipeline drifts.
A clean category structure looks like this:
The category is the rep's call, made out loud, that you can hold them to. The stage stays as the process tracker it was always meant to be. When you separate the two, your forecast review stops being a guessing game about what stage probability means and becomes a conversation about specific deals and specific risks.
Tools like Clari and HubSpot both support forecast categories natively now. The structure is not the hard part. The discipline of calling them honestly is.
The system that makes the number trustworthy
A forecast you can defend to a board is not a model. It is a weekly operating rhythm with four moving parts.
Step three is where most of the accuracy comes from, and it is the cheapest to fix. Companies that track pipeline velocity weekly hit 87% forecast accuracy versus 52% for teams with irregular tracking. That gap is not a tool. It is a calendar invite that actually happens every week.
The inspection question is deceptively simple. For every commit deal, ask: what has to be true for this to close on this date? If the rep cannot answer specifically, the deal is not a commit. If the answer is "they said they are excited," it is not a commit. A commit means the economic buyer has agreed, the paper is moving, and there is a mutual plan to signature. Everything else is best case at most. Run this filter every week and your commit accuracy climbs within a quarter.
A forecast is not a model you build once. It is a weekly conversation you hold reps to.
The teams with the tightest forecasts are not the ones with the best AI. They are the ones who inspect every commit deal every week and score their own calls afterward.
Where AI forecasting actually helps, and where it does not
You cannot read a sales blog in 2026 without being told AI will fix your forecast. Gartner predicts 70% of large organizations will adopt AI-based forecasting by 2030. So let me be specific about what it does and does not do, because the hype is leading teams to buy tools that will not help them yet.
AI forecasting is good at one thing: spotting patterns in your historical data that a human review misses. It can tell you that deals which skip the security review stage close 40% less often. It can flag a commit deal that has gone quiet based on email and call activity, before the rep admits it is slipping. This is real value, and it is the same engine behind most revenue intelligence platforms.
What AI cannot do is fix dirty inputs or invent signal that is not there. If your reps do not log activity, the model has nothing to learn from. If your close dates are fiction, the model learns fiction. AI forecasting on top of a messy CRM gives you a confident wrong number instead of an honest guess, which is worse. The order of operations is fixed: clean data, then category discipline, then a weekly rhythm, and only then is AI worth the spend. Bolting it on first is how teams waste a year and a six-figure contract.
There is a version of this that does work earlier, and it is lighter than a full platform. You can build a simple signal layer with automation that watches for the things a rep will not tell you: a champion who changed jobs on LinkedIn, a deal with no activity in 14 days, a close date that has moved three times. Those signals go to the manager before the weekly review, so the inspection conversation starts from facts instead of vibes. That is the 80% of AI value for 20% of the cost, and it is what I build for most clients before anyone talks about Clari or Gong.
What this looks like for a real team
A Series A client came to me running at roughly plus or minus 30% forecast variance. The CEO had stopped sharing rep-level forecasts with the board because they were embarrassing. Three things changed over one quarter.
First, we cleaned the pipeline. Every open deal needed a close date in the future, a next step, and an amount that matched the latest proposal. We deleted or closed-lost 60 zombie deals that had been propping up the number. The pipeline shrank by 22% and got real.
Second, we replaced stage-weighted forecasting with the four categories above and wrote a one-paragraph definition of "commit" that the whole team agreed on. The definition was what changed things. Once everyone meant the same thing by the word, the number meant something.
Third, we put a 30-minute weekly forecast inspection on the calendar where every commit deal got the "what has to be true" question. Boring. Relentless. It worked.
By the end of the quarter, variance was down to plus or minus 9%. Not because of a tool. Because the inputs were clean, the language was shared, and someone looked at the deals every week. The CEO put rep-level forecasts back in the board deck. That is the whole win: a number you are willing to show people.
How forecasting connects to the rest of your system
Forecasting is downstream of a lot of other things, which is why fixing it in isolation rarely sticks. If your pipeline management is loose, your forecast inputs are loose. If your quota setting is fantasy, reps sandbag the forecast to protect themselves and your commit accuracy stays low on purpose. If your capacity planning is off, you are forecasting against a number the team was never staffed to hit.
The forecast is the output that exposes every other weakness in your revenue system. That is why CEOs ask for it first and why it is the wrong place to start fixing. Start with the data and the deal hygiene. The forecast gets believable as a side effect.
Tired of guessing the number every quarter?
Book a free 30-minute audit. We will calculate your forecast accuracy by category, show you exactly where the variance hides, and name the three fixes we would make first.
Book an audit →Frequently asked questions
What is a good sales forecast accuracy for B2B SaaS?
Most B2B teams run at plus or minus 15 to 25% variance against their start-of-period call. High performers reach plus or minus 5 to 10%. For forecast categories, healthy commit accuracy is around 85% at the median with the top quartile above 95%, and best case accuracy sits near 38% median. If your variance is wider than 25% or your commit accuracy is below 70%, you have a process problem, not a tooling problem.
Is weighted pipeline a bad forecasting method?
Weighted pipeline is fine as a rough sanity check but weak as your primary forecast. It assigns a fixed probability to a deal stage, which ignores deal quality, rep behavior, and what is happening in the buyer's world. Two deals in the same stage can have completely different odds. Forecast categories that reps set deliberately, separate from deal stage, give you a number you can actually hold people to.
How often should we review the forecast?
Weekly, at minimum for commit deals. Teams that track pipeline movement weekly hit around 87% forecast accuracy versus 52% for teams that review irregularly. A 30-minute weekly inspection where every commit deal gets the "what has to be true for this to close" question is the single highest-return change most teams can make.
Do we need an AI forecasting tool like Clari?
Not first. AI forecasting helps you find patterns in clean historical data and flag deals going quiet, but it cannot fix dirty inputs or missing activity logs. If your CRM data is messy, AI gives you a confident wrong number, which is worse than an honest guess. Fix data quality and category discipline first. Add AI once the inputs are trustworthy and you have a real volume of deals for a model to learn from.
Why do reps sandbag the forecast?
Usually because the quota is unrealistic or the comp plan punishes a missed commit harder than it rewards an honest one. Reps protect themselves by under-calling so they can beat their own number. The fix is rarely more pressure. It is a fair quota, a clear and consistent definition of what each forecast category means, and a manager who rewards an accurate call over an optimistic one. Accuracy is a culture you build, not a checkbox you enforce.
Build a forecast your board can trust
The goal is not a perfect forecast. Perfect does not exist in B2B sales. The goal is a number you are willing to show people, that lands within a range a board can plan around, and that gets a little tighter every quarter because you are scoring your own calls and learning from the misses.
That comes from clean data, shared definitions, and a weekly rhythm. The tooling is the last 10%, not the first.
If your forecast is the conversation you dread every quarter, we can help you fix the system underneath it. Tell us where it hurts and we will show you the three changes that move your accuracy the fastest.