Why your sales forecast is wrong: five CRM fields that lie to you

There is a meeting that happens in every sales organisation roughly four times a year. A number that was defended for eleven weeks quietly becomes a different number. Someone asks what changed. The answers are always the same shape: the deal slipped, procurement got involved, the champion went quiet, budget moved to next year.

Nobody in that room lied. The forecast was assembled honestly, from fields that reps filled in honestly, and it was still wrong by a margin that would get a finance team fired.

If you run a sales team of 5–50 reps and you commit a number every quarter, this post is about why sales forecast accuracy is structurally hard — and specifically about the five CRM fields that produce most of the error.

What the data says about sales forecast accuracy

Start with the scale of the problem, because most sales leaders assume their team is unusually bad at this. They are not.

• Gartner's research on sales forecasting puts the median organisation in roughly the 70–80% accuracy band, and finds that fewer than half of sales leaders have high confidence in their own forecast.
• Clari's research found that 87% of enterprises missed their revenue targets in 2025 — a number that is hard to reconcile with the idea that forecasting is a solved discipline.
• Salesforce's State of Sales report puts reps at roughly 28% of their time actually selling. The forecast is built from data entered in the remaining 72%, usually in a hurry, usually on a Friday.

A 75%-accurate forecast sounds tolerable until you translate it. On a 10 million kroner quarter, that is a 2.5 million kroner error in either direction — enough to change a hiring plan, a runway calculation, or a board conversation. Accuracy is not a vanity metric; it is the input to every other decision the company makes.

Five CRM fields that lie to you

Here is the part most forecast-hygiene advice gets wrong. It treats bad data as a compliance problem — reps are not updating the CRM properly, so tighten the process. But look closely at the fields the forecast actually rolls up from, and you find something more specific: each one asks the rep for a fact they do not have, so the rep supplies a plausible substitute instead.

1. Close date

The close date is supposed to be when the buyer will sign. In practice it is the date the rep needs it to be. Close dates cluster with suspicious enthusiasm on the last week of the quarter, and they move in clean 30-day increments, because a rep asked to update a stale date will reach for the next round number rather than re-open a conversation with the buyer.

The tell: a close date that has moved twice and has always landed in the final two weeks of a quarter. That is not a forecast; it is a rep's hope, timestamped.

2. Next step

The next-step field is the single most useful thing in a CRM and the single most abused. A real next step has three parts: a named person, a verb they agreed to, and a date. What the field usually contains is “follow up” or “check in” — which are things the rep will do, not things the buyer committed to.

Gong's analysis of sales calls found that close rates fall sharply when next steps are not concretely discussed on the first call. That is a finding about the conversation, not about the field. The field is downstream evidence of whether the conversation happened.

The tell: a next step with no buyer name in it. If the only person who has to do anything is your rep, the deal is not moving.

3. Stage

Stage is meant to encode buyer commitment. Almost every stage definition I have seen encodes seller activity instead: “demo delivered”, “proposal sent”, “negotiation”. All three can be true of a deal where the buyer has done nothing at all.

The correction is unglamorous: define every stage by something the buyer did. Not “proposal sent” but “buyer confirmed the proposal matches their evaluation criteria”. Not “negotiation” but “buyer named the approver and the signature path”. Stages defined this way cannot be advanced from the rep's side of the table, which is the entire point.

The tell: a deal that advanced a stage on a day with no buyer contact on the calendar.

4. Probability

In most CRMs, probability is derived from stage. Stage 4 is 60%. This is circular: the forecast is weighted by a number that is a restatement of a field the rep already controls, dressed up as a judgement. Two deals in stage 4 — one with a signed security review and a named approver, one where the champion has not replied in three weeks — carry identical weight in the roll-up.

The tell: your probability values are a lookup table with fewer than eight distinct values across the whole pipeline. Real probability estimates are messy; lookup tables are tidy.

5. Amount

Amount is usually priced off the first serious conversation and then never revisited, even as scope moves. Deals shrink quietly — a module drops out, the pilot goes from 40 seats to 15, the three-year term becomes annual — and the CRM keeps carrying the original figure because nobody edits a number downward without being asked to.

The tell: an amount that has not changed since the opportunity was created, on a deal with more than four buyer meetings. Scope always moves across four meetings. If the number did not, the number is stale.

Why discipline is not the fix

The standard response to all of this is a hygiene initiative. New field validation, a mandatory next-step format, a Friday deadline, a dashboard that shames the stragglers. These work for about six weeks.

They fail for the same reason knowledge bases fail, which we have written about at length in why your best rep's knowledge disappears when they leave. Voluntary data entry loses to the next meeting on the calendar, every single time, and it should — the rep who spends twenty minutes tidying fields instead of preparing for a call is making the wrong trade for the business, not the right one.

There is also an incentive problem that no amount of process design removes. Reps are asked to enter, into a system their manager inspects, the information most likely to be used against them. Expecting rigorous self-reporting under those conditions is not a discipline strategy; it is a hope.

The way out is not better self-reporting. It is to stop asking the field to be the source of truth, and start reconciling it against something the rep did not type. If you want the fuller argument for what that looks like across a whole team, our page for sales teams walks through it.

The commitment test: three questions per committed deal

Here is the framework we use, and it takes about ninety seconds per deal. For every deal in commit, answer three questions with evidence from a conversation — not from the CRM, and not from the rep's recollection in the moment.

1. Who said the close date, and when? If the answer is “the rep”, the date is a plan. If it is “the buyer, on the call three weeks ago”, ask what has been confirmed since. Dates decay faster than people expect.
2. What did the buyer commit to do next, in their words, by when? Not what your rep will do — what the buyer said out loud they would do. A deal where the buyer has committed to nothing is not a commit-stage deal, regardless of how well the last demo went.
3. What has changed in the account since the last conversation? Not in the CRM — in the account. New stakeholder in the thread, a reorganisation, a competitor named for the first time, a procurement process nobody mentioned in month one. We wrote about the specific things worth listening for in five pipeline signals hiding in your sales conversations.

Run this on ten commit deals and you will usually reclassify two or three. That is the whole exercise. You do not need a new forecasting methodology; you need one reconciliation pass where assertions get checked against evidence.

How we think about it at Floral

I will be specific rather than vague, because the vague version of this paragraph is what most vendors write.

Floral records and transcribes meetings, and turns each one into a structured summary with three parts: what was discussed, what was agreed, and the resulting to-dos. A human approves that summary, and on approval it syncs to the CRM. The practical effect on forecasting is narrow but real: the next-step field gets written from what was actually said in the room, on the day it was said, rather than reconstructed from memory the following Friday. Commitments made by the buyer end up recorded as commitments, with owners.

What that does not do is score your deals or predict your quarter. We do not claim to. The forecast still requires a manager who asks the three questions above. What changes is that the answers are checkable in under a minute instead of requiring a conversation with the rep who owns the deal — and that they survive that rep leaving.

The uncomfortable summary

Your forecast is not wrong because your team lacks rigour. It is wrong because five fields ask reps for certainty they do not have, and reps — reasonably, professionally, in good faith — fill them with the most defensible available guess. Then the roll-up treats those guesses as measurements.

You can improve accuracy meaningfully without changing your CRM, your process, or your team. Redefine your stages around buyer actions, kill derived probability, and run the commitment test on every commit deal once a week. Most teams find a few points of accuracy in the first month simply from the reclassifications.

If you want to see what the reconciliation looks like when the evidence comes from your own meetings rather than from a rep's recall, book a demo and bring three deals you are unsure about. Those are the interesting ones.