How to Build a Sales Forecast You Can Actually Defend

Forecasting is arithmetic on your pipeline. If the pipeline is unreliable, no method rescues it.

Before forecasting, confirm:

• Every open deal has a value and a close date
• Stages are defined by buyer actions, not seller optimism
• Deals that have stalled are flagged, not quietly counted
• Closed Lost deals carry a reason code

If those aren't true, fix the pipeline first. (See How to Set Up Sales Pipeline Stages.) Everything below assumes them.

The default, and a reasonable starting point.

Forecast = Σ (deal value × stage probability)

An illustrative example:

The critical rule: derive probabilities from your own closed-won history, not from CRM defaults. Of every deal that reached Proposal in the last twelve months, what percentage closed? That's your Proposal probability. Most teams find their real numbers are considerably lower than the defaults suggest.

Where it breaks: it assumes close dates are accurate. They usually aren't — reps set them optimistically, and the same deal slides from March to April to May while still counting in each month's forecast.

Best for: teams with enough closed-deal history to derive honest probabilities, and reasonably consistent deal sizes.

Ignore individual deals. Look at what you've actually closed.

Take your last several quarters of closed-won revenue, establish the trend, adjust for seasonality and known changes (a new rep ramping, a product launch, a large renewal).

Where it breaks: it's blind to what's in the pipeline right now. If your pipeline just doubled or collapsed, run-rate won't know until it's too late.

Best for: established teams with stable motions. Also the best sanity check on any other method — if your weighted pipeline says you'll do triple your historical best quarter, one of the two is wrong.

Ask each rep to categorise their deals:

• Commit — will close this period. Reputation on it.
• Best case — could close if things go well.
• Pipeline — everything else.

Forecast the Commit, treat Best Case as upside.

Where it breaks: it depends entirely on how honest and well-calibrated your reps are, which varies by person and by whether they're above or below quota.

Best for: small teams where a manager knows each rep's bias personally — and even then, alongside another method rather than alone.

The strongest practice is to run weighted pipeline and historical run-rate in parallel.

When they agree, you have a defensible number. When they diverge sharply, that divergence is itself the most useful signal available — something has changed, and you now know to go find out what before you commit to a number.

Present both to leadership with the gap explained. A forecast that shows its working survives scrutiny; a single number with no method behind it doesn't.

Almost nobody does this, and it's the difference between forecasting and guessing.

Forecast accuracy = (Actual ÷ Forecast) × 100

Track it every period. Over a few quarters you learn things no method gives you:

• Consistent over-forecasting? Your stage probabilities are too generous, or close dates are optimistic. Adjust the probabilities down.
• Consistent under-forecasting? You're being too conservative — usually over-correction from a previous miss.
• Erratic? Your pipeline data quality is the problem, not your method.

Track it per rep too. Some people are structurally optimistic; some sandbag. Knowing individual bias lets you adjust intelligently rather than applying a blanket haircut to everyone — which punishes your accurate forecasters.

Forecasting on close dates nobody maintains. If a deal's close date has been pushed three times, it isn't a Q3 deal. Enforce that pushing a date requires a reason.

Counting deals with no recent activity. A deal with nothing logged in six weeks is not in your forecast, whatever its stage says. Time-in-stage flags exist for this.

Ignoring the deal you'd rather not think about. The large deal that's been “closing next week” for two months distorts everything. Downgrade it explicitly.

Forecasting bookings when the business needs cash. A signed annual contract and cash in the bank are different events. Know which one your CFO is planning against.

One number, no range. Give a range — conservative, expected, optimistic. It's more honest and more useful than false precision.

Weekly: review deals that moved, deals that stalled, close dates that shifted.

Monthly: rebuild the forecast; compare against run-rate; log the variance from last month's forecast.

Quarterly: measure accuracy, recalibrate stage probabilities from the quarter's actual close rates, review whether stage definitions still match how you sell.

The recalibration step is what compounds. A forecasting process that learns from its own misses gets meaningfully better within a year. One that doesn't stays a guess indefinitely.

Worth saying plainly: the hard part of forecasting isn't the maths. Every method above is arithmetic you could do in a spreadsheet.

The hard part is having pipeline data that's current, complete, and honest — which is a function of whether your reps maintain the CRM, which is a function of whether the CRM helps them or taxes them. (More on that: What Is a CRM?)

Teams that forecast well aren't better at forecasting. They have cleaner inputs.