Your forecast isn’t broken. Your assumptions are drunk.

A forecast does not become trustworthy because the model calculates to four decimal places.

I wish it did.

Excel would have solved management uncertainty sometime around 1998.

The difficult part of forecasting is not arithmetic. It is deciding which assumptions deserve to be believed, for how long, and what evidence should make us change them.

That is where forecasts usually get into trouble.

I separate the model from the forecast

The model is the machinery.

The forecast is management’s current view of what is likely to happen.

A model can be excellent while the forecast is weak because the assumptions are stale, optimistic or disconnected from operating reality.

I make this distinction because teams often respond to a forecast miss by rebuilding formulas.

Sometimes the formulas are fine.

The business simply stopped behaving like the assumptions.

Pipeline coverage is not a revenue forecast

“We have 4x pipeline” sounds reassuring.

I immediately want to know the composition.

Stage quality. Deal size. Age. source. Rep. close-date movement. Historical conversion by stage. How much pipeline was created recently versus carried forward.

Four times a target in weak pipeline can be less valuable than two times in high-quality late-stage opportunities.

I do not dislike coverage ratios.

I dislike when a summary ratio replaces the underlying behavior.

Win rates need a denominator everyone understands

Win rate can mean wins divided by closed opportunities, wins divided by all created opportunities, or something else entirely.

Change the denominator and the assumption changes.

Then segment mix changes.

Enterprise deals convert differently from SMB. Inbound may behave differently from outbound. New reps behave differently from tenured reps.

A historical blended win rate can be mathematically accurate and operationally irrelevant to the current pipeline.

Close dates have personalities

Some opportunities move once for legitimate reasons.

Some have spent six months moving from “this month” to “next month” with the determination of a gym membership in January.

I like measuring date slippage.

If an opportunity’s close date has changed four times, I do not want the forecast treating the current date with the same confidence as a newly contracted implementation date.

Sales judgment matters.

So does the opportunity’s history.

Churn assumptions should know who is renewing

A flat monthly churn rate is convenient.

Renewal exposure is rarely flat.

If several large customers renew in Q3, that quarter carries different risk from a quarter dominated by smaller contracts.

I want material renewals modeled explicitly and broader populations handled statistically where appropriate.

The method should follow materiality, not the desire to use one formula everywhere.

Headcount timing can make a forecast look artificially good

Hiring slips. Payroll comes in favorable. EBITDA improves.

Everyone enjoys the variance.

Then we remember the hires were supposed to sell, build or support something.

A delayed role can reduce expense and reduce future capacity.

I want both consequences in the model where the relationship is material.

Otherwise the forecast rewards itself for failing to execute the operating plan.

Gross margin assumptions deserve operating drivers

“Margin improves 200 basis points” is not a driver.

Why?

Pricing? Product mix? Vendor terms? utilization? automation? labor mix?

I want to know the mechanism.

If management cannot name the operating change, I am reluctant to book the financial improvement.

A forecast should translate actions into economics, not use economics as a substitute for actions.

Working capital can disagree with the P&L

A forecast can hit revenue and EBITDA and still miss cash badly.

Collections slow. Inventory builds. Customers negotiate longer terms. Vendors require deposits.

I like making DSO, inventory and payment assumptions visible enough to challenge.

Cash has a habit of finding assumptions the income statement managed to hide.

I distinguish facts, trends and bets

A signed lease payment is close to a fact.

A historical collection pattern is a trend.

A new product reaching $5 million of revenue in six months may be a management bet.

All three can belong in the forecast.

I just do not want them wearing the same level of certainty.

Knowing where judgment enters the model tells me where scenario analysis and challenge matter most.

Assumption owners should be close to the information

Sales may own pipeline inputs. HR may own recruiting timing. Operations may own capacity. Finance owns the integrated forecast and challenge process.

Ownership does not mean Finance accepts every submitted number.

It means the person closest to the operating information has responsibility for explaining the assumption.

Then FP&A compares it with history, cross-functional constraints and financial consequences.

I want assumption history

If a launch date moved from March to May to August, I want to see that.

If the revenue forecast changed because conversion fell, preserve the prior assumption.

A forecast should remember how management’s view evolved.

That makes forecast-to-forecast analysis possible and helps the organization learn which assumptions are repeatedly optimistic or slow to update.

Overwriting history makes every new forecast look inevitable.

Forecast-to-forecast matters as much as budget-to-actual

Budget tells me what we once planned.

The prior forecast tells me what management believed recently.

If EBITDA moved $2 million since last month, I want the bridge.

Revenue timing, hiring, margin, working capital, discretionary spend.

That is the current decision story.

A forecast is most useful when it explains how our view changed as information changed.

I do not use forecast accuracy as a morality score

Forecasts will miss.

The question is why.

Was the event genuinely unknowable? Did new information arrive? Was the assumption stale? Did somebody know the risk and fail to surface it? Did the model systematically bias one direction?

Those are different problems.

I want accuracy metrics to improve the process, not turn every miss into a search for a guilty department.

Bad news needs to travel quickly

This may be the most important forecasting control and it is not in Excel.

If people believe lowering the forecast will create punishment, risks arrive late.

Sales keeps the deal in. Product keeps the launch date. Hiring remains on schedule in the model long after Recruiting knows it is not.

Finance cannot forecast information the organization will not say out loud.

A strong forecast process makes updating reality safer than defending an obsolete number.

Scenario planning is where uncertainty becomes useful

I do not need one forecast to contain every possible future.

I do want scenarios around the assumptions capable of changing a decision.

What if conversion falls five points? What if hiring slips a quarter? What if the large renewal churns? What if collections add ten days?

Then management can decide what it would do.

That is more valuable than arguing over whether the base case should be 7.2% or 7.4% growth.

I use ranges when the business is genuinely uncertain

Sometimes one number creates false confidence.

For a major uncertain deal, launch or financing event, a range can communicate the risk better.

The management forecast may still require a single base case for planning.

Fine.

I want the range visible around the decision so everyone remembers which part of the outlook is assumption-heavy.

Forecasts should become simpler as understanding improves

Teams sometimes respond to misses by adding detail.

Another tab. Another driver. Another scenario.

Detail can help.

It can also hide the fact that three assumptions explain 80% of the movement.

I like identifying the small set of drivers management actually needs to understand and giving those the strongest logic.

Complexity should earn its place.

The forecast should tell management where to look

I do not want a model that merely produces an income statement.

I want to know what changed, what matters, where risk increased and which assumptions deserve attention.

The output should help prioritize management conversation.

If every line gets equal formatting and every variance gets equal commentary, Finance has not finished the analysis.

A resilient forecast is not one that never misses

It is one that updates quickly when evidence changes, preserves the reason for the change, exposes the important assumptions and lets management understand the consequence.

I would rather have that than a model that hits one quarter perfectly because the world happened to cooperate.

Precision is useful.

Resilience is more useful.

The goal is not to make the forecast incapable of being wrong.

It is to make the organization faster at recognizing when reality has changed—and better at deciding what to do next.

Revenue assumptions should reconcile to capacity

A revenue forecast can be internally elegant and operationally impossible.

If the plan requires 40% more bookings but sales capacity grows 5%, something else has to explain the difference.

Higher productivity? Better conversion? Larger deals? Faster ramp?

I want that assumption named.

The same applies to production, implementation and customer support.

Financial growth should reconcile to the operating capacity required to produce it.

If it does not, the model is relying on an invisible productivity miracle.

Price, volume and mix should not be blended when they matter

Revenue can grow because the company sold more units, charged more, or sold a different mix.

Margin can change for the same reasons.

I like separating those drivers when they are material because they have different implications.

Price-led growth may be constrained by retention. Volume-led growth may require capacity. Mix can change margin even when total revenue hits plan.

A single growth percentage hides those tradeoffs.

Known events deserve explicit modeling

Historical run rates are useful until we know something history does not.

A contract terminates. A facility opens. A price increase starts. A major hire has accepted an offer. A capital project has a signed schedule.

I want those events modeled explicitly rather than diluted into a historical average.

Forecasting is not choosing between history and judgment.

It is using history until better information exists.

I watch for assumptions that move together

Downside scenarios can become too polite.

Revenue falls 10%, but gross margin, collections and customer retention remain unchanged.

Maybe.

In real stress, drivers can correlate.

Customers may pay slower. Discounting may increase. Utilization may fall. Vendor leverage may change.

I do not need a catastrophe model for every forecast.

I do want scenarios to reflect plausible relationships instead of changing one cell while the rest of the business politely stays still.

Forecast bias is worth measuring

A team that misses high and low randomly has an accuracy problem.

A team that is consistently optimistic has a behavioral problem in the process somewhere.

A team that is consistently conservative may also distort decisions by understating available capacity.

I like looking at bias by driver and by owner over time.

Not to create a leaderboard of shame.

To identify where assumptions systematically lean in one direction.

Materiality should determine forecasting effort

I do not need a custom driver for office supplies if payroll and revenue explain most of the outlook.

I do not need account-level renewal forecasts for 20,000 tiny customers if cohort behavior is stable.

I do need detail where one assumption can move cash, EBITDA or a management decision materially.

Forecasting quality is not proportional to cell count.

It is proportional to how well the model represents the decisions and risks that matter.

I like a short list of “what would change our mind?”

For the most important assumptions, I want leading evidence.

What would make us lower conversion? What would make us delay the hiring plan? What would make us change the churn assumption?

That turns the forecast into an active monitoring process.

Instead of waiting until actuals prove the assumption wrong, Finance watches the evidence that should move it.

This is one of the simplest ways to make a forecast more responsive without making the model more complicated.

The management conversation should end with decisions

A forecast review that spends 90 minutes admiring variances and ends with no decisions is reporting.

Sometimes reporting is all that is needed.

But if the outlook changed materially, I want to know what management will do differently.

Hire? Delay? Invest? Preserve cash? Challenge a sales assumption? Accept the risk?

The forecast earns its keep when the updated view changes a decision—or confirms that no change is needed.

That is why I care so much about assumptions.

They are not cells in a model.

They are management’s beliefs translated into financial consequences.