I've owned demand planning across an entire SKU portfolio at nine-figure revenue. I can tell you with confidence that the hardest part of forecasting has almost nothing to do with forecasting.

The math is the easy part. The hard part is that a forecast is a set of promises made by people with competing incentives, and most companies never address that directly.

Why your sales forecast is wrong in a specific direction

Sales forecasts are not wrong randomly. They're wrong predictably, and the direction tells you exactly which incentive is driving them.

If your team is measured on hitting quota, forecasts come in low. Sandbagging is rational — beat a modest number and you look good. If your team is measured on pipeline or enthusiasm, forecasts come in high. Optimism is rational — big numbers signal ambition and nobody audits them until it's too late.

Either way the number you're planning inventory against isn't a forecast. It's a negotiating position.

The fix isn't a better model. It's separating the two things the number is being asked to do:

  • The commitment — what sales will be held to.
  • The expectation — what operations should actually plan against.

These are different numbers with different owners, and pretending they're one number is why the whole system limps. Once you separate them, you can hold sales accountable to the commitment while planning honestly against the expectation. The conversation stops being a negotiation and starts being a plan.

Forecast the decision, not the number

Most forecasting effort is spent trying to make a number more accurate. That's usually the wrong optimization.

The question isn't "what will we sell?" It's "what decision does this number drive, and how wrong can I be before that decision changes?"

Those are wildly different problems. If a SKU has a four-week lead time and shelf life measured in years, being 30% off costs you almost nothing — you correct next cycle. If it has a sixteen-week lead time, seasonal demand, and a hard sell-through window, being 10% off can cost you the season in both directions.

So stop forecasting every SKU with equal rigor. Segment by consequence of error, not by revenue. The items that deserve real analytical attention are the ones where being wrong is expensive and slow to correct. Everything else deserves a simple rule and your inattention.

Most planning teams have this exactly backwards — they lavish attention on the big revenue items, which are usually the ones with the most stable, most predictable demand.

Lead time is the real variable

If I could get companies to instrument one thing, it wouldn't be forecast accuracy. It would be lead time — and specifically lead time variability.

A supplier with a twelve-week lead time that's twelve weeks every single time is easy to plan around. A supplier averaging eight weeks that ranges from four to sixteen is far more expensive, even though the average looks better. You end up carrying safety stock to cover the worst case, which means you're financing their inconsistency out of your working capital.

Track the variance, not just the average. Then have the conversation with that supplier using their own numbers. In my experience it's one of the highest-return conversations available to an operator, and it almost never happens, because nobody's measuring it.

The meeting that makes it real

Forecasting only works if there's a recurring moment where sales, purchasing, and finance are in the same room looking at the same numbers and are each accountable for their piece.

Not a report that circulates. A meeting where variance gets explained by the person who owns it.

The agenda barely matters. What matters is that the same three questions get asked every cycle:

  1. Where were we wrong last period, and why — specifically?
  2. What's changed since we set this number?
  3. What are we committing to now, and who owns it?

Question one is the one everyone wants to skip. It's also the only one that creates learning. A forecast process with no retrospective isn't a process, it's a ritual.

What good actually looks like

A mature forecasting discipline isn't one that's frequently right. It's one that is:

  • Wrong in known directions, so you can correct systematically instead of reacting.
  • Fast to detect error, so a bad assumption costs you weeks instead of a season.
  • Owned by a name, not by a spreadsheet or a committee.
  • Segmented by consequence, so effort lands where error is expensive.

I'd take a team that's consistently 15% off and knows exactly why over a team that's occasionally perfect and can't explain either outcome.

The first team is running a system. The second is getting lucky, and luck doesn't survive scale.

The leadership part

Here's what makes this a leadership problem rather than a planning one.

Every failure mode above is really a failure to make ownership explicit. Sandbagging happens when nobody's accountable for the expectation. Lead-time variance goes unmeasured when nobody owns supplier performance. The retrospective gets skipped when leadership treats being wrong as a thing to be embarrassed about instead of a thing to be curious about.

You cannot fix any of that with a better model. You fix it by deciding who owns what, insisting on honest numbers, and making it safe to be wrong out loud — while making it very unsafe to be wrong quietly.

That's not a spreadsheet problem. That's the job.