Ask the question you'd rather skip.

Dispatch · to the desk of one working leader

Ask the question you'd rather skip.

The observation

A sales director closes the quarterly forecast on a Thursday and sends it up. Eighty-two percent to plan, with three deals carrying the difference.

The number came from her reps, who work across several markets. Each of them put a percentage against their own deals, and she rolled the percentages together the way she does every quarter.

What she doesn't have is what the percentages meant.

For one of them, seventy percent means the buyer has budget, a signed-off business case, and a date in the diary. For another, seventy percent means the last conversation went well and he'd rather not be the one putting a low number in front of his director in week ten.

Both of them wrote seventy, and neither was careless.

She rolls them up, and the roll-up comes out confident in a way that neither of the numbers underneath it ever was.

Three weeks later, two of the three deals slip. She'll spend the next quarter tightening the forecasting process.

She had a precise answer to a question her team never agreed on.

The pattern

Every forecast is a translation before it's a number.

A number carries less than it looks like it does. The word behind it doesn't survive the trip from one person to the next, and it survives even less when the people at each end work in different languages.

Three researchers at Worcester Polytechnic Institute ran an experiment across English, French, German, Mandarin and Arabic. Fifty in each language, two hundred and fifty in total. Each was shown eighteen ordinary probability expressions, the kind that fill a pipeline review. Likely. Probable. Better than even. Then they were asked to draw what each one meant, filling in a hundred-square grid.

The translations were careful. Three speakers translated each expression independently and disagreements were mediated to agreement before the study ran.

Then the results came back and there was no clean mapping between any two languages.

Of the eighteen expressions, five landed in the same place everywhere. The rest moved. On the widest of them, the gap between where one group of speakers put the word and where another put it ran to twenty-six points of a hundred, and several more ran past twenty.

A twenty-six point gap is the difference between a deal you staff and a deal you let go.

Your reps hand you digits. Each digit started as a word in somebody's head, and the conversion happened before anything reached you.

Ask any of those participants whether they had understood the word and every one of them would have said yes, and been right.

"I said what I meant. It's right there in the sentence."

You ask for an honest number. What comes back is a number, and a number looks identical whatever produced it. The rep who counted a signed business case and the rep who counted a good meeting both hand you two digits, and the two digits match.

No number leaves you uncertain and still asking. A precise one closes the question.

Give AI the pipeline and it returns a weighted summary in seconds, and it will be a good one. It works from what your reps typed. What they meant by it is the one input you have to go and collect yourself.

Cheap precision makes bad precision scale. Every quarter it hands you a tighter, faster, more confident version of the same disagreement, and the part where two reps discover they meant different things is still done one conversation at a time.

Hold this line

When a number comes back clean and confident, don't act on it yet. First find out what the people who produced it thought they were being asked.

The move

This week, take one number you're about to act on. A forecast, or a score you've learned to trust.

Before you build anything on it, ask two or three of the people who produced it. Tell them you're calibrating your own reading, then ask:

"What would move that seventy to a fifty, and what would move it to a ninety?"

Ask them one at a time, so each reading arrives on its own.

The brackets do the work the number can't. For one rep a fifty is a signature that hasn't come back yet, and for another it's a meeting that felt flat. Once you can see both, you know what you're holding.

If your team spans markets, ask what would count as a confident number where they work. How much certainty a number is expected to carry moves from one place to the next, and the convention travels silently.

It's slow and it doesn't scale, and that's what your hours are for. AI hands the roll-up back before you've sat down, and three conversations for one number is what to do with the time it frees. You can't hand those to anybody else. What's being tested is whether your reading matches theirs.

And you'll sometimes find out that a number you already acted on meant something other than what you read into it. That is what the exercise is for, and it's uncomfortable in a way a dashboard never is.

If you ask and then go with your first reading anyway, because it's tidier and already in the deck, you've performed the conversation. They notice. And next time you'll get the version that's easier to say.

Don't let a clean number end the conversation that would have told you the truth.

Two reps can write the same number and mean different quarters.
You cannot evolve faster than you can listen.

What is one number your team reports that you have never once questioned? One line. I read every reply.

See you Tuesday.

Rui

P.S.  Worth your margin: Rakotondravony, Ding and Harrison, "Probablement, Wahrscheinlich, Likely?" Two hundred and fifty people across five languages drawing what "likely" means. The charts are the whole argument.

If someone came to mind while you were reading this, send it to them.

Someone forward this to you? Margin. is free, five minutes, every Tuesday.

Why can two reps write the same number and mean different things?

Because the number is the easy part. For one rep, seventy percent means the buyer has budget, a signed-off business case and a date in the diary. For another it means the last conversation went well and he'd rather not be the one putting a low number in front of his director in week ten. Both wrote seventy, and neither was careless. The roll-up comes out confident in a way that neither of the numbers underneath it ever was.

What did the cross-language study actually find?

Three researchers at Worcester Polytechnic Institute ran an experiment across English, French, German, Mandarin and Arabic. Fifty in each language, two hundred and fifty in total. Each was shown eighteen ordinary probability expressions, then asked to draw what each one meant on a hundred-square grid. Of the eighteen, five landed in the same place everywhere. The rest moved. On the widest of them, the gap between where one group of speakers put the word and where another put it ran to twenty-six points of a hundred, and several more ran past twenty. The translations were careful: three speakers translated each expression independently and disagreements were mediated to agreement before the study ran.

What does a study about words have to do with a forecast?

Your reps hand you digits. Each digit started as a word in somebody's head, and the conversion happened before anything reached you. A twenty-six point gap is the difference between a deal you staff and a deal you let go.

What does AI change here?

Give AI the pipeline and it returns a weighted summary in seconds, and it will be a good one. It works from what your reps typed. What they meant by it is the one input you have to go and collect yourself. Cheap precision makes bad precision scale: every quarter it hands you a tighter, faster, more confident version of the same disagreement, and the part where two reps discover they meant different things is still done one conversation at a time.

What should a leader actually do about it?

Take one number you are about to act on, and go to two or three of the people who produced it. Tell them you are calibrating your own reading, then ask what would move that seventy to a fifty, and what would move it to a ninety. Ask them one at a time, so each reading arrives on its own. The brackets do the work the number can't. If your team spans markets, ask what would count as a confident number where they work, because how much certainty a number is expected to carry moves from one place to the next.