It Already Works, Just Finish It

Six words, and your team will spend the next quarter losing an argument to them.

Not to a person — to a sentence. Nobody has to defend it, it never appears in a document, and it got set the moment something worked on a screen. From then on, every estimate your team gives is heard as a negotiation against a number that nobody wrote down and nobody owns.

I call it the unwritten anchor, and it's the most expensive thing in the room.

Watch it form

Somebody demos. The room relaxes, visibly and physically, in the way rooms do when a risk they'd been carrying turns out to be smaller than they thought. Someone says "oh, that's further along than I expected." A director does the arithmetic out loud and asks whether we're talking a few weeks.

Nothing was decided. But a number now lives in eight people's heads, and it's the only number in the room with no name attached to it.

Then your engineer says four months, and she isn't heard as estimating. She's heard as resisting.

That's the whole failure, right there in the gap between what she said and what the room received. The instant the anchor sets, honesty starts sounding like obstruction, and your best people work out very quickly that the accurate number is the socially expensive one. So they round toward the room, and they absorb the difference in weekends nobody sees.

You've met the second-order version of this even if you've never named it: a team that quotes what the room wants to hear and quietly hopes. Is that a courage problem? It looks like one from the outside. It isn't. It's what you get when the only available alternative to the anchor is an argument, and Amy Edmondson has spent a career showing that people don't speak up in rooms where speaking up costs more than staying quiet.

An estimate delivered after the anchor isn't information. It's a complaint.

What I got wrong at the table

Early in my career I got my first real seat at the product leadership table, and I could not get a single item onto the roadmap.

The director of sales wanted to build something he'd already presold to customers. I had a data-driven alternative, and — this is the part that still stings — I was right. The feature we built didn't move sales and it didn't strengthen one existing relationship. We lost months on it.

Being right cost me nothing and bought me nothing.

He beat me because he arrived with a story and I arrived with a spreadsheet, and because I showed up after the room had already settled on the shape of the thing. We both knew whose fault that was. I was playing at the big-boy table and I hadn't come prepared — not on the analysis, which was fine, but on the delivery, which was the only part that mattered that day.

It took me years to understand that I hadn't lost for lack of evidence. I lost because I turned up holding numbers and asked eight people to feel worse about a decision they'd already made emotionally, which is a thing that has never once worked in the history of meetings.

Data arrives too late by default. Somebody has to move it earlier, and in your org that somebody is almost certainly you.

The remainder list

So move it — before the yes, not after.

The remainder is everything the demo skipped. It always exists, and it's always invisible, because not showing it is the entire point of a demo — nobody builds a prototype to showcase the permissions model. Your job is to put the remainder on the table with a name on it, and that takes three columns and about twenty minutes.

What it skipped, named specifically. Not "hardening," which means nothing and gets discounted to zero. Say "no permissions model, sample data only, no eval harness for output that changes every run." Vague remainders get waved through. Specific ones get scheduled.

Who owns each one, as a person rather than a team. "Platform will handle it" is the precise phrase by which a remainder stays invisible for two quarters and then becomes somebody's resignation.

What each one costs, roughly. You are not forecasting here and nobody will hold you to the figure. You're refusing to let zero be the default, because zero is what the room is currently using.

One page, three columns, and it goes up with the demo — same meeting, not the follow-up, because by the follow-up the anchor has already hardened and you're back to complaining.

A remainder you've written down is a schedule. A remainder you haven't is somebody's unpaid overtime.

Something changes in the room when this exists. Your team stops arguing against a phantom and starts reading from a document, and nobody has to be brave. That's the actual point of the exercise. Nobody should need courage to say a true thing about a schedule.

The number that tells you whether this is your problem

Go and count your last five commitments made off a working demo, and for each one mark whether the estimate got revised upward after the yes was given.

So what number did you come back with? Three or more and your organization isn't estimating, it's anchoring and then correcting — and every correction quietly spends a little of your team's credibility on a number they never chose in the first place.

That's the cost most leaders miss, because schedule slips are visible and everybody sees those. What erodes silently is the belief that when your team says four months, they mean four months. Once that belief is gone you don't get it back with a better process, a new template, or a workshop about estimation. You get it back by never letting them be wrong on somebody else's number again.

Do this before your next demo

Find the next AI demo on your calendar, which is probably this week.

Before it happens, ask whoever built it a single question: what did you skip to get this working? Write the answer down in their words rather than yours, because their words are more specific than anything you'd paraphrase, and carry that page into the room. Put it up the moment the demo ends.

You're not slowing anything down. You're making sure the yes and the remainder arrive at the same time, in front of the same people, while everyone's still in the mood to be reasonable.

Which of your teams stopped giving you real numbers — and how long have you been reading that as attitude?