I gave the estimate on a Friday afternoon. A good one. We’d closed out a tricky piece of work that week, the team was in a light mood, and someone asked how long the next feature would take.
I said two weeks. I said it easily. It took closer to six.
Nothing dramatic went wrong. No outage, no surprise rewrite. Just a steady stream of small things nobody had thought about, one after the other, until the date was a memory and the conversation about it was mine to have.
The number came from the room
When I went back to that moment, I tried to remember how I’d arrived at two weeks. I couldn’t. There was no breakdown, no list, no quick look at the code. There was a feeling of momentum, and the number fell out of it.
That’s the part I didn’t like admitting. The estimate wasn’t really about the feature. It was about Friday. It described how capable I felt after a week that had gone well.
I’ve since caught myself doing the opposite too. On a draining Tuesday, after a long call and a failed deploy, someone asks for a rough idea, and I pad it so heavily that the work looks like a project. Same kind of feature. Very different number. The only thing that had changed was me.
Optimism isn’t the problem
The usual advice is that we’re all too optimistic, so multiply by some factor and move on. I tried that for years. It doesn’t fix anything, because the multiplier gets applied to a number that was never grounded in the first place.
What I’d actually been doing was estimating the version of myself that existed in that moment. Rested me. Tired me. Me with a clean calendar in my head, even though the real calendar had a workshop, two client calls and a holiday in it.
The work doesn’t get done by that version. It gets done by the ordinary one, across ordinary weeks, with interruptions and a team that has its own week going on.
Why I answered so fast
I’ve asked myself why I give numbers on the spot at all. The honest answer is that a quick estimate feels like competence.
When someone asks how long, and you answer straight away, you sound like someone who knows the work. When you say you’ll need to look first, it can feel like hesitation, especially in front of people who are waiting on you.
So I answered fast. And the speed was the thing being performed, not the accuracy. Nobody in that Friday conversation needed the number in the next ten seconds. They needed a number they could plan around. I gave them the first and let them believe it was the second.
What the miss actually cost
The six weeks themselves weren’t the worst of it. Work takes what it takes.
The cost was everything that got built on top of the two weeks. Someone had lined up the next thing to start straight after. A conversation with a client had been shaped around it. I spent a chunk of each of those extra weeks re-explaining, re-planning and apologising, which made the work slower still.
And there was a quieter cost. After the second or third time a confident number of mine slipped, people started adjusting it themselves. They’d hear my estimate and add their own margin without telling me. Which meant my number had stopped being information. It had become something to be corrected.
That one stung more than the delay.
What I do now
None of this is a method. It’s a few habits that make it harder for my mood to sneak into the number.
I don’t estimate in the moment anymore. If someone asks, I give a range and a time when I’ll come back with something firmer. Usually the next morning. The answer gets less exciting overnight, and that’s the point.
I write down what I’d have to believe. Before I commit to a number, I list the assumptions under it. The API behaves the way the docs say. Nobody on the team is out. The client signs off in one round. Reading that list back, I can usually see which line is wishful, and the number moves.
I check the calendar, not my head. Not the ideal week. The actual one, with the meetings and the days I’m already promised elsewhere. It’s a dull step. It has corrected more of my estimates than anything clever I’ve tried.
I ask someone who wasn’t in the room. Not to estimate for me, just to look at the list and tell me what’s missing. They’re not carrying my Friday feeling. They see the gaps I’m stepping over.
I still get it wrong. Last month I gave a careful, considered, morning-after estimate, and still missed it, because a dependency I’d listed as an assumption turned out to be false in week one. But I could see where the miss came from, and I could say so early, instead of discovering it at the deadline along with everyone else.
An estimate is a sentence about the future
I used to think estimating was a skill you get better at with experience. Some of it is. But a lot of what I’d called experience was just confidence, and confidence is the very thing that bends the number.
What I believe now is simpler. An estimate is a small promise about how someone else’s next few weeks will go. It deserves the same care as anything else I put my name on. Not more time than it needs. Just not less.
If you want to see whether this applies to you, look at your last few estimates and try to remember what kind of day you gave them on.
You may find the work had less to do with it than you thought.