The estimate that became a promise
The scene repeats everywhere. Someone asks how long it will take. The team looks at the work and says three weeks. Someone more senior mentions that leadership has already said two. And then the part that matters: the team does not argue. It revises its own number downward and says it can probably try.
Nobody lied in that room. Nobody was incompetent. What happened is that the question changed halfway through, and almost nobody noticed.
What the word "estimate" starts to mean
A study presented at CHASE in 2021 followed how estimation actually works inside a software company. The authors describe a three step process: each person gives an individual number, the team settles on an internal number, and the team leader converts that internal number into a commitment to the customer.
The finding lives in that third step. When the leader knows the number will have to be defended in front of someone, the estimation session stops searching for the most likely figure and starts searching for the most defensible one. And the estimators, sensing this, revise their own numbers toward more optimistic values before anyone applies pressure.
That is the uncomfortable part. The pressure does not need to be applied. It only needs to be anticipated.
The number that comes out is still called an estimate. But it no longer answers "how long will this probably take." It answers "what number can I hold up in a difficult conversation." Those are different questions and they produce different answers.
Where the buffer lives
The same research examines padding, the amount someone adds on top of the value they honestly believe is required. What makes it interesting is that padding is not one behavior. It is three.
It can be a contingency buffer, protection against risk and uncertainty. It can be room to finish work that could not be padded at all, including work carried over from earlier projects. And it can be disguised budget for testing, cleanup and quality that nobody funded directly.
The authors frame this as a way of buying time, balancing short term and long term commitments. They also name the cost: padding hides that balancing act from the customer.
But there is a worse variant, and it shows up in the study. Leaders sometimes add the buffer without telling the estimators. The person who will do the work gets held accountable for a number they did not give, inflated by someone else, without knowing it was inflated.
That is where a mechanism turns into a culture. The developer learns that their honest figure was never the point. Next time, they supply a different kind of figure.
What the team learns in two cycles
It is worth looking at the incentive facing whoever estimates.
If the low number is met with relief and becomes a date, while the high number gets challenged in front of everyone, the team has just received a clear instruction. No policy required. No meeting required. Two rounds is enough.
After that you still receive numbers. They are simply no longer information about the work. They are information about what the person thinks you want to hear, which is far less useful for planning anything.
The symptom is easy to spot. When nobody disagrees with an estimate out loud anymore, the process did not become mature. It became quiet.
The criterion I use
The separation that fixes most of this is small and tedious to maintain.
The estimate is what the team believes. The commitment is what the business decides to promise after weighing that belief against everything else it knows: competitors, contracts, cash, a market window. Both are legitimate. Both can exist on the same project.
What cannot happen is one silently becoming the other somewhere between the team meeting and the slide.
In practice that means three things. The estimate leaves as a range rather than a point, and the range arrives intact at whoever receives it. The buffer, if there is one, is known to the people doing the work. And when the business decides to promise a shorter date than the team estimated, that gets said out loud, in the room, as a business decision.
That last part changes the climate more than the rest. Shortening a deadline for commercial reasons is a defensible choice. The choice was never the problem. The problem is it happening without an owner, so that the gap between estimated and promised shows up later as the team's execution failure.
When the opposite is the right answer
Not every context supports this separation, and pretending otherwise is just a different way to be wrong.
Some work does not accept a range. Fixed bids, sealed proposals, contracts with penalty clauses, an integration that must be live the day a partner launches. There the number is a commitment from minute one, and maintaining two artifacts creates paperwork without changing anything. What helps instead is cutting scope until the commitment fits, rather than pretending the estimate fit.
There is also the case where a silent buffer is the correct call. A new team, an unstable context, a customer who treats any range as an invitation to negotiate the floor. An experienced lead protecting the team with a number rounded generously upward is doing management, not manipulation. The difference is who the buffer is hidden from. Hiding it from the customer is negotiation. Hiding it from the people doing the work is what produces the damage described above.
And there is the simplest case of all: small, well understood work you have done ten times. There the estimate is closer to measurement, and this entire discussion is ceremony.
What I look for now
When I join a team and want to know whether estimates mean anything there, I do not look at the technique. Points, days, shirt sizes or gut feel barely matters.
I look at what happens to the person who gives the high number.
If they can explain their reasoning and leave the meeting without a social cost, that team's numbers probably carry information. If they have to justify themselves while the person who estimated low does not, the numbers have already become something else wearing the same name.
What happens on your team when an estimate comes in higher than the date leadership already announced?