A consulting engagement begins with a number: this will take six weeks. Months later, someone asks why it took ten, and the room fills with explanations about scope, tools and people. Rarely does anyone ask the more useful question, which is why the six-week number felt so believable in the first place. Estimating beliefs are stubborn, and many of them are quietly wrong. Here are six common myths about estimating technical work, each set against what a simple record of guesses and actual results tends to show.
Myth 1: Experienced people estimate accurately
Years in the field do sharpen judgement about what a task involves, but they do not cure optimism. Seniority often makes people more confident, and confidence is not accuracy. A veteran engineer can still picture the smooth route through a migration and forget the three odd systems that nobody documented.
What the record shows: experienced people tend to be more accurate on work they have done many times and about as optimistic as everyone else on novel work. The fix is not to trust the senior person more, but to hold their numbers to the same comparison as anyone else’s: guess, actual, ratio.
Myth 2: A more detailed breakdown guarantees a better estimate
Splitting a project into fifty line items feels rigorous. It is helpful for visibility, but it can leave a false sense of precision. The items you did not think of are not in the list, and they are the ones that cause overruns. Breakdowns catch forgotten steps, yet they cannot catch forgotten problems.
What the record shows: detailed plans improve the ratio modestly and often leave a stubborn residue, typically a tail of unplanned work that shows up as a roughly steady percentage across projects. Track that percentage and add it as its own line, labelled honestly as unknowns.
Myth 3: Padding the estimate fixes the problem
Many teams quietly add twenty percent to every number. It feels safe and it hides the habit rather than curing it. Padding is applied evenly, but errors are not even: some task types run at nearly the guess while others double. A uniform pad overcharges the predictable work and still under-covers the risky work.
What the record shows: ratios differ sharply by kind of task. A calibrated factor per task type, based on a log, beats a blanket pad because it spends the margin where overruns actually occur.
Myth 4: Overruns mean the team is slow
A story from a training programme shows how wrong this can be. An instructor team scheduled six hours of hands-on labs for a workshop and finished at nine. The instinct was to blame the participants. When the team looked back at its notes, it found that lab setup alone had consumed two of the three extra hours, and the participants’ pace was almost exactly as predicted. The delay lived in the preparation, not the people. (The workshop is an illustrative composite, not a real event.)
What the record shows: when you split actual time into its parts, overruns usually trace to setup, waiting and re-work, rarely to slow execution. Without a breakdown of actuals, teams punish speed that was never the issue.
Myth 5: You cannot estimate work you have never done
It is true that first-time work is hard to size. It does not follow that estimating is pointless. What you can do is borrow a ratio from the nearest similar task and widen the range. If every first-time tool adoption in your log ran at about double the guess, your first guess for the next new tool should be doubled before anyone starts arguing.
What the record shows: novelty has a signature. Tasks involving an unfamiliar tool, an unfamiliar team or an unfamiliar environment cluster at the high end of the ratio range. Naming that cluster turns “we cannot know” into “we usually need twice as long”.
Myth 6: Calibration slows teams down
Logging guesses and actuals takes a minute or two per task. Teams worry that this overhead eats into delivery. Measured honestly, it pays for itself within a quarter. The minute spent logging is small beside the hours lost to a missed commitment, a rescheduled client meeting or a weekend of catch-up.
What the record shows: teams that calibrate typically see ratios settle toward 1 over a few months, and fewer last-minute reshuffles of the plan. The saved effort rarely appears on a timesheet, but people notice it in calmer planning meetings.
The reason these beliefs stick around
Every one of these beliefs is a comfortable one. Believing that experience guarantees accuracy lets a manager relax. Believing that padding solves everything lets a team stop looking. Believing that overruns mean slowness gives everyone a simple villain. Myths last because they save effort, and the cost of that saving arrives later, spread thin across missed dates that nobody connects to the original belief.
There is also a social reason. In many workplaces an estimate is treated as a promise, so changing one feels like admitting failure. A culture that treats estimates as guesses to be checked, not vows to be defended, makes it much easier to retire a myth. Consultants see this difference immediately when they join a new client: where the record is visible, plans get honest quickly; where it is hidden, the same surprises repeat every quarter.
How to test these myths against your own work
None of this needs to be taken on trust, and our own claims deserve the same scrutiny. Pick the last ten finished tasks from your tracker, find the original guess for each, and compute actual divided by guess. Then ask four questions. Did the senior people’s numbers do better than the others’? Did tasks with a long breakdown come closer to the guess? Which task types sit above 1.5? And how much of the overrun was setup or waiting?
Ten tasks will not prove anything, but they will tell you which myth your team believes most. That is the one to attack first, and the log will show whether the correction is working.
A short guide to talking about this with clients
Clients are rarely upset by a range; they are upset by a surprise. When you present an estimate, show the guess, say which task type it belongs to, and quote the ratio your records give for that type. A sentence such as “similar work has run about one and a half times our first guess, so we have planned for that” lands far better than a bare date. It shows that the number was checked against evidence, and it invites the client to share the uncertainty instead of resenting it later.
Reading wider
If you are a data or platform team, the same habit applies to machine-learning delivery. Two earlier posts here, one on why operational practice matters for machine-learning projects and another on a staged learning path for data and engineering staff, give plenty of task types to start logging.
For something outside delivery work altogether, there is a nonprofit organisation’s website at cotocus.org, which presents itself through program, research, community and blog pages and talks about using technology, innovation and education for social good. It is general background reading, not part of any estimating method.
One last story
An operations team once pinned its ratio table to the wall of a shared office. At first it felt like a confession. Within two months it became a conversation starter: someone would point at the “new vendor” row, then at their upcoming plan, and quietly double a number. No lecture, no policy, only a visible record. Myths fade quickly when the evidence sits where everybody walks past it.