Estimating is the art of predicting the future while knowing you're probably wrong. The goal isn't perfect accuracy - it's narrowing the cone of uncertainty enough that stakeholders can make good decisions. If you ever meet someone who claims their estimates are always right, nod politely and don't let them near your project.
Three techniques that actually work
Analogous estimating: look at similar work from the past and use that as your baseline. 'We built a customer portal last year in 12 weeks. This one is 30% more complex, so let's estimate 16 weeks.' It's fast, it's simple, and it's only as good as your historical data - which, to be honest, is usually a spreadsheet someone forgot to update.
Parametric estimating: use a statistical relationship between variables. 'Each user story takes our team an average of 1.3 days. We have 40 stories. That's 52 days, assuming the coffee machine doesn't break and nobody goes on leave.' This is more precise than analogous, but garbage in, garbage out - your '1.3 days' figure is only useful if it was actually measured, not guessed.
Three-point estimating: the PMP exam's gift to the world. For each task, estimate the best case (optimistic), worst case (pessimistic), and most likely case. Then apply the PERT formula: (Optimistic + 4×Most Likely + Pessimistic) / 6. The result is a weighted average that accounts for uncertainty. The real benefit isn't the math - it's that the conversation forces you to think about what could go wrong before it does.
What not to do (field notes from the hall of shame)
Don't pad estimates secretly. Adding 50% 'just in case' without telling anyone destroys trust and makes everyone else's planning unreliable. If you need contingency, be transparent about it. 'I estimate 8 weeks, and I'd recommend adding 2 weeks of contingency for the integration risk we identified.' That's honesty. 'The estimate is 10 weeks' when you know it's 8 is a future fight waiting to happen.
Don't let the first number spoken become the anchor. In meetings, whoever throws out the first estimate tends to set the range everyone argues around. If a stakeholder says 'this should take two weeks' and you think it's six, don't start negotiating down from two - start from the data. 'Based on similar work, our team averages five to seven weeks for this scope. Let me walk you through why.'
Don't estimate in isolation and then spring it on the team. The people doing the work have the best information about how long it takes. Ask them. Their estimates will be more accurate than yours, and they'll be more committed to a timeline they helped create. This is not abdication - it's delegation of the part you're worst at.
Summary & next steps
Analogous, parametric, and three-point estimating each have their place. The common thread: good estimates come from data, historical records, and the people doing the work - not from optimism or pressure. Be transparent about uncertainty, and never let the first number spoken become the only number considered.
Next: budget management - turning estimates into something the finance department won't immediately return marked 'insufficient rigour.'