Estimating IT Projects: Analogous, Parametric, Three-Point and Bottom-Up Techniques

Ask any project manager what keeps them awake, and estimation will feature high on the list. Stakeholders want a firm date and cost; engineers know the work contains unknowns. The truth is that an estimate is a prediction with uncertainty, and good estimating is the craft of making that uncertainty visible and manageable. This lesson covers the main techniques, a worked example and the habits that improve accuracy.

Estimates, Targets and Commitments

Three things are often confused. An estimate is a prediction of effort, duration or cost based on what is known. A target is what the business wants (“launch by 1 March”). A commitment is a promise to deliver. Trouble begins when a target is simply relabelled as an estimate. If the estimate does not meet the target, the right response is to change scope, resources or date, not to pretend the estimate was different.

Effort, Duration and Cost

Effort is the amount of work, in person-hours or person-days. Duration is the elapsed calendar time, which depends on how many people work on it and how available they are. Cost follows from effort and rates plus other expenses. Ten person-days of effort is not ten calendar days: with two people at 50 per cent availability it is ten days; with one person at 50 per cent it is twenty.

Main Estimating Techniques

Analogous (top-down) estimating

Use the actual figures from a similar past project, adjusted for differences. It is fast and cheap, and suits early stages, but is only as good as the comparison and the quality of historical data.

Parametric estimating

Apply a statistical relationship between variables, for example “testing takes 0.5 days per use case” or “200 screens at 3 days each”. It is more accurate than analogy when the unit rates are well founded.

Bottom-up estimating

Estimate each work package or task, then add them up. It is the most accurate when the WBS is detailed, but takes the most time and relies on the people who will do the work.

Three-point (PERT) estimating

Ask for three numbers: optimistic (O), most likely (M) and pessimistic (P). Then calculate the expected value and spread:

  • PERT expected estimate = (O + 4M + P) ÷ 6
  • Standard deviation = (P − O) ÷ 6
  • A simple triangular average = (O + M + P) ÷ 3

Expert judgement and Delphi

Consult people with relevant experience. The Delphi technique collects anonymous estimates over several rounds so that dominant voices do not skew the result.

Planning poker and relative sizing

Agile teams estimate relative size using story points rather than hours; see Planning Poker in Scrum and Estimation and Story Points in Scrum.

Worked PERT Example

A developer is asked how many days it will take to build an integration with a payment gateway. Optimistic: 8 days. Most likely: 12 days. Pessimistic: 22 days (the vendor’s sandbox is unreliable).

  • PERT estimate = (8 + 4×12 + 22) ÷ 6 = 78 ÷ 6 = 13 days
  • Standard deviation = (22 − 8) ÷ 6 ≈ 2.3 days
  • Triangular average = (8 + 12 + 22) ÷ 3 = 14 days

Roughly two-thirds of the time, the outcome should fall within one standard deviation of the expected value, between about 10.7 and 15.3 days. For a more cautious commitment, quote the expected value plus two standard deviations, about 17.7 days, which covers roughly 95 per cent of outcomes under the technique’s assumptions. To combine several tasks on a path, add the expected values and add the variances (the squares of standard deviations), then take the square root.

Comparing the Techniques

TechniqueAccuracyEffort to produceBest used
AnalogousLow to mediumLowEarly, with little detail
ParametricMedium to highMediumWhen reliable unit rates exist
Bottom-upHighHighDetailed planning
Three-pointMedium to highMediumUncertain tasks and risk analysis
Relative sizingMediumLowAgile teams with stable velocity

The Cone of Uncertainty

Early in a project, estimates can be off by a large factor in either direction. As scope is clarified and work is completed, the range narrows. Present early estimates as ranges, such as “between ₹70 and ₹110 lakh”, and promise to refine them at defined points. A single precise number given at the start of the project implies certainty that cannot exist.

Habits That Improve Estimates

  1. Let the doers estimate. The person who will do the work usually has the best insight, though they may be optimistic.
  2. Estimate in ranges and state assumptions. Write down what is included and excluded.
  3. Include all work: testing, reviews, meetings, rework, environments, documentation and project management.
  4. Use historical data. Record actuals against estimates and learn from the gap.
  5. Estimate independently, then discuss differences. Disagreement reveals hidden assumptions.
  6. Account for availability. People are rarely productive for a full day on project work.
  7. Re-estimate regularly, as knowledge grows.

Padding, Parkinson and Student Syndrome

People protect themselves by padding estimates. Then Parkinson’s Law says work expands to fill the time available, and the student syndrome says people start at the last moment. The padding is consumed, and the project is late anyway. Avoid hidden padding: estimate honestly and hold explicit contingency at project level. See Project Budgeting and Cost Control.

Common Mistakes

  • Giving a single-point estimate under pressure, without ranges or assumptions.
  • Estimating only development, forgetting testing, deployment and management.
  • Letting someone else estimate work they will not do.
  • Never comparing estimates with actuals.
  • Accepting a target as if it were an estimate.
Key takeaways
  • An estimate is a prediction with uncertainty; a target is a wish and a commitment is a promise.
  • Choose techniques by stage: analogous early, bottom-up and three-point for detail.
  • PERT: (O + 4M + P) ÷ 6, with standard deviation (P − O) ÷ 6.
  • Give ranges, record assumptions and compare estimates with actuals.

Related Lessons