IntermediateLesson 16 of 304 min read

Estimating Better

Use your own past jobs to correct your estimates, learn three-point estimating and a safe estimate on sample data, and turn estimates into honest quotes.

In this lesson you will

  • Compare estimates with real hours
  • Use a three-point estimate
  • Add a principled buffer
  • Improve with each job

About 24 minutes

Dev is always a bit wrong, in the same direction

Dev looks back over 8 finished jobs and notices something. Almost every job took longer than estimated. A single bad estimate is bad luck. Seven out of eight is a habit. Estimating is the freelancer’s most important and most underrated skill, because every price rests on it. The good news is that you can improve it with data you already have: your own time log from lesson 7.

Use your history

The first block of the program compares estimated hours with real hours for 8 made-up jobs. It divides actual by estimated to get a ratio for each job. A ratio of 1.25 means the job took a quarter longer. The average is 1.25, the middle value is 1.25, and 7 of the 8 jobs ran over. So Dev’s honest estimates are about a quarter too low. One simple fix follows: multiply your gut estimate by your historical ratio. A gut feeling of 8 hours becomes 10.0.

Three-point estimating

A single number hides uncertainty. A better habit is to write three: an optimistic estimate if all goes well, a most likely one and a pessimistic one if problems appear. A well-known formula called PERT combines them: add the optimistic, four times the most likely and the pessimistic, then divide by six. It gives the most likely case the most weight. A rough measure of spread is the pessimistic minus the optimistic, divided by six.

import statistics
from statistics import NormalDist

# MADE-UP past jobs: (estimated hours, actual hours)
history = [(8, 10), (6, 7), (10, 14), (5, 5), (12, 15), (4, 6), (9, 10), (7, 9)]
ratios = [a / e for e, a in history]
print("jobs:", len(history))
print("average actual / estimate:", round(statistics.mean(ratios), 2))
print("middle ratio:", round(statistics.median(ratios), 2))
print("jobs that ran over:", sum(1 for e, a in history if a > e), "of", len(history))

o, m, p = 6, 8, 13
expected = (o + 4 * m + p) / 6
spread = (p - o) / 6
print("three-point estimate:", round(expected, 2), "hours")
print("spread:", round(spread, 2))
safe = expected + NormalDist().inv_cdf(0.80) * spread
print("80% safe estimate:", round(safe, 2), "hours")
print("history-adjusted guess:", round(m * statistics.median(ratios), 1), "hours")
hours = round(safe * 2) / 2
print("quote hours:", hours, "| fee at floor:", round(hours * 24.55))
jobs: 8
average actual / estimate: 1.25
middle ratio: 1.25
jobs that ran over: 7 of 8
three-point estimate: 8.5 hours
spread: 1.17
80% safe estimate: 9.48 hours
history-adjusted guess: 10.0 hours
quote hours: 9.5 | fee at floor: 233

For the next job Dev guesses 6, 8 and 13 hours. The three-point estimate is 8.5 hours, a bit above the most likely 8 because the pessimistic case is further away than the optimistic one. The spread is 1.17 hours. If Dev wants an estimate that is not exceeded about four times in five, the program adds a margin of 0.84 spreads and gets 9.48 hours. That is a model, and real work is messier, but it gives a principled reason for a buffer instead of a vague worry. Rounding to the nearest half hour gives 9.5 hours, and at the floor of 24.55 an hour the fee is 233. Compare that with the history-adjusted guess of 10.0 hours. Two different methods land within an hour of each other, which is reassuring.

Make estimates better each time

  1. Record the estimate before you start, so you cannot fool yourself later.
  2. Log your real hours, as in lesson 7, including the awkward ones.
  3. When the job ends, compute the ratio and note what caused any gap.
  4. Update your rule. If your ratio is 1.25, plan on it until it changes.
  5. Break big jobs into small tasks and estimate each one. Small tasks are easier to judge.

Where estimates go wrong

  • Forgetting the extras. Emails, questions and rounds of changes.
  • Hope. The optimistic case feels like the real plan.
  • Unclear scope. If the job is not defined, the estimate cannot be right.
  • New kinds of work. The first time you do a task, add extra for learning.
  • Pressure. A client’s budget should not change your honest hours.
Common mistake: estimating to fit the budget

Estimating to fit the client’s budget. If the honest estimate is 10 hours and the client has room for 7, the work will still take 10. Change the scope instead of the estimate.

Sample data and no promises

The jobs, hours and ratios are made up, and the three-point formula is a standard estimating tool and not a guarantee. Real jobs surprise everyone. Fees use sample money, and no income is promised.

Find your last five jobs. Compute actual divided by estimated for each one, and write the average. Use it to correct your next estimate.

Practice for this lesson

Check what you learned

Five quick questions. Pick an answer and you will see straight away whether it is right and why.

What does a ratio of 1.25 mean?

  1. It took a quarter less
  2. It took twice as long
  3. It took exactly as estimated
  4. The job took a quarter longer than estimated

Actual divided by estimate of 1.25 is 25 per cent over.

What is the PERT three-point formula?

  1. Optimistic plus four times most likely plus pessimistic, over six
  2. The sum of three, over three
  3. The largest of three
  4. The smallest of three

It weights the most likely case four times.

Why is a single-number estimate risky?

  1. It is illegal
  2. It is too precise always
  3. It hides uncertainty
  4. It is free

Three points show a range.

What should you record before starting a job?

  1. The invoice
  2. The profit
  3. The tax
  4. Your estimate

Recording first stops you fooling yourself later.

What should you change when the client’s budget is too small?

  1. The honest estimate
  2. The scope
  3. Your floor
  4. The hours log

Cut scope, not truth.

Summary

  • Your history shows how wrong your estimates tend to be.
  • A three-point estimate shows a range, not one number.
  • A buffer can be calculated, not just feared.
  • Record estimates first and learn from every job.

Next lesson

Lesson 17 covers cash flow and putting money aside.

Questions

Is the PERT formula exact?

No. It is a standard estimating tool that gives a weighted average. Real jobs vary, so use it as a guide and check it against your own history.

What if I have no history yet?

Start logging now. Until you have a few jobs, add a generous buffer, such as a third more than your gut feeling, and learn from the first results.

Are these numbers real?

No. They are made-up sample numbers to show the method.

Progress is saved in this browser.