- Break the Work Into Tasks You Can Picture
- Use a Three-Point Estimate for Uncertain Tasks
To estimate project hours accurately, list the work as small tasks and estimate each one (bottom-up), give the uncertain tasks a three-point estimate, then check the total against how long similar past projects actually took. Don't skip the last step: an estimate built only from how this project should go comes out low. Then track the actual hours, so your next estimate starts from better data.
Why estimates come out low has been studied. In a paper published by the Project Management Institute, Bent Flyvbjerg summarizes Kahneman and Tversky's finding that "people will underestimate the costs, completion times, and risks of planned actions", which Lovallo and Kahneman call the "planning fallacy" (Flyvbjerg, From Nobel Prize to project management, PMI, 2006, opened October 29, 2026). The fix the paper describes is to compare your plan with the outcomes of similar finished projects, which is step 3 below.
Break the Work Into Tasks You Can Picture
Split the project until you can picture each task from start to finish, such as a day's work or less. A task like "build the website" hides the parts that take time. A task like "wireframes for 5 pages" can be estimated.
Then add the work that's easy to forget, because it isn't a deliverable:
- client reviews and the revision rounds after them;
- testing and fixing before launch;
- entering content the client supplies;
- launch and the fixes just after it;
- calls, emails and project admin.
Add up the hours for every task. That's your bottom-up estimate.
Use a Three-Point Estimate for Uncertain Tasks
For a task you're unsure about, estimate it three times:
- O, optimistic: the hours if nothing goes wrong.
- M, expected: the hours you think it will take.
- P, pessimistic: the hours if the problems you can foresee do happen.
Then combine them. The techniques wiki on ProjectManagement.com, a PMI site, gives two formulas (3-Points Estimating, a community-edited page last edited July 15, 2024, opened October 29, 2026):
- PERT (beta): E = (O + 4M + P) ÷ 6, which gives M four times the weight of O or P.
- Triangular: E = (O + M + P) ÷ 3, a plain average.
- The same page gives (P − O) ÷ 6 as the standard deviation of the PERT estimate. A larger value means a less certain task.
For example, a design task with O = 12, M = 18 and P = 30 hours has a PERT estimate of (12 + 72 + 30) ÷ 6 = 19 hours, and a standard deviation of (30 − 12) ÷ 6 = 3 hours. Replace the task's single estimate with the PERT figure in your bottom-up total.
Check the Total Against Your Past Projects
Flyvbjerg's paper describes reference class forecasting in three steps: identify "a relevant reference class of past, similar projects", establish the distribution of their outcomes, and compare the new project with that distribution.
For a small business, the reference class is your own finished projects of the same kind. For each one, divide the hours it took by the hours you estimated. If your past estimates were 25% low, a new estimate built the same way is likely to be low by a similar amount, unless something about this project is different and you can say what.
The paper notes that the class must be "broad enough to be statistically meaningful, but narrow enough to be truly comparable". Five-page websites compare with five-page websites, not with online stores.
Decide on Contingency From Your Own Track Record
Contingency is the hours you add for problems you can't place in a specific task yet. There's no universal percentage that fits every business or project. Base it on your own record:
- If you have comparable past projects, the gap between your estimates and the actual hours is your evidence. Plan for the total the reference class check gives you.
- If you don't, choose a contingency you can explain, write it down as a separate line, and check it against the actual hours when the project ends.
Keeping contingency as its own line, not spread across tasks, shows you afterward whether you needed it.
Worked Example: Hours for a 5-Page Website
A designer and developer estimates a 5-page marketing website. All the figures are a sample.
1. Bottom-up, in hours
- Discovery and brief: 3
- Wireframes, 5 pages: 6
- Design, 5 pages, desktop and mobile: 18
- Client review, round 1: 4
- Design revisions, round 2: 6
- Front-end build, 5 pages: 22
- Content entry: 4
- Analytics setup: 2
- Testing across browsers and devices: 6
- Launch and fixes: 4
- Total: 75 hours
2. Three-point estimates for the two uncertain tasks
- Design: O = 12, M = 18, P = 30, so (12 + 72 + 30) ÷ 6 = 19 hours, 1 more than the bottom-up figure.
- Front-end build: O = 18, M = 22, P = 32, so (18 + 88 + 32) ÷ 6 = 23 hours, 1 more.
- Revised total: 75 + 1 + 1 = 77 hours
3. Check against past projects
The designer has tracked 8 similar websites: estimates totaling 520 hours, against 650 hours actually worked. 650 ÷ 520 = 1.25, so their estimates ran 25% low. Applied here: 77 × 1.25 = 96.25 hours.
4. Plan and price
The designer plans for about 96 hours. At a sample rate of $85.00 an hour, 96.25 hours is $8,181.25. The bottom-up 75 hours alone would have priced the job at $6,375.00, which is $1,806.25 less: the amount that past projects say the estimate would have missed.
Track Actual Hours Against Every Estimate
The check in step 3 only works if you record what happened. At the end of each project, record:
- the project type, using the same label each time, such as "5-page website";
- the estimated hours, by task and in total;
- the actual hours, by task and in total;
- the ratio of actual to estimated hours;
- the task with the biggest miss, and one sentence on why.
Track time on fixed-price projects too. The client doesn't see the hours, but you need them for the next estimate.
Billed publishes this guide. Its timesheets page says you can "Log time for each task as you work, and see how much time goes into each project," that "Billed breaks your time down by day and by week" and that "Past entries also give you a record to estimate the next project from." It adds that "Billed's reports include a time tracking report," which "can be filtered by date range and exported." For hourly work: "Create an hourly rate project, set its hourly rate, and Billed tracks the charge for each task as you log it. From the project dashboard you can invoice one task or several at once." See timesheets and time tracking.
When a Project Runs Over Its Estimate
Tell the client as soon as you see the overrun coming, not when the final invoice goes out. Then work out the cause:
- The scope changed. The client asked for more, or something the estimate excluded turned out to be needed. Agree on the extra hours and cost in writing before you do the work.
- The estimate was wrong for the same scope. Decide whether to absorb the hours or raise it with the client, and record the miss so the next estimate allows for it.
Related guides
- How to create a project estimate
- How to write a winning estimate
- How to handle estimate revisions
- Estimate vs invoice best practices: from an accepted estimate, through change orders, to the final invoice
- How to follow up on estimates
- Construction estimating basics: includes AACE estimate classes and a worked example
- Guide to hourly invoicing
- Guide to milestone invoicing
- Guide to progress invoicing
- Freelance hourly rate calculator: works out the hourly rate that covers the pay you want, your business costs and a tax set-aside
FAQ
What's the formula for a three-point estimate?
The PERT (beta) version is (O + 4M + P) ÷ 6, where O is the optimistic hours, M the expected hours and P the pessimistic hours. The triangular version is (O + M + P) ÷ 3. For O = 12, M = 18 and P = 30, PERT gives 19 hours and the triangular formula gives 20.
How much contingency should I add?
There's no universal percentage. If you have tracked similar projects, divide their actual hours by their estimated hours and apply that ratio to the new estimate. If you haven't, choose a contingency you can explain, keep it as a separate line, and compare it with the actual hours at the end.
What if I don't have past projects to compare with?
Start with a bottom-up estimate and three-point estimates for the uncertain tasks, add a contingency you can explain, and track the actual hours from your first project onward. Each finished project of the same kind adds to the reference class you can check against.
Why do my estimates come out low?
Research on forecasting finds that people underestimate how long planned work will take, which Lovallo and Kahneman call the planning fallacy. Estimates built from how the project should go miss the delays, revisions and rework that similar past projects had. Comparing with your own past actual hours corrects for this.
Should I give the client a range or a single number?
A range suits an early ballpark, before the scope is detailed: say what moves the price toward the low or high end. Once the scope is set, a three-point estimate gives you both ends of the range and a planning figure in between.
Should I track time on fixed-price projects?
Yes, if you want your next estimate to be more accurate. Fixed-price clients don't see the hours, but without them you can't tell how far your estimate was off, or which tasks caused the difference.
