A study of some 16,000 major projects by Bent Flyvbjerg and Dan Gardner found that only 8.5% were delivered on time and on budget, and just 0.5% also produced the benefits they promised (SSRN, 2023). Many of those overruns start with project cost estimation itself, on the day someone writes the first number down, because an estimate made when almost nothing is defined behaves very differently from one made after the work is broken into tasks.
The cost estimation standard most engineering firms follow puts this in numbers. AACE International’s classification system says a concept stage estimate can land 20% to 50% below the actual cost or 30% to 100% above it, while a fully detailed one narrows to roughly minus 3% to 10% and plus 3% to 15%. The range is lopsided on purpose: early estimates miss high far more than they miss low.
This guide explains which project cost estimation technique fits each stage of definition and how accurate each one can honestly be, then shows how to turn the result into a number you can defend. It compares the main cost estimation techniques and walks through a worked example that estimates the same project three ways. It also covers a contingency calculation and the failure modes that keep estimates wrong from one project to the next.
- What cost estimation in project management produces
- How accurate project cost estimation can be at each stage
- 8 cost estimation techniques for projects
- Which cost estimation technique to use and when
- Worked example: one project estimated three ways
- How to set contingency without padding
- Where cost estimates go wrong
- Feeding estimates with actual hours
- FAQ
What cost estimation in project management produces
Cost estimation in project management is the work of predicting what a defined scope will cost before the money is spent. The output is not a single figure. A usable result of cost estimation in project management has three parts: a point value, a range that says how far the actual cost could reasonably move from it, and a basis of estimate that records the scope, assumptions, rates and data sources behind the number.
Keep four related numbers apart, because teams often use the words interchangeably and then argue about the wrong thing:
- Estimate: what the work is likely to cost, including uncertainty.
- Budget: the amount approved for spending, usually the estimate plus contingency, sometimes cut by a sponsor.
- Quote or price: what the client pays, which adds margin and commercial judgment on top of cost.
- Actual cost: what the recorded hours, purchases and overheads add up to once work happens.
What goes into the estimate (direct and indirect, fixed and variable costs) is covered in our guide to project costs and how to calculate them, and the way to lay those costs out line by line is in the project cost breakdown guide. This article focuses on the estimating itself: which method to apply, and how much to trust what it gives you.
How accurate project cost estimation can be at each stage
AACE International Recommended Practice 18R-97 sorts estimates into five classes by how much of the project is defined when the estimate is made. It was written for engineering and construction in the process industries, but the logic carries over to cost estimation in project management for software, consulting and internal projects: accuracy is limited by definition, not by the effort you spend on arithmetic.
| Class | Project defined | Typical purpose | Typical method | Expected accuracy (low / high) |
|---|---|---|---|---|
| Class 5 | 0% to 2% | Concept screening | Parametric models, judgment or analogy | -20% to -50% / +30% to +100% |
| Class 4 | 1% to 15% | Study or feasibility | Factored or parametric models | -15% to -30% / +20% to +50% |
| Class 3 | 10% to 40% | Budget authorization or control | Semi detailed unit costs | -10% to -20% / +10% to +30% |
| Class 2 | 30% to 75% | Control or bid | Detailed unit costs | -5% to -15% / +5% to +20% |
| Class 1 | 65% to 100% | Check estimate or bid | Detailed unit costs with full takeoff | -3% to -10% / +3% to +15% |
Source: AACE International RP 18R-97, Table 1, process industries. The same ranges appear in the US Department of Energy’s cost estimating guide, DOE G 413.3-21A.
Three details in the standard matter more than the table, and most summaries leave them out:
- The ranges apply after contingency. AACE describes them as the typical variation of actual cost from the estimate once contingency has been added to reach a 50% chance of overrun versus underrun, and says about 80% of projects should fall inside them when contingency is handled well. A bare estimate with no contingency should be expected to do worse.
- The high side is wider than the low side. A Class 5 estimate can be 50% too low or 100% too high. Quoting it as “plus or minus 50%” understates the risk on exactly the side that hurts.
- The class does not guarantee the accuracy. The standard notes that a Class 5 estimate for a repeat project with good cost history can be as accurate as a Class 3 estimate for a project built on new technology. Historical data moves you up faster than extra hours of estimating.
Project management guides that follow PMI terms use simpler labels: a rough order of magnitude estimate at initiation, a budget estimate during planning, and a definitive estimate once the work breakdown is complete. They map loosely onto Classes 5, 3 and 1.
8 cost estimation techniques for projects
Each of these cost estimation techniques answers a different question, and each needs a different kind of input. The order below runs roughly from what works with almost no definition to what needs the most. If you need to estimate duration and effort rather than money, our separate guide to project estimation techniques covers that side.
1. Analogous estimating
You take the actual cost of a similar finished project and adjust it for known differences in size, complexity, location or rates. If last year’s client portal cost $84,000 and the new one has about 20% more scope, the analogous estimate is $100,800.
Use it at concept stage, when you have a comparable project and little else. Watch for the comparison project’s recorded cost being incomplete. If staff did not log all their hours on it, the analogy inherits that gap and every estimate based on it starts low.
2. Parametric estimating
You find a cost driver that scales with the work, measure its historical unit cost, and multiply: cost per square foot, per page template, per integration, per training seat. The formula is simply unit cost x quantity + fixed costs.
A related rule from engineering, often asked about as the “0.6 rule”, scales cost by capacity rather than linearly, because doubling the size of a plant rarely doubles its cost. For service projects a linear driver is usually enough, provided it was measured on several projects rather than one.
Use it when the work is repeatable and you have a reliable unit rate. Watch for drivers that look causal but are not; pages of documentation can track effort on one type of project and mean nothing on another.
3. Bottom-up estimating
You break the scope into work packages using a work breakdown structure, estimate hours for each package by role, multiply by cost rates, and add materials, licenses and subcontracts. It is the slowest method and the most accurate one, because every line can be checked by the person who will do the work.
Use it for budget approval and fixed price bids. Watch for the sum of many small optimistic numbers. Each task estimate tends to describe the case where nothing goes wrong, so a bottom-up total needs a risk adjustment rather than blind trust in its precision.
4. Three-point estimating (PERT)
For each uncertain item you record an optimistic (O), most likely (M) and pessimistic (P) value. The beta, or PERT, formula weights the most likely case: E = (O + 4M + P) / 6, with a standard deviation of (P – O) / 6. A simple triangular average, (O + M + P) / 3, gives more weight to the extremes.
Use it on the few line items where the risk sits, not on every task. Watch for pessimistic values that are only slightly worse than the most likely case. A pessimistic number should describe a bad but plausible outcome, which on integration or regulatory work is often double the most likely figure.
5. Top-down estimating
You start with a total, often a sponsor’s budget or an analogous figure, and allocate it down to phases by historical percentages. If discovery usually takes 5% of cost and testing 12%, a $200,000 envelope gives $10,000 and $24,000.
Use it to test whether a fixed budget is realistic and to set phase targets. Watch for treating the allocation as an estimate. It tells you what each phase may spend, not what it will cost.
6. Expert judgment and the Delphi method
Experienced people estimate from what they have seen. The Delphi version collects anonymous estimates from several experts, shares the spread and reasoning, and repeats until the range narrows, which keeps the most senior or loudest voice from anchoring the group.
Use it for new kinds of work without historical data. Watch for experts remembering how long work felt rather than what it cost. Ask them to name the comparison project they are drawing on.
7. Vendor bid analysis
For outsourced parts of the scope you collect quotes from several suppliers and compare them with each other and with your own should cost figure. A bid far below the rest usually signals a misunderstanding of scope, not a bargain.
Use it wherever subcontracts or purchased services are a large share of cost. Watch for exclusions buried in the quotes; compare scope line by line before comparing totals.
8. Reserve analysis
Reserve analysis sizes the money set aside for risk. Contingency reserve covers identified risks and sits inside the cost baseline the project manager controls. Management reserve covers risks nobody has identified and sits outside it, released only by the sponsor. It is not a way to estimate the work itself, but an estimate without it is incomplete, which is why it gets its own section below.
Which cost estimation technique to use and when
No single method covers a whole project. Good cost estimation in project management uses a sequence that starts coarse and gets finer as definition improves, with a second method checking the first at every stage.
| Stage | What you know | Primary technique | Cross check | Roughly equals |
|---|---|---|---|---|
| Idea or sales inquiry | Goal, rough size, one similar project | Analogous | Expert judgment | Class 5 |
| Business case or proposal | Main deliverables and a measurable size | Parametric | Top-down allocation | Class 4 |
| Budget approval or fixed price bid | Work breakdown with roles and rates | Bottom-up | Parametric, vendor bids | Class 3 |
| Risky line items | Named risks and their range | Three-point | Reserve analysis | Any class |
| During delivery | Actual hours and spend to date | Reforecast from actuals | Remaining bottom-up | Class 2 to 1 |
Combining cost estimation techniques is the part that adds accuracy. When two independent methods land close together you have some evidence the estimate is sound; when they disagree by more than the class range allows, the gap points at a scope assumption one of them got wrong.
Worked example: one project estimated three ways
Here is project cost estimation applied to a 12 week client portal build by a small agency. The figures below are illustrative, and the cost rates are internal hourly costs per role, not billing rates.
Step 1: analogous and parametric figures
Step 2: bottom-up estimate
| Work package | Hours | Cost rate | Cost |
|---|---|---|---|
| Discovery | 60 | $70 | $4,200 |
| Design | 220 | $65 | $14,300 |
| Front end development | 420 | $75 | $31,500 |
| Back end and integrations | 300 | $85 | $25,500 |
| Quality assurance | 140 | $55 | $7,700 |
| Project management | 150 | $80 | $12,000 |
| Labor total | 1,290 | — | $95,200 |
| Licenses and user testing | — | — | $5,400 |
| Bottom-up total | — | — | $100,600 |
Step 3: three-point adjustment on the risky item
The integrations package depends on a client API nobody has seen yet, so it gets a three-point estimate instead of a single figure.
What the three numbers tell you. The three cost estimation techniques land between $97,400 and $102,583, a spread of about 5%. That is reassuring, but it is not proof, since all three lean on the same agency’s history. If the parametric figure had come in at $70,000, the gap would have been the most useful output of the exercise, because it would mean either the template count or the rate per template does not describe this project.
Sensitivity. If front end work runs 15% over, that is 63 extra hours or $4,725, which the $10,258 contingency absorbs. If the integration hits its pessimistic case, the extra 220 hours cost $18,700 and exceed the contingency on their own. That single line deserves a scope clarification with the client before the price is fixed, not a bigger buffer.
How to set contingency without padding
Contingency is where project cost estimation turns into a budget. Padding hides risk inside task estimates, where nobody can see it, manage it or give it back. Contingency names the risk and puts a separate number on it. Three practical ways to size it, from simplest to strongest:
-
Percentage by class.
Apply a percentage that shrinks as definition improves. It is quick and defensible for small projects, but the percentage should come from your own past overruns, not from a figure copied from another industry.
-
Expected value of named risks.
List the risks, estimate the cost impact and the probability of each, and add up probability x impact. A risk with a 30% chance of adding $18,700 contributes $5,610.
-
Range based.
Use the three-point results to model the total, then set contingency so that the budget has a stated chance of being enough. AACE’s ranges assume contingency sized for an even chance of overrun or underrun; a sponsor who wants more certainty needs a larger reserve.
Whichever method you use, record what the contingency is for. When a named risk passes without happening, its share can be released, which is the opposite of what happens to padding.
Where cost estimates go wrong
The same handful of failures shows up in project cost estimation across industries, and most of them have nothing to do with which of the cost estimation techniques you chose.
- Estimating effort and calling it cost. Hours x billing rate is a price. Cost needs each person’s own cost rate, plus overtime premiums, leave, tools and overhead that do not appear on any task.
- History that was never recorded. Analogous and parametric estimates are only as good as the actual cost data behind them. Teams that track time loosely, or only on billable tasks, build every future estimate on numbers that are too small.
- Anchoring on the client’s budget. Once someone has heard “we have $80,000”, estimates drift toward it. Estimate first, then compare with the budget.
- Scope that was assumed, not written. Most overruns trace back to work nobody listed. Write exclusions and assumptions into the basis of estimate, and treat later additions as changes, not as scope creep the estimate should have absorbed.
- One number, no range. A single figure turns project cost estimation into a promise. Present the point estimate with its class and range.
- Never reforecasting. The estimate made at approval is the least informed one the project will ever have. Compare actuals with it at every milestone and update the estimate at completion.
Software projects add their own failure modes, such as unclear requirements and integration risk; they are covered in detail in our guide to estimating software development cost.
Feeding estimates with actual hours
Project cost estimation gets easier with every finished project, provided the actuals were recorded. Labor is the largest cost on most service and internal projects, so the quality of your project cost estimation depends on how well you know what the work took last time. That is a time tracking problem before it is an estimating one.
In actiTIME, the loop works like this:
- Task estimates. Set estimated hours on each task when you scope the work, so the bottom-up estimate lives in the same place as the time that will be tracked against it.
- Estimated vs. Actual Time Report. Shows the variance per task, project or customer, which tells you which kinds of work you consistently underestimate.
- Cost of work rates. Each user has regular, overtime and leave cost rates, so tracked hours turn into labor cost automatically in the Cost of Work Report.
- Budgets. Time, cost and billing budgets can be set on a customer, project or task, with a progress bar that turns red once a budget is exceeded.
- Profit/Loss Report. Compares billable amounts with costs by customer, project or task, which shows whether the estimate behind a fixed price held up.
After a few projects, those reports become the historical data that analogous and parametric cost estimation techniques need: real cost per deliverable, real hours per role, and a measured record of how far your estimates usually miss.
Now we can better predict future project requirements
Our company needed a simple way of tracking time used on multiple projects – and actiTIME fit the need. Its interface is simple and easy to maintain. We use the application for time management, task estimation and also to communicate deadline information to our team members. Now having actiTIME we can better predict future project requirements!
actiTIME costs from $5 per user per month, and there is a permanent free version for teams of up to 3 users. You can start a free trial to run your next estimate against real tracked hours.
FAQ
What is the 0.6 rule in cost estimating?
The six tenths rule scales the known cost of one facility or system to a different capacity: new cost = known cost x (new capacity / known capacity) raised to the power 0.6. It reflects economies of scale, so a unit twice as large costs about 1.5 times as much rather than twice as much. It is a concept stage tool for plants and equipment and rarely fits labor heavy service work, where cost tends to grow in line with scope.
How long should a cost estimate take to prepare?
Effort grows with the class. In the AACE matrix a Class 5 estimate relies on judgment, analogy or parametric models and can be done quickly, while Class 1 and 2 estimates need detailed unit costs from a full takeoff of the scope. A practical rule for project cost estimation is to spend only as much time as the decision needs: a go or no go screen needs a ranged analogous figure, a fixed price bid needs a bottom-up estimate.
Should a project cost estimate include profit margin?
No. Keep the cost estimate free of margin and add margin when you set the price. Mixing them hides how much of a quote is cost and how much is profit, so when a client negotiates the price down, nobody can see whether the cut came out of margin or out of the money needed to deliver the work.
