Test estimates are usually wrong in the same direction. They are too low, not because testers are careless, but because the estimate counts the work everyone can see (writing and running test cases) and misses the work that only appears once the project starts: waiting for builds, chasing environment issues, retesting fixes, and re-running regression after a late change. A good estimation method is mostly a way of making that invisible work visible before someone commits to a date.

Start by breaking the work down

The most reliable foundation is a work breakdown structure: split testing into tasks small enough that someone can picture doing them. A single number for “testing the new loan module” is a guess. Separate numbers for each of these are judgements:

  • Understanding requirements and raising clarification questions
  • Test design: scenarios, test cases, data requirements
  • Test data and environment preparation
  • First execution cycle, including defect logging
  • Retesting fixed defects and a second regression cycle
  • Automation of selected cases, if in scope
  • Reporting, meetings, coordination with developers and business users

The last three items are the ones most often left out. Retesting alone can be a third or more of execution effort on a project with a typical defect count, because every fix needs verification and a check that nothing nearby broke.

Three numbers instead of one

A single-point estimate hides uncertainty. Three-point estimation asks for an optimistic (O), a most likely (M) and a pessimistic (P) figure for each task. The weighted average used in the PERT technique is:

Expected = (O + 4M + P) ÷ 6

The most likely value gets four times the weight of the extremes, but a long pessimistic tail still pulls the expected value upwards. That is realistic: tasks rarely finish much earlier than planned, but they often finish much later. A rough measure of spread for each task is (P − O) ÷ 6.

Here is an estimate, in person-days, for testing a mid-sized feature:

TaskOMPExpected
Test design46106.33
Data and environment setup2383.67
Execution cycle 157127.50
Retest and regression cycle 235105.50
Reporting and coordination1232.00
Total15234325.00

Notice what happened. If you had simply added the most likely values, you would have committed to 23 days. The PERT expected total is 25. The difference comes almost entirely from environment setup and the execution cycles, where the pessimistic case is far from the likely one, which is exactly where experience says delays come from.

You can go one step further. Combining the task spreads (by adding their squares and taking the square root, which assumes the tasks vary independently) gives a total spread of about 2.2 days. Committing to the expected value means roughly even odds of overrunning. Adding about two spreads, for a figure near 29 to 30 person-days, gives a much safer commitment under the method’s simplifying assumptions. Whether the business wants the 25-day plan or the 30-day plan is then a conscious choice about risk, not an accident.

Other methods and when they fit

Ratio or percentage methods. Many organisations estimate testing as a share of development effort, for instance a ratio derived from past projects of a similar kind. This is fast and useful for early budgeting, but it breaks when the testing burden does not scale with code: a small change to a payment flow can require far more testing than its development effort suggests.

Historical data. If your test management tool records how long previous cycles took and how many test cases they had, you can derive productivity figures, such as test cases designed or executed per person-day for your team on similar work. This beats any textbook number. The caution is consistency: “a test case” means very different things in different teams.

Wideband Delphi and planning poker. Several people estimate independently, compare, discuss the outliers and re-estimate. The discussion is the point. When one tester says three days and another says twelve, one of them knows something about the work the other does not.

Size-based methods. Function point analysis and its testing derivative, test point analysis, estimate from the size and complexity of the functionality. They are more formal and appear in some large outsourcing contracts in India where vendors must justify effort, but they need calibration to be trustworthy.

From person-days to calendar days

Effort is not duration. Twenty-five person-days does not mean five testers finish in five days. Testing depends on builds arriving, defects being fixed, and environments staying up. If developers take two days on average to fix a defect, the second cycle cannot start until those fixes land, however many testers are waiting. Testers also rarely spend a full eight hours on project tasks; meetings, support for other projects and leave reduce real availability. Many planners assume around six productive hours a day, though the right figure depends on the team.

A defensible schedule therefore states its dependencies explicitly: “This plan assumes the build is delivered on the 10th, that critical defects are fixed within one working day, and that the UAT environment is shared with no other project.”

Assumptions are part of the estimate

An estimate without its assumptions is just a number someone will hold you to after the conditions change. Write down the scope included and excluded, the number of test cycles assumed, the expected defect volume, the environments and devices covered, and any third-party dependencies. When the client adds three more payment methods in week two, the documented assumptions show that the estimate needs revisiting, rather than the team silently absorbing extra work.

Estimates should also be re-estimated. After the first execution cycle, you know the real defect rate and the real stability of the build. Updating the forecast then is not an admission of failure; it is the estimate doing its job.

When the business says “do it in half the time”

The estimate’s real usefulness shows when someone asks for a two-week test effort in two days. Because the work is broken down, you can show exactly what disappears at each budget: perhaps only high-risk areas get tested, regression is limited to automated suites, and the second cycle is dropped. The business can then choose what to give up. Without the breakdown, the only possible responses are “yes” or “no”, and neither is helpful.

Written by the RP Strativa Team. All task figures and the productive-hours assumption are illustrative.

Leave a Reply

Trending