Estimation is used to try to predict future outcomes related the the iron triangle elements of time, money, and, to a lesser degree quality (or features or performance). The BABOK essentially only discusses the first two. Estimates can be made of both costs and benefits. While all aspects of this process are in a sense entrepreneurial, the biggest component of entrepreneurial judgment is predicting future benefits, particularly for potential sales.
Any aspect of an effort or solution may be estimated for any part of its full life cycle. Examples include the time, cost, and effort (in terms of staff and materials) of any activity; capital, project, and fixed and variable costs of delivered solutions, potential benefits (e.g., sales, savings, reduced losses), and net performance (projected benefits minus projected costs).
The most important thing to know about estimation is that it tends to be more accurate when more information is available. This is especially true when making estimates about outcomes very similar situations from the past.
There are many methods of estimation including:
- Top-down and Bottom-up: Estimates can be performed from both ends depending on what is known about the engagement and the solution (the project and the product). Breakdowns can be made from the highest levels down to more detailed levels, or aggregations can be made from detailed low-level information which is then grouped and summed.
- Parametric Estimation: This method has a lot in common with bottom-up estimation. It attempts to multiply lesser-known input information (how many of A, B, and C) by better-known parametric information (e.g., the known prices for each individual example of A, B, and C). Levels of skill and experience can figure in to such calculations as well.
- Rough Order of Magnitude (ROM): This is basically an educated guess, based on experience, impressions, and entrepreneurial judgment. There are a few pithier names for this method!
- Rolling Wave: This involves making continuous estimates of elements throughout an engagement, which ideally become more accurate over time as more is known and less is unknown.
- Delphi: This technique seeks estimates from a wide variety of participants, potentially over multiple iterations, until a consensus is reached. This allows certain kinds of knowledge to be shared across the participants. As an example, think of a group of coders bidding on tasks during sprint planning. Most participants might make similar judgments of the complexity of a task, but if one or two team members make very different estimates they could share that they’re aware of a simple or existing solution to the problem that will reduce the effort required, or know about hidden requirements and other stumbling blocks that will increase the effort required. As another example, the first issue of Omni Magazine included a Delphic poll of its readership asking about when certain developments, discoveries, and accomplishments might take place. The results were published in a subsequent issue.
- PERT: This technique asks participants to estimate best-case, expected, and worst-case outcomes, which are then averaged, with the expected outcome given a weighting of four times, i.e., result = (best + 4*expected + worst) / 6.
As mentioned above, the accuracy of estimates is likely to improve when more information is available. This information can come from similar or analogous situations, historical information, expert judgment, or a combination of any or all of these.
Estimates can be given as point values or as a range, the latter of which will also indicate the degree of uncertainty. A measure called the confidence interval describes the expected range of outcomes, and it is generally expressed as (1 – expected maximum error), where the expected maximum error is a percentage of the central value. For example, an estimate of 100 plus or minus 10 would indicate a confidence interval of 90%. In the case of 100 plus or minus five, the confidence would be 95%. Certain statistical and Monte Carlo techniques generate confidence intervals. In these two examples, the maximum absolute error in one direction is sometimes called the half-width, because it is half of the full range of possible outcomes (the upper and lower bounds do not have to be the same distance from the expected value.). This information can come into play when determining needed sample sizes.
Estimates should generally be made by those responsible for the outcome of the effort for which the estimate was performed. These can, however, be checked against estimates from additional parties.