The science of estimation is an art rooted in risk management

I frequently find people mixing up estimation techniques between plan-driven and change-driven approaches thinking they are all completely different. While there are small changes, the art of estimation is simply looking at the level of accuracy when estimated and the extent of confidence in the given estimate for working on it! From that angle, the approaches like the analogous estimation, parametric estimation, triangular estimation, the special PERT estimation, affinity estimation, relative sizing, and storypointing can all be categoized into top-down, budget, and bottom-up estimation. 

Top Down Estimation

• The top-down estimate is often based on gut feel. It draws on the experience of previous projects (hence analogous). It could be driven off of someone's expertise (expert judgment) or based on knowledge tracked in historical records (corporate knowledge base). It is frequently done at the early stages of a project (initiation) where minimal effort is required to get a feel for whether an initiative should be undertaken or not! As a result, the level of accuracy is very low (-25% to +75%). This is why this technique is called rough order of magnitude (ROM) or order of magnitude (OOM). 
• In adapative projects where features representing a collection of stories and tasks not yet broken down are estimated, these collection of work closer to each other (hence affinity) are estimated in abstaction units such as T-shirt or Coffee-cup sizes. Hence such estimates are called affinity estimates in the backlog!

Budget Estimation

• As the project continues with planning, we look at increasing our confidence in the estimate. So, we get down to decomposing the details, such as the parameters required to estimate or seek opinion from multiple experts to narrow our estimate. Since we apply the parameters (number of rooms to paint * amount of paint required * price per paint can; number of virtual machines on the cloud * number of active hours * price per hour), we call it parametric estimation. The details of the parameters applied vary based on the industry and the project. 
• When it comes to seeking expert's opinion, instead of seeking estimate from estimate alone, we seek 3-points (optimistic, most likely, and pessimistic) to get the average. This average applies the statistical principle of central tendency to even out the variations. The same logic applies with PERT (program evaluation review technique) where multiple points are observed where the princple of normalization (bell curve logic) can be expected. So, PERT becomes a special case of triangular estimate. 
• For adaptive projects, the parametric thoughts carry forward. So, a team looks at a feature in relation to another feature (either delivered or to be delivered). So, if the new feature is twice the size of another feature which is either of small T-shirt size or 3 points, then, the new feature is either medium size or 8 points. These estimates may apply at the release level backlog giving the definition of ready (DoR) following the DEEP (Detailed Appropriately, Estimable, Emergent, Prioritized) property.
• Naturally, there is more time required to identify the parameters, seek opinions from multiple people, and perform this estimation. As the level of acuracy increases between -10% to +25%, the duration taken to estimate also increases.

Bottom Up Estimation

• When the project continues in its later stages of planning, then, increased level of accuracy and confidence are required to allocate work to someone and spend the money. Consequently, teams engage in more granular breakdown of tasks (activities) or modules (function points) so that these activities can be estimated, dependencies understood, and lead/lag factored appropriately to compute the 'definitive' estimate.
• In the case of adaptive projects, here is where the features or user stories are further broken following the INVEST (independent, negotiable, valuable, estimable, small, testable) property so that they are good candidates for sprint/iteration planning. In the sprint planning, the team does the cumulative estimate applying the central tendency to arrive at the stroy point (hence story pointing). The planning poker player at this stage is to faciliate the team level estimate as the expectation at the scrum level is that anyone should be able to pick up any story (cross-functional, multidisciplinary, T-shaped skills based, self-managed team). As work is constantly refined through backlog refinement, release level planning, and iteration level planning, the agile teams gain significant efficiency in sprint planning as the time taken to do this estimation is spread over time. 

Figure: Estimation Illustration by Sriram Rajagopalan

As you can see, the entire art of estimation is a classic risk management exercise where not only the management stakeholders but also the delivery team converge on the constraints, assumptions, risks, and dependencies (that I call as the CARD of the business goals and objectives). No wonder project management is not only a science but an art!

Thoughts? What do you think?