If you ever find yourself wanting to inject some energy into a cross-functional gathering of delivery staff and stakeholders, ask for their opinion on estimates. For every person who adamantly insists that estimates are needed to support proper governance, someone else will argue that the inherent wrongness of an estimate and how estimates are abused will wipe out any benefits of defining them.
I'm not here to rehash the #estimates vs. #noestimates debate, but I felt it might be useful to understand what estimation techniques are being used by teams who are using agile approaches for delivering the scope of their projects.
There are numerous ways to estimate effort, cost or duration, but given the limitations of LinkedIn's polling capability, I ran a poll in the PMI Project, Program and Portfolio Management group with only the following four choices:
- Traditional techniques - this covers the range of estimation methods commonly associated with predictive delivery including analogous, parametric and bottom-up.
- Relative sizing - this covered such techniques as story pointing, T-shirt sizing, affinity estimating and others.
- Throughput and Monte Carlo - this refers to the process of forecasting based on an understanding of how many work items of a certain size can be completed within a fixed amount of time as well as using simulations to get a confidence level on how many work items might be completed by a given date.
- #NoEstimates - whereas the original idea presented by Woody Zuill in 2012 really meant not to estimate release cost, effort or duration and to focus on delivering small, valuable work items as quickly as possible to stakeholders, it has evolved since to refer to a more realistic use of estimation only when and where it makes sense.
Given the higher level of uncertainty and complexity present in projects which lend themselves to adaptive approaches, my assumption had been that the votes would mostly be cast in the latter three categories. Traditional estimation techniques work very well when there is reliable historical data or expert judgment which is relevant within the context of the current work being done.
At the end of the voting period, there was a tie for first place between the first two categories, each getting 42% out of the 37 votes cast. Throughput and Monte Carlo received 11% of the votes and #NoEstimates only received 5%.
The low volume for #NoEstimates is not surprising. While it might be the most pragmatic way of handling things, few teams have the support from leadership, customers and governance authorities to work in this manner.
Similarly, while a throughput-based forecasting approach combined with Monte Carlo simulations might be a better empirical approach, it does have a number of prerequisites to be successfully used, and also requires a fairly mature, disciplined team for it to work well.
I was expecting to see relative sizing receiving the majority of the votes, so I was surprised to see that traditional methods continued to dominate.
Just because something is popular, doesn't mean its effective.
