If passion were a measure of utility, the proponents who believe in #NoEstimates (the twitter hashtag representing the “No Estimates” movement) would be more valuable than oxygen. They believe there are better ways to manage and fund projects that don’t require pulling your best and brightest from their real job to estimate (and re-estimate) how long each part of a software project will take.

It’s hard not to be with them in spirit. Anybody who has spent time in the software industry has felt the futility of estimating raw ideas that will inevitably change over time and evolve toward something far different than the original single-line description. But we can’t forget that we are being paid by a business (someone else’s money), and all businesses have limitations of one or more resources—people, money or time—which must be spent in the most effective manner possible.

And it is here that some form of forecasting is required for new ideas, to make sure that the expenditure of scarce resources are yielding the biggest bang for the buck. Impact of potential revenue or cost savings needs to be assessed over a timeframe, and the cost to achieve those benefits needs to be broadly understood. Someone needs to plan and understand cash flow to keep the company thriving. Enter the estimate.

Blame for poor estimation outcomes is wrongly placed. Poor mathematics, failing to estimate the impact of risks and dependencies, and ignoring uncertainty (entirely) are having much more impact than a few developer stories being underestimated. Narrowly defining estimates to “developer story size,” and then recommending doing away with all forms of estimation, is a shortsighted approach.

There are good reasons to avoid estimation, the most striking being when the work is a foregone conclusion. Extending “No Estimates” to all possible work is nonsense unless there is truly no option of altering staffing levels, or reordering feature priorities based on time to market. In the idea-rich software atmosphere, there is always more demand than supply.

Why do current estimate techniques fail so badly? Looking at the most common technique of story points and velocity projection, we start to form a picture of hopelessness. If a team is stable in size, has little turnover in experience level, no external dependencies or risks, and all scope is known upfront, then maybe, just maybe, velocity will give an accurate forecast.

The concept behind velocity is pure simplicity and genius; it’s a way to convert one unit of team performance and effort into a unit of time. Problem is, it doesn’t cope well with the kind of changes in the (abbreviated) list from a few sentences ago. Another shortcoming is it’s an average linear projection and becomes inaccurate when modeling a non-linear process, such as team ramp-up time, risks, and external team dependencies that occur rarely but wreak vengeance on velocity when they do—much more significant than a few inaccurate developer estimates.

The forecast we need is how long it takes work to flow through the entire organization, as most aspects beyond pure coding time have little to do with the work complexity. Waiting for a tester or production hardware isn’t correlated to small, medium or large story size values. Velocity assumes there is.