Neil Kellick has a presentation on the alternatives to estimation on agile projects. Neil is a proponent of #NoEstimates.
This presentation has several interesting points around the #NoEstimates discussion.
- When you split work – page 16 – how do you know the size of something and when that size is the “right” size?
- What's the STOPPING rule.
- You make an estimate of when the size will be right and stop when you recognize that size has been reached.
- When throughput is in the denominator how do you know what is “possible” for throughput to the cycle time can be understood to determine when the customer will be receiving the feature?
- You use the probability distribution function shown on page 23 to estimate what possible throughput's can be achieved.
- You then look at the backlog, talk to the customer, and have a conversation about the acceptability of the current throughput in reaching the business needs.
- If that is not acceptable, you do something about it. Drop things or get more capacity for work
- That's called the Estimate to Complete
- Page 22 is an “estimate” of when the work will be done.
- I can see the backlog, know something about the capacity for work, and make an estimate of when we are going to be done with the project if no more work arrives in the backlog.
- If more work does arrive, I can estimate the impact on the project visually and with Little's Law if I wrote down the throughput number.
- Or I could make a probabilistic estimate using the distirbution of possible throughput on Page 23.
- Page 23 is an “estimate” of the cycle time. The probability distribution of possible cycle times shows what possible cycle times will impact the project. This is the method required by DI-MGMT-81861 on large defense programs for schedule. Similar requirements are in place for cost.
By the way, don’t use the mean, use the Mode – the most likely – this is identical to the Monte Carlo Simulations we use on large complex programs.
So while some in the #NoEstimates groups are adimate that "no estimating" is required, Neil has showed ways of forecasting completion dates, and forecasting cost using Little's Law and other standard agile practices. He's estimating. Or at least he could estimate if he chooses to. That's a different topic. But starting with the simple queuing model, estimates are built in to Neil's very nice presentation.
