How can we be Agile and still satisfy estimation requirements for ML5?

I work in an ML5 organization. We're experimenting with scrum to see what benefits can be gained over our traditional waterfall model. But I'm stuck on the estimating part! When I've done scrum in the past we've done some planning poker and called it a day. But the head of my SEPG tells me that wideband delphi estimation is frowned upon at the high maturity levels because it is akin to guessing. Can you recommend how we can perform agile estimation in a way that's compatible with HM organizations?

So what's wrong with guessing?

I'm not sure your problem is as much about "Scrum" as it is about the inflexibility and lack of experience within your SEPG! The CMMI does not speak to specific estimating methods and while "planning poker and calling it a day" may not be enough to satisfy the CMMI, there isn't a requirement to perform more "traditional" methods such as SLOC, KOKOMO, or the like either.

There have been whole books written on high maturity and estimating, so now is not the right time to get into all of the glorious and fun-filled detail but I'll touch on some of the concepts.

As to Wideband Delphi being "akin to guessing" I guess I would say "it depends" (I'm not saying it's a "high-maturity practice" either). To say that is akin to saying that a design that has been put together by an expert software architect, and then peer reviewed by other experts, and then prototyped is akin to guessing on a design. It isn't. We call all of those things TS, VER, and VAL don't we?

But, on to the meat of it.....

The difference between a ML2/3 organization and a ML5 organization isn't in how they do things, but it is in the data used to make the decision on how to do things. The essence of ML5 is trying new and innovative techniques to improve performance based on data gathered through the execution of the process (like Scrum for instance).

Someone in your organization should be gathering appropriate data, developing baselines (determining the natural bounds of process performance), and performance models using various statistical methods. That data should be used to estimate and plan projects as well as select potential innovations (such as the use of Scrum) for piloting and eventual deployment.

So, if I were to use historical performance baselines as input into a wide-band delphi process that was intended to "feed" a scrum project, and I was doing so because I had identified Scrum as an innovation that could improve performance (supported by data that told me that phase and cost overruns were a problem we a problem) then I would be behaving exactly like a High Maturity organization was supposed to perform.

Of course, that's a hypothetical situation. You'd have to determine your own thread for getting value from Scrum and ML5, but if you read our latest SEI Technical note "CMMI and Agile: Why not Embrace Both" you'll learn a lot more about it. Get it at: http://www.sei.cmu.edu/publications/documents/08.reports/08tn003.html

http://www.broadswordsolutions.com/