View my profileagile
Why Relating Story Points to Hours and Money is Risky
05/08/09 In: Agile estimating
One of my clients is a small software development house that does custom development in the form of development projects for clients . I helped them to successfully introduce Agile (Scrum with XP) and both the team and business managers are really happy with it.
As they liked our methods of planning and estimating (story points and velocity) the account managers and sales team were discussing the options to relate story points to dollar values.
To explain why I think this is very risky and not advisable I wanted to give them some background.
“How big” vs. “How long”
Story points are units that are used to size a piece of functionality or work. Sizing in this case means that story points indicate "how big" a piece of work is.
This is often confused with "how long" it takes to implement it but in fact "how big" and "how long" are very different things:
- The "how long" is highly dependent on which developer is performing the work
- The "how big" bears no relationship to who is performing the work
If we use an analogy of several piles of dirt in a room I can say that one pile of dirt is bigger than the other. However, if the work then consists of moving one pile of dirt from one end of the room to the other by carrying small amounts of dirt on a spade it might turn out that Peter is a lot stronger and faster than Paul and that it would take him 2 hours to move the pile while it would take someone slower and more junior like Paul 6 hours to complete the job.
Don’t assign work upfront
Therefore, in order to estimate the "how long" at the beginning of a project or iteration we would need to know which programmer will do the work at the time we estimate. This is highly impractical as it would lead to bottlenecks when developers are waiting for each other to finish their pieces of work. It could cause a chain reaction of delays triggering down the project and people being over or under-utilised.
How big?
What we can do on the other hand is to estimate the "how big" part. Using the piles of dirt analogy we can say that a pile of dirt is 40ccm with much more certainty than we can estimate how long it will take someone to move it to the other end of the room - especially if we don't know who will carry out the work yet. As software functionality is a lot harder to measure than the volume of a pile of dirt we introduced story points. Instead of saying that a piece of work is 40ccm we say that its size is e.g. 3 story points. Essentially we are applying the same principle of measuring size rather than time.
How long?
We still need to come up with a measure though for how long it will take to move all piles of dirt to the other end of the room, ie how long it will take to finish the project. One way of doing this is to size all known pieces of work and then to perform the work in timeboxed iterations of e.g. 2 weeks. At the end of each iteration we will know how many story points the team have implemented and after a number of iterations we will know that the capacity of the team is roughly X points pr. sprint.
Using this capacity number (called velocity) we have achieved the following:
- We can plan ahead in terms of time: If a team normally achieves 10 story points pr iteration for a project with 100 story points we would need 10 sprints.
- We can plan ahead in terms of budget: Given the number of resources on a project is constant we know the number of hours in an iteration and therefore the price of a sprint. If we know roughly how many sprints are needed in a project we have an indication of how much it will cost overall.
- We have managed to make estimates independent of who will be doing the work. Incorporating the "how long" into team capacity rather than the piece of work unit makes estimates independent of future work allocation.
Sizes are relative
As we, unfortunately, can’t measure the size of software functionality as objectively as piles of dirt we size pieces of work relative to each other. We have no way of saying that a piece of work is e.g. a universal size of 3 which is why we arbitrarily start with some piece of work and assign it a size. All subsequent pieces of work (user stories or features) can then be measured against this base story as being smaller, bigger, much bigger etc. For this we use the Fibonacci number scale (1,2, 3, 5, 8, 13) which helps us size all user stories relative to each other.
Which number we start out with does not really matter as the relative sizing in combination with team velocity is sufficient for us to plan ahead.
Sizes depend on context
It is important, though to note that story sizes are calibrated against an arbitrary base chosen by the team and that those sizes are by no means absolute or objective. In fact the sizes and velocity are only valid for this particular team and velocity will also change over time as the team increases capacity by getting better at what they're doing. Also, often functionality implemented by one team is assigned a widely different number of story points from another team implementing the exact same functionality (the same code might be implemented using 30 story points by one team and 140 points by another).
Overall, this system works well as long as story sizes and team velocity (ie the "how big" and "how long") remain independent of each other and as long as we keep in mind that story sizes are derived as arbitrary values relative to each other within a project.
Using story points, potentially even across projects, to equate directly to $$ is risky at best.
(Credits for coming up with the “pile of dirt” analogy to Mike Cohn)
|
Tools - the new kid on the block
26/10/08 In: Agile tools
Mingle has received a fair bit of attention lately and the latest version, 2.1, has just been released.
I’m quite happy with Rallydev but as Mingle is made by Thoughtworks themselves and the guys at Silverstripe keep raving about it I thought I’d better check it out. Also, as much as I like Rallydev and VersionOne - they’re both a bit weak on the usability side.
Below a list of what I like and dislike about Mingle:
Things I like:
1. Installation & Support
- Free web/phone conference with a Mingle representative to guide you through configuring the system after you have received a free hosted trial. Very helpful!
- A breeze to install the downloadable version
- Runs on all operating systems (incl. Max OSX, Linux and Open Solaris) and most database systems incl. Postgres
- It just looks nice and appealing
- Tasks can be completed with relatively few mouse clicks
- If you choose the scrum template the card hierarchy of sprints - stories - tasks is very intuitive and standard scrum
- You can create epics for your product backlog
- Given you’re willing to spend enough time playing with configuration settings and are inclined to learn Mingle’s proprietary SQL-like language you can configure the tool to fit exactly your needs. Hell, you could probably even extend their JRuby ;-)
- Mingle contains a fully fledged wiki - great for sharing project information!
- All reports are configurable on the wiki
- As expected, very easily done from Excel (For switchers it’d be great to have an automated import/export to/from Rallydev and VersionOne)
- Great that it lets you preview your import before starting the actual import.
Things I don’t like:
1. Concept of cards for everything
- Representing everything as cards is confusing: It does make sense for user stories, which in the physical world are written on story cards but using cards for releases, sprints and tasks seems forced and non-intuitive
- I don’t want to fit my mental models around a tool - it should be the other way around! Scrum concepts such as sprints and releases are a bad fit for Mingle’s generic object model of cards
- I miss the nice overview I have in Rallydev where I can see all my release and sprint dates
- Card trees fill quite a big part of the screen and it’s hard to get an overview at a glance (There’s another, smaller view but it doesn’t show the relations very well)
- Sprints and releases don’t have fields for start and end dates (I can add them myself by configuring cards but not adding them by default seems odd)
- There seems to be no standard way of getting started with a project quickly. There is no step by step process to quickly set up your releases, sprints, product backlog and user stories. Unlike Rallydev which guides you through with a pre-defined process and video documentation Mingle makes you figure it out for yourself.
- Mingle expects you to invest a considerable amount of time learning how to configure the system and how to make it work for you.
- There is no free SAAS version available (By comparison Rally Community edition is free for up to 10 users and Version One’s team edition provides similar functionality for free)
- Hosting Mingle on your own server is free for up to 5 users but I don’t want to wast my time installing tools and maintaining a server.
- The hosted trial for one project is great but only being allowed to keep the instance for 2-3 week is not enough for me to try Mingle on a real project and to make a decision to move all my projects from Rally or VersionOne
Overall, this is a promising tool with an attractive user interface.
It is nice that the tool is extensively configurable but it is way too time intensive to set up. The point is that even though I can configure everything I don’t want to: I just want to get going quickly with 80% of what I need and focus on getting my projects done. Spending time on configuring a tool seems like a waste of time.
As Mingle still seems a bit rough around the edges and does not provide a killer feature that would justify the hassle, learning curve and additional licensing cost I won’t switch from Rallydev this time.
I’ll definitely keep an eye on future releases though!
Agile requirements
08/10/08 In: Agile user stories
How to pick a Scrum team
28/09/08 In: Agile teams
I was recently asked by a friend how to pick a Scrum team.
My friend is a project manager within a New Zealand government department and to deliver an important project he was given complete freedom of choice with regards to project methodology and people.
He chose Scrum as a delivery framework and needed 8 people to form his dream team. The big question was which people to pick and which skills to focus on.
I've never been fortunate enough to pick my own scrum team. Normally, I get a certain degree of choice and then a combination of who happens to be available and people who I really, really fight for. And sometimes I fight to avoid getting someone on the team.
Having complete freedom I think I'd focus on:
Sure, people should be qualified and good at what they do but it is much easier to teach people skills than to perform personality transplants.
My friend is a project manager within a New Zealand government department and to deliver an important project he was given complete freedom of choice with regards to project methodology and people.
He chose Scrum as a delivery framework and needed 8 people to form his dream team. The big question was which people to pick and which skills to focus on.
I've never been fortunate enough to pick my own scrum team. Normally, I get a certain degree of choice and then a combination of who happens to be available and people who I really, really fight for. And sometimes I fight to avoid getting someone on the team.
Having complete freedom I think I'd focus on:
- Personality and work attitude - people who are solution oriented, like to deliver, like to succeed, are easy to work with, can communicate and just get the job done
- People who like each other or have a chance of getting along, i.e. have respect for each other and can have fun together
- Absolutely no primadonnas: We can all be flexible and accommodate a lot but the lone genius who lives by his own rules just does not work (I usually remove them and I have never regretted it)
- People who are confident enough to handle the visibility and transparency that come with Scrum
Sure, people should be qualified and good at what they do but it is much easier to teach people skills than to perform personality transplants.
Design Chunking with Scrum
27/09/08 In: Agile user experience
When I started with agile (Scrum) software development five years ago one of the main challenges I faced was combining an agile development approach with user-experience driven website design.
Especially, as we were working on a global, consumer-oriented web site with a strong focus on product branding we needed to make sure that we could benefit from the agile development approach without losing focus on a consistent user experience (UX) and information architecture (IA) across the site.
Agile frameworks such as SCRUM and XP say absolutely nothing about the IA and user interaction design processes, which meant that we had to learn mainly by trial and error.
Most of the problems we faced originated from a clash of two fundamentally different world views:
Our experiments:
Attempt 1: Concurrent IA, design and development resulted in “piece meal” design and user interaction. We lost track of the big picture and lacked overall IA/design consistency across site. Also, our product owners found it difficult to keep an overview of the project.
Attempt 2: Upfront IA and graphic design caused problems with unused and out-of-date designs and feature specifications. Sometimes we finished an iteration or release with partially implemented features - very non-agile! We also found it difficult to adapt to feedback after the IA/design phase was completed.
Attempt 3: Post skinning (applying design after functionality has been developed) caused our implementation to drift from the design and user interaction intent and we often faced excessive and unnecessary re-coding and re-factoring work. The product owners and end users found it difficult to give feedback as non-shippable increments were demonstrated.
After three failed attempts we still wanted to utilize the advantages of an agile approach that would allow for continuous inspection and adaption and could help us to keep the focus on the most important parts of the site. Basically, we wanted to keep the agile advantages without sacrificing the overall user experience.
Attempt 4: Design-chunking - success!
(This term was originally coined by Lynn Miller)
With this approach we defined initial concepts upfront and subsequently broke the IA and user experience design apart into smaller pieces that could be delivered incrementally.
Design-chunking process:
The most important break-through came from two changes:
Decoupling of IA/design and development:
This approach worked very well for us, we delivered nine releases to a happy client and in terms of process we felt we got the best of both worlds.
I would be very interested to hear your thoughts on this approach and particularly any experiences you may have had with similar issues in your projects - especially large and/or global web-based projects.
Also, an absolutely excellent description of this approach was done by Lynn Miller in her paper on Adapting Usability Investigations for Agile User-Centered Design. Much of my article is based on and inspired by her work. I only wish I had know about it before we started our extensive period of trial and error.
Especially, as we were working on a global, consumer-oriented web site with a strong focus on product branding we needed to make sure that we could benefit from the agile development approach without losing focus on a consistent user experience (UX) and information architecture (IA) across the site.
Agile frameworks such as SCRUM and XP say absolutely nothing about the IA and user interaction design processes, which meant that we had to learn mainly by trial and error.
Most of the problems we faced originated from a clash of two fundamentally different world views:
- UX people are driven by a holistic approach (big picture)
- Dev teams focus on iterative and incremental development
Our experiments:
- Attempt 1: Concurrent IA, design and development
- Attempt 2: Upfront IA and graphic design
- Attempt 3: Post skinning
Attempt 1: Concurrent IA, design and development resulted in “piece meal” design and user interaction. We lost track of the big picture and lacked overall IA/design consistency across site. Also, our product owners found it difficult to keep an overview of the project.
Attempt 2: Upfront IA and graphic design caused problems with unused and out-of-date designs and feature specifications. Sometimes we finished an iteration or release with partially implemented features - very non-agile! We also found it difficult to adapt to feedback after the IA/design phase was completed.
Attempt 3: Post skinning (applying design after functionality has been developed) caused our implementation to drift from the design and user interaction intent and we often faced excessive and unnecessary re-coding and re-factoring work. The product owners and end users found it difficult to give feedback as non-shippable increments were demonstrated.
After three failed attempts we still wanted to utilize the advantages of an agile approach that would allow for continuous inspection and adaption and could help us to keep the focus on the most important parts of the site. Basically, we wanted to keep the agile advantages without sacrificing the overall user experience.
Attempt 4: Design-chunking - success!
(This term was originally coined by Lynn Miller)
With this approach we defined initial concepts upfront and subsequently broke the IA and user experience design apart into smaller pieces that could be delivered incrementally.
Design-chunking process:
The most important break-through came from two changes:
- We introduced an iteration (sprint) 0
- We de-coupled IA/user experience/graphic design from development
- done before development begins
- short (2-3 week) IA/user experience/graphic design only iteration to lay the groundwork
- defines “big picture” on which subsequent designs and not yet fully defined areas of the site will be based
- outcome 1: overall strategy for IA (site-wide elements such as page grid, navigation, rough placement of modules)
- outcome 2: overall design concept including site-wide design guidelines
- exact content balance between IA and design will be based on whether an IA or design led approach is used
- signed off by client (partial IA and design sign-off) and discussed with developer to assess technical complexity
- iteration 0 only covers fundamentals and potentially the details of the most important page. Details for other pages and areas of the site are still to be refined incrementally
Decoupling of IA/design and development:
- after iteration 0 decoupling IA/design and software development into two independent tracks
- tracks iterate separately but simultaneously and provide frequent feedback to each other
- the IA/design track feeds a refined design of the most important features/page(s) to the development track at the end of each design iteration
- the output from the IA/design track at the end of an iteration is used as input to the subsequent development iteration
- for the IA/design track to feed the development track the IA/design track needs to be ahead in time - ideally this would be one iteration ahead with minor refinements still possible during the development iteration (e.g. change an image)
- focus on potentially shippable increments of functionality to provide feedback opportunities for the client/users
This approach worked very well for us, we delivered nine releases to a happy client and in terms of process we felt we got the best of both worlds.
I would be very interested to hear your thoughts on this approach and particularly any experiences you may have had with similar issues in your projects - especially large and/or global web-based projects.
Also, an absolutely excellent description of this approach was done by Lynn Miller in her paper on Adapting Usability Investigations for Agile User-Centered Design. Much of my article is based on and inspired by her work. I only wish I had know about it before we started our extensive period of trial and error.