Which Estimation Units Should Your Teams Use For Work Items?Why estimate in the first place?
- At the feature level, you want estimates so you can figure ROI (Return On Investment), so the business and product owners can prioritize one feature against another
- At the roadmap & backlog level, most organizations want to be able to forecast, be it at the feature level or release level, so you need to be able to understand what is your organization’s capacity, and translate it into plans with dates and deliverables, milestones etc.
- At the finer-grained, work-item level (typically User Story), you want to help teams understand their capacity so they can take on the right amount. In particular in teams that use a frequent planning cadence, such as that prescribed by Scrum
- Still at the team level, team members setting expectations on the amount of work they plan to complete can surprise themselves, learn, and improve.
- At the team level, you want to understand whether a work item is small enough, might want teams to discuss estimates as part of creating a common understanding about work and align their views
What attributes should you look for in units you use for work-items-level estimates?I have listed here some attributes, prioritized based on my view (most important first):
- Intuitive – The units should make sense to newcomers, and to stakeholders outside the team itself. If it is opaque, it hurts transparency, and makes it hard for team members to get their heads around it.
- Encourage team thinking, which is another way to say seeing the big picture. One of the main points in agile methods is the emphasis on the team, rather than the individual expert. Additionally, Lean Thinking puts an emphasis on optimizing the whole system, meaning that if I’m a software developer, it’s better for the system if I focus on getting user stories done-done, rather than focusing on finishing as many of my programming tasks as I can.
- Just enough precision – Estimation can be wasteful, in particular if we do it too early and too detailed. Estimates will be wrong anyway (the very definition of “estimate” is “an approximate calculation”). Think about estimating the duration of a trip from Vegas to LA next month. Does it really make sense to argue if it is going to take 4:00 hours or 4:10?
- Be indifferent to the expertise of individual team members – If there are two back-end developers in the team, one is very experienced whereas the other is a newbie, they are bound to provide very different estimates if asked for hours to complete this. If we can somehow find a way to neutralize that effect, it would be swell.
- Scales up to multiple teams – In cases where a product or project requires the work of multiple teams, you will want to aggregate estimates, and combine teams velocities, so you can plan and track the whole effort, and coordinate dependencies using the same language in terms of progress and size.
- Minimize the risk of confusing effort and duration – One of the struggles of software development estimation, planning and tracking, is the common practice of using time-based estimates (as in “this will take 5 days”). The problem here is that the very same unit is used for duration. After all, planning in software development is a type of rate problem, like the time and distance problems your kids hate in middle school. Think about discussing a trip from Vegas to LA. How far are they? (distance) – 4 hours away? No, it really depends on your speed and whether you stop along the way etc. If you take a plane, the duration will be about 30 minutes. If you drive on Saturday (peak time) and you stop for a shopping spree at Ontario Mills, it may take well over 12 hours. The distance, however, is around 260 miles. It would be confusing to say that the distance is 4 hours, yet, this is exactly what we do in software development. Can we find units that sidestep that problem?
- Promote ingestion of small work items – One of the main struggles of teams moving to agile, is the need to break down work into small chunks. If the estimation “language” we use can challenge us to slice those work items thinner and thinner, that would be a benefit.
- Easy to reconcile with feature-level estimates. You will typically use a hierarchy of work items. From the large initiatives and roadmap items, down to features (or MMFs – Minimally-Marketable-Features), to team-level work items, the building blocks we typically call User Stories. If you use one unit for the features, and another unit for work-items, you want to have an easy way to map from one to the other, so you can track progress without suffering a discontinuity.
- Creates a fair economy of velocity – This is a culture thing. Measuring productivity in software development teams is impossible (there are saner alternatives, though), and any naive and misguided attempt to do this using numbers amounts to senseless bean-counting that is pretty much guaranteed to achieve the exact opposite of what you want. Ideally, the unit you use would make it hard for the naive manager (or executive) to go down that path. Your goal with work-item estimation units is to help the team do their work and plan their capacity, not help executives wring the last bit of “productivity” out of the team.
- Good at accommodating a variety of item sizes – Some teams find themselves handling both tiny user stories one team member can bring to completion in a single day, and large user stories they can barely complete in two weeks, even if multiple team members work on them. Ideally, the unit we use can accommodate both extremes.
- Help validate capacity allocation for different skills in the team. For example, if all the stories the team takes on are all very heavy on back-end development, and the team has a limited amount of back-end development expertise, the total estimate of all those stories won’t be a good representation of the team’s ability to complete these.
What are your options?
- Task-level time. Work items are broken down into tasks, and each of those tasks is estimated by the expert in the matching area. The breakdown can be done by expertise, where tasks look like “design”, “back-end development”, “front-end development”, “testing”, or can be broken down in a more fine-grained level like “Add field to API”, “Document API change”, “Write unit test” and so on. Task estimates are then rolled up to the work-item level, and you have a total estimate, in addition to skill-based estimates (e.g. this User Story will take 12 hours of back-end development, 3.5 hours of front-end development, and 4.25 hours of testing). This is often the default for teams that are just getting started with agile, and they will often keep doing it this way. Note that some teams use one of the methods listed below, and later on (just before the sprint/iteration starts, as part of some pre-planning) will perform that task breakdown.
- Story-level ideal-days. The team just estimates the total amount of effort needed to complete this work item, including testing, development, design etc. The effort is calculated in days of work. Ideal in the sense that we are assuming that a single person is working on this work item at any point in time, without any interruptions and distractions.
- Team-weeks. This is equivalent to the Ideal Days, but it emphasizes the team aspect by asking “If the whole team could, and would work solely on that work item, how many weeks would elapse before it was complete (done-done)”. This does take into account the normal amount of interruptions and distractions. Numbers will typically be less than one, assuming a large enough team and small enough work items.
- Story points. This is a relative, unit-less indication of effort. The team identifies a small (possibly the smallest) work item, and, looking at other work items, asks “how many times much effort is needed to complete this work item, relative to that small item”. This relative sizing has a few advantages.
- No-estimates – This approach, proposed by Vasco Duarte, suggests that estimating is dysfunctional by nature, and that counting the number of stories that can be complete in a period of time is the best predictor of progress. A variation proposed by Yuval Yeret proposes using “Equivalent # of stories” at the feature level, and perhaps using this as well for unusually larger stories
Okay, now let’s compare those:
Task-level time (hours or days)
No-estimates (just count stories)
Definitely, this method is very common even pre-agile
Story points is an abstract concept which is hard for some to grasp. Definitely not intuitive to business stakeholders who see that as a chinese wall for teams to hide behind
Encourages team thinking?
No, counting hours by expertise (testing, back-end…)
Estimate aggregates all expertise needed to complete this item
Yes, the estimates reflect overall effort for the team
Yes, the estimates reflect overall effort for the team
Very precise… too precise to be used until the work is about to start. Requires a lot of effort in estimating
Uses Fibonacci series which provides a variable resolution
Indifferent to individual team member’s experience?
May require assignment of tasks to individuals so they can estimate how much it will take them
For me it is 4 days, for you it may be 1 day (you have more experience, knowledge of that area…)
The team is an amalgam of skills and level of expertise, whose totality is reflected in this estimate
Regardless of whether you are an ace developer or a newbie, a 5-pointer should be around 5 times more effort than a 1-pointer
Scales up well to multi-team level?
Somewhat. Can easily be translated to days or weeks of work, although it is likely to be tedious
Not very well. The whole concept is based on a team deciding what does the 1-point story represent in their world. When you scale to the release train and value stream if is very hard to aggregate the story-points-based velocity of teams into a value-stream, release-level velocity
Story-based velocity can be aggregated across teams.
Same as story points. My typical story is not necessarily your typical story.
Risk of confusing duration and effort?
Bean-counting at its best
“Why does that 5-days story take two weeks? You have 4 spare days in your sprint, can’t you squeeze this one in?”
Easy to reconcile with feature-level estimates?
Time arithmetic is straightforward
At least at the single team level, it is fairly straightforward
Whether you use story points as well for features, or T-shirt sizes, you can map these to the story points you use in stories
Somewhat. For a specific team, you’d need to estimate features as “number of stories”, which is feasible.
Creates a fair economy of velocity?
No, teams are managed by the numbers, creating a pressure to deliver hours, and likely to create a burn-out and tech-debt
Makes it harder to
Yes, team focuses on getting things done, and getting it done the right way, vs. just “making good time”
Good at accommodating large variety in work-item sizes?
Yes, anything that can be broken down to tasks can be estimated
The scale goes to infinity pretty fast, and you need to revert to using a different scale for large items
At the sprint level, if a team has a significant variety in the size of work items, # of work items won’t be very helpful in deciding how much can the team take on
Helps in validating capacity allocation at the expertise level?
Yes, albeit in a dysfunctional way which keeps the team in their comfort zone of expertise silos. ( As opposed to tasks being created, estimated and assigned close to when the work-item is about to be worked on)
Estimates may hide imbalance in expertise needed
So, bottom line, which unit would I recommend?
- For the single team, story points are a good option if people don’t get goosebumps when they hear about the concept. So far my success rate is about 50%. Even if you do use story points, perhaps start with just using 1, 2, 4, too-big, or 1, 2, 4, 8, too-big
- For multiple teams, I would start with story-level ideal-days (which is effectively what SAFe uses), and use a variable resolution scale (e.g. Fibonacci).
- For both cases, you should collect data for a some time (e.g. three months), and examine if, in hindsight, counting stories is a good enough approximation for velocity, and if so, consider moving to a no-estimates-based approach
- If the story-based velocity is pretty stable for all teams, you can move to a “no-estimates” kind of setup, and liberate your teams from bean-counting mentality
- Joshua Kerievsky on why you should stop using story points https://www.industriallogic.com/blog/stop-using-story-points/
- Is Velocity killing agility? http://jimhighsmith.com/velocity-is-killing-agility/
- Vasco Duarte on no-estimates, see http://noestimatesbook.com/ , and Yuval Yeret’s blog on using this in SAFE, with some suggested improvements. Also at http://plan.io/blog/post/145902333978/noestimates-6-software-experts-give-their-view