The Purpose Of Estimation
A common scenario runs like this:
- Developers are asked for (or given) estimates for upcoming work.
- These tasks and estimates are turned into release plans tracked with burn-down charts
- Time and effort goes into monitoring progress against these plans. Everyone is upset when actuals end up being more than estimates. In effort to increase pace with the estimates, developers are told to sacrifice quality, which only makes things worse.
People are optimists, so these estimates tend to be too low, even without pressure to make them low (and there's usually at least some implicit pressure)
In this narrative, effort put into estimates is, at best, waste - since "an estimate is a guess in a clean shirt" . Usually estimates end up being actively harmful as they encourage FeatureDevotion, a nasty condition where people start valuing ticking off features more than tracking the real outcome of the project.
Estimates also set expectations, and since estimates are usually too low, they set unrealistic expectations.  Any increase in time or reduction in features is then seen as a loss. Due to loss-aversion, these losses have a magnified effect. 
Faced with situations like this, it's easy to see how people turn their angry glares towards estimation. This leads to an increasing notion that anyone indulging in estimating is an Not a True Agilist. Critics of agile say this means that agile development is about developers going off and doing vague stuff with promises that it'll be done when its done and you'll like it.
I don't share this view of estimation as an inherently evil activity. If I'm asked if estimation is a Bad Thing my answer is the standard consultants' answer of "it depends". Whenever someone answers "it depends" the follow-up question is "upon what". To answer that we have to ask why we are doing estimation - as I like to say "if it's worth doing well, it's worth asking why on earth you're doing it at all".
For me, estimation is valuable when it helps you make a significant decision.
My first example of an estimation-informed decision is allocation of resources. Organizations have a mostly fixed amount of money and people, and usually there are too many worthwhile things to do. So people are faced with decisions: do we do A or B? Faced with such a decision it's useful to know how much effort (and cost) each will involve. To make sensible decisions about what to do, you need to have a feel for both the cost and the benefits.
Another example is to help with coordination. The blue team wants to release a new feature to their web site, but cannot do so until the green team builds a new service to give them crucial data. If the green team estimates they will be done in two months and the blue team estimates that it will take them a month to build the feature, then the blue team knows it's not worthwhile to start today. They can spend at least a month working on some other feature that can be released earlier.
So whenever you're thinking of asking for an estimate, you should always clarify what decision that estimate is informing. If you can't find one, or the decision isn't very significant, then that's a signal that an estimate is wasteful. When you do find a decision then knowing it focuses the estimate because the decision provides context. It should also clarify the desired precision and accuracy.
Understanding the decision may also lead you to alternative actions that may not involve an estimate. Maybe task A is so much more important than B that you don't need an estimate to put all your available energies into doing it first. Perhaps there is a way for blue team members to work with the green team to get the service built more quickly.
Similarly, tracking against a plan should also be driven by how it informs decision making. My usual comment here is that a plan acts as a baseline to help assess changes - if we want to add a new feature, how do we fit it into the FivePoundBag? Estimates can help us understand these trade-offs and thus decide how to respond to change. On a larger scale re-estimating a whole release can help us understand if the project as a whole is still the best use of our energy. A few years ago we had a year-long project that was cancelled after a re-estimate a couple of months in. We saw that as a success because the re-estimate suggested the project would take much longer than we had initially expected - early cancellation allowed the client to move resources to a better target.
But remember with tracking against plans that estimates have a limited shelf life. I once remember a gnarly project manager say that plans and estmates were like a lettuce, good for a couple of days, rather wilty after a week, and unrecognizable after a couple of months.
Many teams find that estimation provides a useful forcing function to get team members to talk to each other. Estimation meetings can help get better understanding of various ways to implement upcoming stories, future architectural directions, and design problems in the code base. In this case any output estimation numbers may be unimportant. There are many ways such conversations can happen, but estimation discussions can be introduced if these kinds of conversations aren't happening.  Conversely if you're thinking of stopping estimation, you need to ensure that any useful conversation during estimation still continues elsewhere.
Go to any conference with agile leanings and you'll hear talks of teams that work effectively without estimation. Often this works because they, and their customers, understand that making estimates isn't going to affect significant decisions. An example is a small team working closely with business. If the broader business is happy with allocating some people to that business unit, then work can be carried out in priority order; often this is helped by the team breaking down work into small enough units.  A team's level in the agile fluency model plays a big role here. As teams progress they first struggle with estimation, then can get quite good at it, and then reach a point where they often don't need it. 
Estimation is neither good or bad. If you can work effectively without estimation, then go ahead and do without it. If you think you need some estimates, then make sure you understand their role in decision making. If they are going to affect significant decisions then go ahead and make the best estimates you can. Above all be wary of anyone who tells you they are always needed, or never needed. Any arguments about use of estimation always defer to the agile principle that you should decide what are the right techniques for your particular context.
1: A recent read is Estimation is Evil an excellent discussion by Ron Jeffries of the problems that estimates can cause.
2: I got this analogy from Ron Jeffries, although I don't have a written reference for it.
3: This situation is an excellent example of an inappropriate use of metrics.
4: Estimates and expectations
I particularly liked a comment on this by my colleague Angela Ferguson "the way that estimates set expectations is up to us - it is poor project management (whether by project managers or other team members) that results in a client who thinks estimates are fixed, or that raw estimates = actual effort/duration"
"I, in fact, try to practice delivery bad news on a weekly basis with my key client, even when things are travelling as expected ... 'so we're looking quite well on track now, but if we had discovered something that took longer than expected, or a requirement had blown up to be larger than expected, or we found something new and very important, what do you think the best course of action would be?' And then you explore the options - cut stories, add time, add capacity, etc. This means that when the expected unexpected thing happens (because we know it will happen), the conversation doesn't seem new and scary to the client."
5: Very roughly people feel twice as much pain for a loss as pleasure for a gain.
6: If you do this an approach like ThrownEstimates can help the discussion move at a good pace.
7: Of course breaking work down into small units requires some implicit estimation, but that's really a different animal to the more common explicit estimation activity.
8: James Shore has a recent blog post that details his observations about how fluency influences estimation practice. I think a similar analysis of practices at various stages of fluency could be very useful.
AcknowledgementsI repeat myself again in thanking varous ThoughtWorkers on internal lists for their comments. In particuarly I'd like to call out Angela Ferguson, Dave Pattinson, and Pat Kua. I should also thank James Shore for responding so rapidly and well for my question about the links to the fluency model.
(Note: Opinions expressed in this article and its replies are the opinions of their respective authors and not those of DZone, Inc.)