I teach project management and estimation workshops. Every time I teach estimation, someone asks, "How can we get the perfect estimate?"

You can't. Estimates are guesses. And, if you are multitasking, every estimate you have is wrong. Guaranteed.

But you still need to figure out when things will be done. Instead of estimating how long a task will take, I prefer to estimate how many small things I can complete in a week. And, I make sure I get to done on each task before I take on another task.

One of the problems of estimating large tasks is that we tend to estimate in units that are too large. If you estimate in weeks, you'll be off by weeks. In days, you'll be off by days. In hours, you'll be off by hours. Over the course of a project, even estimating in hours can obscure the actual duration remaining for the project.

Instead of using scope as the base, I use a timebox. Instead of asking, "How long will this take?" I ask, "What are all the small tasks required to complete this piece of scope?" Then I see how many of those will fit into a week.

If you're working with other people, and are using a Delphi method for estimation, you'll discuss your estimate. You'll hear things like "If Danny does it, it's only a 3. If I do it, it's a 5." I always take the upper estimate. Why? Because if Danny is so good, he's not going to be available. He'll have questions to answer or some other task that's a higher priority. I'll do it anyway. And, if I practice doing these tasks I'm not so good at now, I'll get better pretty quickly.

Think hard about whether you want a good estimate or if you want the work done. Estimating at the beginning of the project is often a waste of time. The more work you get done, the better your remaining estimate will be. 

I attended Microsoft's Professional Developers Conference last month, with one of my goals being to learn more about Azure, the company's cloud computing platform. I have been meaning to write a post on the subject for some time, but it did not come together until this week, when I listenened to Marc Benioff expounding the Salesforce.com version of cloud computing.

Benioff is ebullient and outspoken; a contrast to Microsoft's Ray Ozzie who gave the Azure keynote at the PDC and conveyed little excitement. Of course, it's easier to be ebullient when your company is growing at 20% per year; but another factor is that the Saleforce.com vision is easier to articulate. Throw out your servers, says Benioff, and move everything to the cloud, that is, to us.

In reality Salesforce.com has a significant integration story, so you can have your on-premise applications talk to your cloud applications, but it is not something Benioff talks about much. Why should he? His company profits when you migrate stuff to his platform; the integration piece is merely an enabler.

Microsoft by contrast makes its money from on-premise software. Its internet services lose money, according to its own accounts. It also has a vast partner infrastructure that is sustained by installing and maintaining its products.

Still, Microsoft knows that it has to do cloud, because the economic benefits to its customers are unavoidable, and because any IT company without a cloud strategy will be punished by the financial markets. In consequence, there is an array of consumer services under the Live brand, and an emerging platform of enterprise services under the Azure brand.

Azure is Windows server in the cloud. You get SQL Server; you get IIS; you get ASP.NET; and it is all pay-as-you-go. So is Microsoft suggesting that we throw out our servers? Considering the profitability of its server division, that would be madness; and indeed, the PDC presented mixed messages about the company's cloud strategy.

This was brought home to me by a session I attended on Bridging On-Premise and the Cloud, given by Windows Azure Distinguised Engineer Yousef Khalidi. Cloud computing, said Khalidi, is a "style of computing with dynamically scalable and virtualized resources provided as a service through the network ... this definition doesn't even say if it has to be on the internet or not. It's a way to think about how an application is structured to fit in a cloud-like model."

Khalidi went on to describe a "spectrum" of computing platforms, from the traditional server or datacenter to private and then public clouds. The forthcoming AppFabric Server will make it easier to run cloud-like applications (according to the definition above) on your internal network, with the option to move it to Azure later should you so desire. "Our strategy, our basic thesis guys, is that there are benefits across the whole spectrum and we'll continue to support the whole spectrum," he said.

The snag is that co-ordinating cloud and on-premise gets complicated, as a glance at one of his slides illustrates.

Well, computing is a complex business and I can understand the appeal of this model to Microsoft-platform companies that require some specific benefit, like pay-as-you-do scalability. As an overall proposition though, it is less attractive than the kind of thing Benioff talks about; it can feel like adding complexity rather than reducing it.

Another issue with Azure is that while it lifts much of the IT administrator's burden, it does little to speed development. You still have to write the code, test it, debug it, deploy it, maintain it. Saleforce.com on the other hand is not only a multi-tenant platform, it is a multi-tenant application. If you are lucky, a little bit of customisation is all you need.

You might not be lucky. You might end up having to write mountains of non-portable code in Apex, the Force.com language. You might run into limitations of the platform, like its difficulty with long-running data processing operations that can interfere with the responsiveness of the system. Salesforce.com is also an expensive platform, with per-user per-month fees for ever.

Still, at least Benioff has a coherent story. I'm not sure that Microsoft does.

The latest news is that Microsoft has re-organized Azure into a new Server & Cloud Division. So now the internal division that has most to lose if Azure succeeds is running the whole show. Technically that makes sense; but the marketing message is going to be even harder to articulate.

If you want to know how to estimate without using time units, here's how I do it:

For my work, I block out half-day chunks of time. I find that I can do some tasks in that half-day chunk (1, 2, or 3), and finish enough on any one thing that I can safely mark it as done enough and go on to the next task.

Some days, I have tons of little things to do: return calls, make lists, prepare to do something. Some days I have larger tasks: write a blog entry, draft a workshop, write a ton of email. Some days I have quite large tasks: draft 1 of a piece of a chapter, draft 1 of an article, draft 1 of a report.

Most weeks, I have a mix of things that are small, medium, and large. I no longer worry or estimate how long any of these things will take. I know about how many of the small, medium, and large tasks I can finish in a week. I can make my lists taking that into account.

If you are a developer or tester or BA or technical writer or a whatever (don't want to leave anyone out :-), you can do the same thing. Here's my recipe:

1. Make sure you've taken any task larger than a half-day chunk and broken it down into smaller tasks. That's why you see draft 1's for articles. I can write a draft in less than a half-day. I can't finish all the editing and have a great article at the end of a half-day. So, I don't try to do so.
2. Be clear on what done means for a particular task. If you don't know what done means, you can't get there. You'll either "finish" without getting to a reasonable stopping point and then you worry, or you can't finish in less than a half-day chunk.
3. Do just that one task until it's done. Commit to it. Finish it.
   

Now you have a little recipe for getting the tasks done. Here's how you apply it to your estimation. When someone asks how long something will take, you look at all your little tasks. Make sure they small, medium, or large, where large still fits in a half-day. You get no more than 10 half-days in a week. How many of these tasks can you do in a week? Add them up, and there you have your estimation without having to define how many hours a given task is.

If you're part of a project team, and want to try this on a larger scale, I particularly like the Fibonacci series (1, 2, 3, 5, 8, 13, 20) for estimating how large and complex a task is. As a team, use a Delphi approach (everyone gets together and as a group agrees on a number. For me, and when I teach my estimation workshop, any task larger than a 13 means "we have no clue and 20 is as good a number as any other number." When you hear someone is a specialist and can finish something earlier,  take the larger of the two values.

Other folks use t-shirt sizing: XS, S, M, L, XL, 2XL, 3XL. I don't like that as much, because, in my experience, teams don't learn to break tasks down enough, and have too many M, L, XL tasks. YMMV.

Now, it doesn't matter if you work in iterations or not. Take a week. As a group, ask "How many of these tasks can we do in a week?" Discuss the question. Arrive at a number you can all live with. You're done!

Remember that estimates are guesses. They are likely to be wrong, especially if you use time-based units. That's because we are either optimists or pessimists and because we never have ideal hours, days, or weeks. Try this and see what happens to your estimations.