Low Level Design Part 3: Estimates

You should include your task breakdown and estimates in the low level design in the appendix if it's available.

Design your project with the idea of disambiguating hard questions in mind. Frontload as much ambiguity in the project as possible. Changes in estimates months in advance are preferable to ones right before the deadline universally. Usually, the best way to do this is to write a thin vertical of your project first and then start fleshing out the functionality after that.

If you haven't worked with a library, platform, language, etc. your estimates on that topic will have low accuracy due to unknowns. You can reduce this effect with additional reading, especially of code, or by writing code using that library yourself. Call out everything that you don't know. Low accuracy estimates almost always take more time to complete rather than less. If you estimate a task as a week, the shortest it can be is an hour. The longest it can be is infinity. This asymmetry means estimates are usually close to the upper limit for how quickly something can be done.

Your estimates may appear unintuitive sometimes. It's likely that your gut feeling about an estimate comes from something. Explore that feeling and find the reasons why you think that task might take longer than it would seem. Incorporate this into the task description so it's intuitive. It's usually the case that a highly simplified explanation of project tasks will not include the relevant details that say why it takes as long as you think it is. It's very, very important to include these details.

Per the Mythical Man Month,every engineer has a base startup cost of reading documentation on the project and becoming familiar with it. If an engineer is only on the project for a single task, make sure they need to learn as little as possible about the project itself to complete their task by giving them interfaces, clear requirements and boxing their component. Be aware of how many engineers are on the project, how many libraries, platforms, etc. each will need to learn. Learning will take a considerable amount of time. For a single task, it will almost always take longer than implementation. Each additional person added to a project increases its communication-based costs.

Make sure to include features you would often take for granted if you are estimating a new project. If you are unsure, ask another engineer to check. Some examples for a service include: Auth, HTTPS/End to End TLS, Pipeline infrastructure, Monitors, Alarms, Logging, Dependency set up time, etc.

The most accurate estimation mechanism is comparison to previous projects. Find a previous project that has similarities for yours and see how long it took for each component. This will be helpful in estimating using real world data. This is especially value if this part of the project has high uncertainty.