I read an article in the Wall Street Journal a couple of months ago that resonated strongly with me. Although the article deals with the problem of project planning for infrastructure projects, it’s really about irrational optimism. Why is it that infrastructure projects always cost more and take longer than expected? (As anyone reading this post knows, this problem isn’t limited to infrastructure projects!)
Here is one of the insights:
The pattern of budget excess echoes findings in other areas where people allow their best hopes to dominate the planning process. Irrational optimism afflicts even individuals who have experience with a given situation, and should know better. Researchers have linked it to business analysts’ earnings forecasts and to students predicting when they will finish assignments?
Magne Jørgensen, who has studied persistent underestimates of completion times in software development, says businesses would do well to maintain data on prior forecasts. “Looking forward makes you more optimistic,” says Prof. Jørgensen. “Looking backward makes you more realistic.” (italics mine)
Cardinal Peak has always strived to make realistic estimates for its customers, and over time we have become better at doing so. Mike Deeds and I, who often end up estimating projects together, have found a technique that seems to work pretty well. We start with a bottom-up estimate: We create a list of the tasks that need to be performed; we estimate the time required to complete each task, and we multiply the aggregate time by 1.2.
The “1.2 factor” helps to account for two things. First, the need for interaction among developers to complete a task — after all, system components, whether hardware or software, never exist in isolation, and interaction takes time. Second, the fact that developers often encounter unforeseen challenges — things that are hard to know when planning and are only learned when you sit down and start actual design and coding.
Next, we step back and ask ourselves which project that we’ve done in the past is most similar to this one. Once we’ve identified it, we ask ourselves how many engineers were working on that project and how long it took them. We make any obvious adjustments to account for the differences between the projects, and if after that, the bottom-up and top-down estimates are not within 10% of each other, we dig in and ask why. Have we missed some obvious tasks (e.g., debugging time)? Have we forgotten how long it takes to do QA and fix bugs in the real world? Have we underestimated how long it will take to develop manufacturing tools and support a handoff?
Unfortunately, convincing customers that our estimates are truly realistic is sometimes challenging, especially when the estimates are higher than a customer had hoped. We understand this and do our best to help them understand that our estimates seem “high” because they are realistic and were generated by experienced engineers. The WSJ article notes that:
Researchers say another factor exacerbates the effects of irrational optimism: Projects planned most optimistically will look more attractive to funders than those that are planned realistically. So even if not every planner is optimistic, those who are have the best shot of moving forward. “Optimism is tremendously inspiring,” says Dale Griffin, professor of marketing. “Accuracy, perhaps not so much.”
Over time, we’ve also learned that clients remember the lowest number that was ever placed in front of them, regardless of how early in the planning process it was generated and how many feature requests and changes they have made since then! The best way we’ve found to deal with this is to resist the temptation to throw out a number before the requirements are understood and to insist on a solid requirements document and payment for the time needed to generate a plan. Generating a good plan is a lot of work; real engineering occurs during the plan generation process.
Sometimes, when the pressure to get overly optimistic feels particularly intense, we have to remind ourselves that those who make overly optimistic estimates frequently end up suffering from what the WSJ calls “the consequence of rosy projections.” We’ve been there too, and it isn’t pretty.
Mike Perkins, Ph.D., is a managing partner of Cardinal Peak and an expert in algorithm development for video and signal processing applications.