In the first post of this series, I told the backstory of an ill-fated project estimation effort that cost us the trust of an important client. What went wrong? Why was our estimate so bad?
Our estimate was wrong because we relied on individual judgement—when individual judgement should have been our last resort.
Individual Judgement relies heavily on intuition, and intuition simplifies how we understand experiences in ways that support coarse comparisons (is X larger than Y?) but limit the detailed comparisons we need for good estimation (how much larger is X than Y?). This simplification can also suppress details from our experience in unpredictable ways for different people, meaning we shouldn’t expect combinations of judgement-based estimation errors to cancel out.
And it is why, for example, the first and second estimates in our story didn’t relate sensibly with the costs of developing the proof-of-concept. The estimates did not take into consideration the proof-of-concept development costs, even though it would have been relatively easy. They could have, and they should have, but they didn’t.
Life Gets a Vote
Beyond individual judgement, we didn’t consider the probability of our estimates coming true. Our estimates may be the best we can make them, but they won’t predict every outcome exactly—they can’t. When we estimate, we have no fool-proof insight into the day-to-day decisions and circumstances that will define how long work will actually take.
A clue that probabilities were not considered in our story was the task estimates recorded and combined as single numbers (in our case, hours). No ranges, percent confidence, or agreement on what these estimates were supposed to represent. This made our combined estimates arguably useless, since they represented an unknown mix of individual probability assumptions that couldn’t be reliably combined.
Through Mud-Colored Glasses
Lastly, we let experience distort our thinking. Recalled experience doesn’t represent everything that really happened or could happen. This limitation will typically lead us to believe good outcomes come more from good decisions than circumstance and vice-versa, but the reality is that most outcomes are traceable to both. And decisions and circumstances will not remain consistent over time.
This can be critically important when considering especially recent, memorable, or painful outcomes, and was probably a factor when we made both estimates in our story. In our first tally, for example, optimistic impressions of the proof-of-concept work led to a way-too-rosy estimate. Once the work proved harder than anticipated, however, the team expected every task to come with time-consuming technical problems and interactions.
Similar to the probability assumptions, the team didn’t agree on the significance of recent experience. This meant task estimates were bound to vary more across the team than they should, since they represented another unknown mix of assumptions—assumptions made particularly negative and extreme by the time we worked on the second estimate.
Insult to Injury
In summary, our judgement-based task estimates were low and incomplete, our combined estimates represented conflicting assumptions, and our biases further distorted the results. When we consider these influences together, it’s understandable that both our estimates were unrealistically low.
In the next post, I’ll lay out what we’ve learned to do differently.
(Please see the third post in this series here.)