Last year, I worked on an inception trying to work out how big a three way product merge and rebuild would take. The business wanted to know what they could have by the start of summer this year.
During this intense four week inception we identified a huge number of stories – way more than I had ever identified in previous inceptions. Almost 500 storie by the end. I can’t recommend anyone going through this experience though we had drivers which meant we couldn’t avoid it this time.
My previous experience and gut feel tells me 100-ish stories (when I’m working with people to break it down) is probably enough work for a small dev team (3 dev pairs) for about 3 months. This was definitely a whopping year long programme of work (if done right).
We also had a lot of pressure to estimate them all. Up front. Obviously, attempting to estimate a year’s worth of work accurately is going to be pretty inaccurate. The longer the piece of work, the more assumptions will change, the more estimates made on those assumptions will be wrong. I know. However people still wanted numbers to understand how large this programme of work would be.
Some statistics
We ran incremental estimation sessions using relative story sizing, following fibonacci planning poker and estimating in points. Our maximum point size was 8 points. 5 was generally the highest though we tended to have 1 in 30 cards about this size.
We even iterated over a few estimates at random intervals to see if our relative sizing of stories changed significantly.
Interestingly enough, we stored some spreadsheets for various time during out estimation and I’ve pulled out some statistics from them, laid out in the table below:
| Spreadsheet Version | # Stories Identified | # Stories Estimated | Total Estimates in Points | Average Point / story |
|---|---|---|---|---|
| 0.22 | 135 | 129 | 340 | 2.63 |
| 0.26 | 529 | 395 | 1037 | 2.62 |
| 0.30 | 494 | 488 | 1346 | 2.75 |
What can we learn this from?
Firstly, one can see that the average story size isn’t significantly different over this large spread of stories. One could argue that given the dataset, it could be enough to extrapolate further estimates.
The next thing to consider is why do the numbers tend to average out? One could argue the story breakdown process for this project, leads to stories of the same size. It would be dangerous to assume all projects have similar story breakdown process.
Alternatively one could argue that the estimation process helped us breakdown stories to be approximately the same size. Nevertheless, an interesting observation and one I’ll continue to explore.
I’ve always suspected that the average will be slightly larger than midpoint of the range (slightly skewed from normal distribution).
What forces do you think make it skewed?
Stories tend not to take zero time
I didn’t think that would matter as much. I remember being told about people’s heights being bell curve distribution and that people do not have zero height either.