3 ms·

> Stop estimating. I have analysed every team I have been on that use estimation. Those teams have been 99% incorrect. In my experience, it does not work. If yo

> Stop estimating. I have analysed every team I have been on that use estimation. Those teams have been 99% incorrect. In my experience, it does not work. If you need dates, I would recommend a more modern approach like forecasting. What does this even mean? I've never heard of a forecast that didn't use a combination of past data and an assessment of complexity of future work to assemble the forecast. And assessment of complexity is the heart of estimation. Can anyone help me because my imagination is failing me here.

Ensorceled 4y ago
The problem is that article is confusing the management failure of assuming estimates should be accurate and then punishing/rewarding based on accuracy of estimates with the proper practice of using estimates for scoping projects so they can properly prioritized and loosely road-mapped. It's a throw the baby out with the bathwater problem.
drdec 4y ago
I think the difference is that a forecast is developed by a team (or formula/ML) separate from the engineers doing the work.
- BaseballPhysics 4y ago
  Once again: no team or model or formula can produce a forecast without an assessment of complexity for the work effort being forecast. So if what you're saying is true, this would involve someone other than the devs assessing complexity, then coming up with a date, and then holding the devs accountable. And that sounds like a complete nightmare that's far worse.
  - drdec 4y ago
    > without an assessment of complexity The point of ML is that that can happen without human intervention. And yes, I agree that if devs are held accountable to such timelines that would be terrible. But if they are used for business planning with the understanding that they will not be accurate in the specific, but hopefully in the aggregate, then that's fine. (This is how engineer's estimates should be used as well for that matter.)
    - BaseballPhysics 4y ago
      > The point of ML is that that can happen without human intervention. What kind of wizardry is this? What are the inputs to this ML? The vague requirements supplied by the customer and filtered through a sales rep? Please. ML isn't magic. Hell, what would this magical model even be trained on??

Ensorceled 4y ago

The problem is that article is confusing the management failure of assuming estimates should be accurate and then punishing/rewarding based on accuracy of estimates with the proper practice of using estimates for scoping projects so they can properly prioritized and loosely road-mapped. It's a throw the baby out with the bathwater problem.

drdec 4y ago

I think the difference is that a forecast is developed by a team (or formula/ML) separate from the engineers doing the work.

BaseballPhysics 4y ago
Once again: no team or model or formula can produce a forecast without an assessment of complexity for the work effort being forecast. So if what you're saying is true, this would involve someone other than the devs assessing complexity, then coming up with a date, and then holding the devs accountable. And that sounds like a complete nightmare that's far worse.
- drdec 4y ago
  > without an assessment of complexity The point of ML is that that can happen without human intervention. And yes, I agree that if devs are held accountable to such timelines that would be terrible. But if they are used for business planning with the understanding that they will not be accurate in the specific, but hopefully in the aggregate, then that's fine. (This is how engineer's estimates should be used as well for that matter.)
  - BaseballPhysics 4y ago
    > The point of ML is that that can happen without human intervention. What kind of wizardry is this? What are the inputs to this ML? The vague requirements supplied by the customer and filtered through a sales rep? Please. ML isn't magic. Hell, what would this magical model even be trained on??