際際滷

際際滷Share a Scribd company logo

Forecasting for beginners

Download as PPTX, PDF
Dan Brown
Dan Brown

This document provides an introduction to forecasting and predictive modeling using cumulative flow diagrams (CFDs). It discusses how to gather initial data by breaking down work into epics and stories and tracking lead times. With enough data points, a CFD can be generated showing work in progress over time. Frequency charts of lead times can then help set expectations. The document encourages playing with a sample Excel workbook to generate one's own CFD and charts for forecasting. It argues that even a small amount of data is better than relying solely on estimates to plan work.

1 of 37
Downloaded 23 times
Forecasting For Beginners Dan Brown
About Me Dan Brown Kanban Coach & Teacher KanbanDan@gmail.com @KanbanDan
We often talk about estimates as if they are something meaningful We normally mean forecast when we say estimate Forecasts arent guesses
Maths is our friend (honest) Maths often gets a bad rep We often think of complex things we had to learn from first principles at school But you dont need to understand the inner workings of an engine to be able to drive a car Its a good idea to have an expert on hand when the engine needs fixing Solve x and y where: y = x2 - 5x + 7 y = 2x + 1
The problem with Disneyworld Whats the downside of Disneyworld for guests? Avoiding the queues Theres an App for that It uses historic data and the Travelling salesman problem The right maths in the right place Longest Disney queue Ive stood in was 15 minutes
Gameshows? The Monty Hall Problem drives maths undergrads mad 3 doors, one prize What are the odds you pick the right door? 33% What if Monte removes a losing door? What are your odds now? Should you change your choice?
Agreeing terms PRODUCT A service that makes sense to a customer EPIC a big story. Too big for a team to finish in a fortnight STORY a single unit of work that finishes in between 2 and 9 days
Lets think in Asteroids Epics Products Stories
Ready Player 1 What happens if we shoot an Product sized asteroid?
Shoot a Product Slow Products break into 3 medium paced Epics
Shoot an Epic Medium paced Epics break into 4 fast paced Stories
How did this strategy work out? Oops!
Then this usually happens We get an Expedite work item to deal with!
Back to sanity We could finish epic 1s 4 stories, then the next epic of product 1 That way we always finish something valuable rather than showing progress on lots of things
Conclusion (The answers near the front of the book)
Step one - Workshop Run a workshop to break down your initial product into epics
Step two Break down the first 5 epics into stories Count the stories in each epic Ignore the middle 3 numbers Assume the Biggest and Smallest represent the range Assume the mid point of the range is the median number of stories per epic X X X X XX Fewest stories Most stories
Step three get to work! Measure the Lead Time to complete each of the first 11 stories. Initial data gathering is done! You can also use the 85th Percentile as your story SLA KEEP CALM AND START WORK!
Graph time You now have enough data to draw a Cumulative Flow Diagram (CFD). Number of stories on Y axis against date on the X axis Shows To Do, Doing and, Done Plot a cone of certainty using 15th and 85th Percentiles
To Do Done Doing X
CFD Forecasting Key Points Always use ranges, not individual dates Make it visible Teach people how to read it The truth is the truth. This makes it visible, undeniable and non-negotiable Moves the conversation on to business decisions This is real data from a real development team
Frequency chart 85th %ile Lead time frequency chart will show YOUR Weibull distribution Use this to help decide when to start time bound stories
Where do I start Go to github.com/kanbandan Click on PredictiveCFD Download the Excel workbook Make yourself a new copy and open the workbook You need to play with 2 sheets Setup On The Board
Setup sheet I used the standard Excel formatting for Input cells You can only change the salmon coloured cells Blank out the two dates hereSet this date to the first date of your delivery Set this dropdown to 11 Set to your work item types
On the Board This is all of the data for the sample sheet
On the Board Clear it off and start adding your stories No gaps in dates entered My favourite cheat formula =IF(ISNUMBER([@[Ready For Demo]]),[@[Ready For Demo]],"") (If the cell to my right is a number, show it here too. If not show a blank cell here) Lets you skip columns you dont want to use Remember weighting of 1
And thats it You can now look back in wonder at your wonderful Cumulative Flow Diagram Lead Time Frequency Chart
Why it all works Explaining the magic numbers (just in case you don't trust me)
Let's talk WWII tanks The Panzer V was a big heavy tank. It had better armour, range and accuracy than the Sherman. The Allies needed to know how many were in France to plan D-Day
How many tanks? Eisenhower asked both Military Intelligence and the Bletchley Park Boffins to work on it This is known as "The German Tank Problem" MI BPB June 1940 1000 169 June 1941 1550 244 Aug 1942 1550 327 Real 122 271 342
Maths beats estimates So do we need to do lots of maths? Good news - you don't. There IS a formula, but I'm not going to bother you with it today.
The answers With 5 samples you are 12.5% likely to find a bigger value and 12.5% likely to find a smaller value than your existing range. 75% chance within range With 11 samples you make that 90% chance inside range, 5% above and 5% below.
Putting it to use It works for: tank gearbox serial numbers story sizes or even dating partners
Why not just estimate? How do you weigh something big on bathroom scales? Cut it up and weigh all the small parts? The problem is the tolerance cumulates and makes the measurement so inaccurate its useless 200 days 賊 120 days isnt much use to us
Should we stop estimating? Estimates are useless, estimation is essential The benefit of whole team estimation is the sharing of tacit knowledge, just before working on the thing were talking about. It deliberately introduces conflict No groupthink
Getting started all you need is: a date stamp (or a pen) a spreadsheet (or some graph paper) https://github.com/kanbandan/PredictiveCFD
Forecasting for beginners

More Related Content

Forecasting for beginners

  • 2. About Me Dan Brown Kanban Coach & Teacher KanbanDan@gmail.com @KanbanDan
  • 3. We often talk about estimates as if they are something meaningful We normally mean forecast when we say estimate Forecasts arent guesses
  • 4. Maths is our friend (honest) Maths often gets a bad rep We often think of complex things we had to learn from first principles at school But you dont need to understand the inner workings of an engine to be able to drive a car Its a good idea to have an expert on hand when the engine needs fixing Solve x and y where: y = x2 - 5x + 7 y = 2x + 1
  • 5. The problem with Disneyworld Whats the downside of Disneyworld for guests? Avoiding the queues Theres an App for that It uses historic data and the Travelling salesman problem The right maths in the right place Longest Disney queue Ive stood in was 15 minutes
  • 6. Gameshows? The Monty Hall Problem drives maths undergrads mad 3 doors, one prize What are the odds you pick the right door? 33% What if Monte removes a losing door? What are your odds now? Should you change your choice?
  • 7. Agreeing terms PRODUCT A service that makes sense to a customer EPIC a big story. Too big for a team to finish in a fortnight STORY a single unit of work that finishes in between 2 and 9 days
  • 8. Lets think in Asteroids Epics Products Stories
  • 9. Ready Player 1 What happens if we shoot an Product sized asteroid?
  • 10. Shoot a Product Slow Products break into 3 medium paced Epics
  • 11. Shoot an Epic Medium paced Epics break into 4 fast paced Stories
  • 12. How did this strategy work out? Oops!
  • 13. Then this usually happens We get an Expedite work item to deal with!
  • 14. Back to sanity We could finish epic 1s 4 stories, then the next epic of product 1 That way we always finish something valuable rather than showing progress on lots of things
  • 15. Conclusion (The answers near the front of the book)
  • 16. Step one - Workshop Run a workshop to break down your initial product into epics
  • 17. Step two Break down the first 5 epics into stories Count the stories in each epic Ignore the middle 3 numbers Assume the Biggest and Smallest represent the range Assume the mid point of the range is the median number of stories per epic X X X X XX Fewest stories Most stories
  • 18. Step three get to work! Measure the Lead Time to complete each of the first 11 stories. Initial data gathering is done! You can also use the 85th Percentile as your story SLA KEEP CALM AND START WORK!
  • 19. Graph time You now have enough data to draw a Cumulative Flow Diagram (CFD). Number of stories on Y axis against date on the X axis Shows To Do, Doing and, Done Plot a cone of certainty using 15th and 85th Percentiles
  • 21. CFD Forecasting Key Points Always use ranges, not individual dates Make it visible Teach people how to read it The truth is the truth. This makes it visible, undeniable and non-negotiable Moves the conversation on to business decisions This is real data from a real development team
  • 22. Frequency chart 85th %ile Lead time frequency chart will show YOUR Weibull distribution Use this to help decide when to start time bound stories
  • 23. Where do I start Go to github.com/kanbandan Click on PredictiveCFD Download the Excel workbook Make yourself a new copy and open the workbook You need to play with 2 sheets Setup On The Board
  • 24. Setup sheet I used the standard Excel formatting for Input cells You can only change the salmon coloured cells Blank out the two dates hereSet this date to the first date of your delivery Set this dropdown to 11 Set to your work item types
  • 25. On the Board This is all of the data for the sample sheet
  • 26. On the Board Clear it off and start adding your stories No gaps in dates entered My favourite cheat formula =IF(ISNUMBER([@[Ready For Demo]]),[@[Ready For Demo]],"") (If the cell to my right is a number, show it here too. If not show a blank cell here) Lets you skip columns you dont want to use Remember weighting of 1
  • 27. And thats it You can now look back in wonder at your wonderful Cumulative Flow Diagram Lead Time Frequency Chart
  • 28. Why it all works Explaining the magic numbers (just in case you don't trust me)
  • 29. Let's talk WWII tanks The Panzer V was a big heavy tank. It had better armour, range and accuracy than the Sherman. The Allies needed to know how many were in France to plan D-Day
  • 30. How many tanks? Eisenhower asked both Military Intelligence and the Bletchley Park Boffins to work on it This is known as "The German Tank Problem" MI BPB June 1940 1000 169 June 1941 1550 244 Aug 1942 1550 327 Real 122 271 342
  • 31. Maths beats estimates So do we need to do lots of maths? Good news - you don't. There IS a formula, but I'm not going to bother you with it today.
  • 32. The answers With 5 samples you are 12.5% likely to find a bigger value and 12.5% likely to find a smaller value than your existing range. 75% chance within range With 11 samples you make that 90% chance inside range, 5% above and 5% below.
  • 33. Putting it to use It works for: tank gearbox serial numbers story sizes or even dating partners
  • 34. Why not just estimate? How do you weigh something big on bathroom scales? Cut it up and weigh all the small parts? The problem is the tolerance cumulates and makes the measurement so inaccurate its useless 200 days 賊 120 days isnt much use to us
  • 35. Should we stop estimating? Estimates are useless, estimation is essential The benefit of whole team estimation is the sharing of tacit knowledge, just before working on the thing were talking about. It deliberately introduces conflict No groupthink
  • 36. Getting started all you need is: a date stamp (or a pen) a spreadsheet (or some graph paper) https://github.com/kanbandan/PredictiveCFD