Hi Charles. I use a spreadsheet to record throughput of small stories each e.g. week, and calculate the STDDEV. I then add this to the average throughput to create an optimistic future throughput, and subtract it to create a pessimistic future throughput. These form the outer ranges of the “cone” in the graph.

Example 1:

7 dev weeks complete, story throughput = 4, 4, 5, 4, 5, 4, 4
Avg throughput = 4.3, Std dev = 0.5
Forecast throughput for weeks 8+ (using 1 STDDEV) = 3.8 to 4.8 stories per sprint
Forecast throughput for weeks 8+ (using 2 STDDEV for wider range representation of uncertainty) = 3.3 to 5.3 stories per sprint

[Forecast Pess throughput (1 STDDEV) = 4.3 — 0.5 = 3.8
Forecast Pess throughput (2 STDDEV) = 4.3–1.0 = 3.3
Forecast Opt throughput (1 STDDEV) = 4.3 + 0.5 =4.8
Forecast Opt throughput (2 STDDEV) = 4.3 + 1.0 = 5.3]

Example 2 (same average throughput but greater variance/volatility — here I’m highlighting the importance of this approach over only using averages):

Throughput: 4, 8, 2, 3, 5, 4, 4
Avg throughput = 4.3, Std dev = 1.9
Pess throughput = 4.3 — 1.9 = 2.4
Opt throughput = 4.3 + 1.9 = 6.2

You can also go lo-fi (as Henrik Kniberg suggests in his seminal “PO in a nutshell” video) and simply draw lines at the outer ranges of the cumulative throughput line to create the cone, without needing to calculate the STDDEV (or “statistical hoodoo” as Henrik calls it).

I call the forecasts “optimistic” and “pessimistic” because they represent things going “well” or “not well”. The wider we make the range, the more well or not well we would need things to go in order for that particular future to pan out. This creates greater “accuracy” in the forecasts (i.e. we are less likely to be wrong about what happens), but sacrifices “precision” (i.e. the range may be too wide to be useful for decision-making or communicating milestones with customers).

You need to find what works in your context to get the right balance of precision and accuracy, as well as having transparent conversations about uncertainty and reliance on being predictive (make decisions based on what we think might happen) over being empirical (make decisions based on current state of artefacts and “yesterday’s weather”).

I hope this answers your questions?