Velocity Prediction Techniques
How do you predict velocity for roadmap planning when the team is new or changing? Executives want delivery dates but we have 3 sprints of history and just hired 2 people.
Use Monte Carlo simulation with your limited data. Take your 3 sprints of actuals, generate 1000 random samples assuming normal distribution, run simulations. Gives you probability ranges: "70% chance we complete in 8-10 sprints, 90% chance in 7-12 sprints".
Way better than single-point estimates with high uncertainty. I use SimulationCraft or just Python numpy.
Conservative approach: use your lowest historical velocity for predictions. If you've done 25, 30, 28 points, plan for 25. Under-promise, over-deliver. Accounts for team changes and unknown unknowns. Stakeholders hate it but you hit dates.
Account for ramp-up time explicitly. New team members are 50% productive first sprint, 75% second, 100% by third. Adjust capacity calculations accordingly. We model this in Excel and it's been scary accurate. Don't assume new devs = instant velocity boost.