AI-Based Estimation: The Future?

The Promise

Feed your Issue Tracker history and Git logs into an AI. It analyzes your "Cycle Time" distribution and predicts: "Feature X has a 85% probability of shipping in 14 days."

The Reality

AI predicts the past.

• If your team has stable patterns (e.g., maintenance mode), AI estimation is terrifyingly accurate.
• If you are building something new (Innovation), AI is useless. It has no training data for "Unknown Unknowns".

Tools in the Market

1. LinearB / Haystack: Analyze cycle time and PR throughput.
2. Issue Tracker Intelligence: Summarizes tickets but rarely predicts dates accurately.
3. 7 Geese / Monte Carlo: Statistical simulation (not really AI, but math) is often better than AI.

The Consultant's Take

Use Monte Carlo Simulation, not "AI".

• Don't ask ChatGPT "How long will this take?".
• Run a simulation: "Based on our last 50 tickets, what is the P95 delivery time?"
• Math > Magic.