How to Use AI to Improve Agile Estimating
Artificial intelligence is making its way into every area of software development — and agile estimation is one of the most promising frontiers.
What AI Already Does Today
Automated Estimates Based on History
Tools like ActionableAgile and Forecast analyze historical sprint data and suggest automatic estimates:
“Out of the 50 tasks similar to this one your team has completed, the average was 5.2 points with a standard deviation of 1.8. Suggestion: 5 points.”
Complexity Classification
AI can analyze a user story description and classify its complexity:
“This story mentions ‘third-party API integration’ and ‘complex validation’. Based on the team’s history, this tends to be 8-13 points.”
Risk Detection
By analyzing patterns in past estimates versus actual results:
“Your team consistently underestimates tasks involving ‘legacy database’ by 40%. Consider applying a correction factor.”
Release Forecasting
Predictive models calculate completion probability:
“At current velocity with 150 points remaining, there is an 80% chance of completion in 5-7 sprints and a 50% chance in 6 sprints.”
How AI Complements (Not Replaces) Planning Poker
AI as a Reference Point
Before Planning Poker, AI can show:
“Historical estimates for similar items: 3-8 points. Average: 5.”
The team then votes with that reference, avoiding completely baseless guesses.
AI as a Post-Vote Validator
After voting, AI can flag:
“You estimated 3 points. For items with these characteristics, the historical average was 7. Consider whether something is being underestimated.”
AI as a Continuous Calibrator
AI automatically adjusts recommendations as the team calibrates:
“Over the last 2 sprints, the team’s accuracy improved 15%. Recommendations now carry more weight.”
What AI Does NOT (Yet) Do Well
Business Context
AI does not know that “this client is strategic” or “this deadline is political.” These factors shift priorities but are invisible to algorithms.
Undocumented Complexity
AI analyzes text, not code. If the complexity lives in an undocumented legacy module, AI will miss it.
Team Dynamics
AI cannot detect that “the developer who knows this module is leaving” or “we have 2 new members still calibrating.”
Scope Negotiation
AI can say “that takes 5 sprints.” But it does not suggest “if we cut X, it takes 3.”
Emerging Tools
Jira with AI (Atlassian Intelligence)
- Suggests estimates based on history
- Groups similar tickets together
- Detects delay patterns
Third-Party Plugins
- Jira Assist: predictive estimation
- Corrello: user story quality analysis
- Trendlyzer: release forecasting with confidence intervals
Standalone Solutions
- LinearB: engineering metrics with AI-powered insights
- Pluralsight Flow: workflow analysis
- Axify: dashboards with predictive forecasting
The Future: AI as an Estimation Co-Pilot
In 1-3 years, it is likely that:
- AI will suggest estimates automatically when a story is created
- Planning Poker will incorporate AI as a “reference voice” during voting
- Releases will be planned by AI with 90% confidence intervals
- Risks will be identified proactively before the estimation session even starts
But human judgment remains essential. AI suggests, the team decides.
Conclusion
AI in agile estimation is a complement, not a replacement. It brings data and pattern recognition that humans easily miss — but it does not replace the conversations that divergent estimates trigger, business context, or the collective judgment of Planning Poker. Use AI as a consultant, not as the decision-maker.