Skip to content
Graph of Plan

Software Project Prediction Is Possible

Software project delivery timelines can be predicted with useful accuracy (within 30% of actual) by AI analysis of project data.

The Assumption

Murphy’s entire value proposition is predicting when software projects will deliver. But is this even possible?

The hard truth: “Why is software always late?” has been asked for 50 years. No tool has solved it. Maybe it’s unsolvable:

  • Software projects are complex adaptive systems
  • Requirements change mid-project
  • Dependencies are hidden until they bite
  • Human factors dominate technical factors

If prediction isn’t possible, Murphy is selling snake oil.

Evidence

Supporting signals:

  • Some patterns are predictable (velocity trends, scope creep signals)
  • AI can process more data than humans
  • Monte Carlo simulations provide probabilistic forecasts
  • Early warning is valuable even if not precise

Counter-signals:

  • 50 years of failed prediction tools
  • Fundamental uncertainty in creative work
  • Garbage in, garbage out (project data is messy)
  • Goodhart’s Law: measured metrics get gamed

What Would Prove This Wrong

  • Predictions consistently off by over 50%
  • No better accuracy than naive estimates (e.g., “double the estimate”)
  • False positives cause alarm fatigue
  • Agencies don’t trust the predictions

Impact If Wrong

If prediction isn’t possible:

  • Murphy fails regardless of execution
  • Pivot to different value prop (project visibility, not prediction)
  • Or abandon Murphy entirely
  • Agency expertise becomes less valuable

Testing Plan

Technical validation:

  • Build prediction model on historical data
  • Backtest against known outcomes
  • Measure accuracy: % of predictions within 30% of actual

Customer validation:

  • Are predictions more useful than gut feel?
  • Do early warnings provide actionable lead time?
  • Do agencies change behavior based on predictions?

Kill criteria: If predictions aren’t better than “multiply PM estimate by 1.5”, pivot the value prop.

Depends on:

Affects:

  • Murphy — entire product viability

Assumption

Software project delivery timelines can be predicted with useful accuracy (within 30% of actual) by AI analysis of project data.

Depends On

This assumption only matters if these are true:

How To Test

Build prediction model. Test on historical project data. Measure accuracy in real projects.

Validation Criteria

This assumption is validated if:

  • Predictions within 30% of actual delivery date 70% of the time
  • Predictions more accurate than PM gut feel
  • Early warning of delays provides actionable lead time

Invalidation Criteria

This assumption is invalidated if:

  • Predictions consistently off by over 50%
  • No better than naive estimates
  • False positives cause alarm fatigue

Dependent Products

If this assumption is wrong, these products are affected:

Dependent Milestones

If this assumption is wrong, these milestones are affected: