AI Can Now Help Your Boss Decide What to Build First

Every company that builds software faces the same awkward weekly argument: which projects do we start first? Managers use gut feel, spreadsheets, past experience and, increasingly, AI predictions. Those sources often disagree. Two researchers, Azzeddine Ihsine and Sara Ihsine, published a paper describing how to combine all of them into one decision that stays visible and can be adjusted later — rather than a black box nobody can argue with.

Their system scores each candidate project on business value, effort, risk, and dependencies (meaning some tasks must finish before others can start). It then weights each estimate by how reliable that source has proven to be over time. In controlled simulated tests, this reliability-weighted approach reduced average effort-estimation error by 42.6% compared with the single best estimator, and beat a simple average of all opinions. Just as usefully, the method spotted which priority estimates were most likely to be wrong, catching the highest-error guesses about 90% of the time.

The researchers also tested how stable the rankings are. Small tweaks to how much a company values business value barely changed the results. But bigger strategic shifts — a sudden change in priorities — reshuffled the top 10% of projects noticeably. That's an honest warning: the math is only as steady as the goals you feed it.

The catch is that all of this happened in simulations, not real companies. The authors ran 800 artificial scenarios, which is a strong theoretical start but not proof it works with messy real-world teams, office politics and shifting deadlines. Treat it as a promising blueprint for making AI-assisted planning more transparent — not a finished product you can buy today.