A shorter route can be a worse day
Imagine a maintenance schedule with fewer miles and a missed appointment. On one chart it looks like progress. At the customer's front door, perhaps less so. Before asking AI to optimize an operation, we would ask the people doing it what better actually means. That conversation can be awkward. It is also cheaper than building a system around the wrong answer.
The research has its own setting
MIT's October 2 profile of Cathy Wu describes reinforcement-learning research on complex systems, including transportation. It revisits a 2023 result improving training efficiency across related problems. This is historical research context, not a new promise of savings for a business reading the article today.
What interests us is the care needed to define the problem. A business does not become ready for a learning system because the research is impressive. It becomes ready for an experiment when the team can explain a decision, the constraints around it and the evidence that would make a proposed change worthwhile.
Source: MIT News: Computational tools for society's most complex challenges · 2026-10-02
Listen to the person who fixes the schedule
In our hypothetical maintenance company, the dispatcher knows that one technician has a specialist skill, another appointment has a firm deadline and urgent calls need breathing room. Some of that knowledge may never reach the spreadsheet. Ask what gets changed by hand and why. Those corrections are clues to the real job.
Then look at approved records of the current process. A visit marked complete may hide a late arrival or an extra journey. Compare a candidate approach with an ordinary baseline that staff recognize. Include a busy day and a missing detail. Let the dispatcher challenge the assumptions before anyone mistakes a neat recommendation for a workable schedule.
A useful result might be modest. It might show that a simple rule helps, that the data needs work or that the existing method is doing surprisingly well. All three answers are worth having. Our takeaway: spend the first meeting agreeing what better means and whose day should improve. Choose the method after that. The research can widen your imagination; the operating question gives it somewhere useful to go.