Ask why an AI project stalls and the answer usually points at the model. Too many errors, not enough accuracy, the wrong data. Nitin Murali, supply-chain excellence chief at Gallo and founder of Category 2, argues the real fault sits elsewhere. Writing for the Forbes Technology Council, he says the problem is the org chart, not the technology.
The gap is simple to state. AI now makes decisions at scale, but the processes around it leave no named human accountable for them. The machine acts. Nobody owns the outcome. When something goes wrong, there is no one to explain what happened or why.
Murali identifies three roles that show up in every deployment that works, and that almost no one hires for.
The first is the Signal Architect. This person owns which signals the organisation sees, what they mean in context, and when the system's confidence is high enough to act on. They decide what counts as a real alert rather than noise.
The second is the Judgment Engineer. Their job is to design where humans step into an automated workflow, and which exceptions need a person rather than an automatic resolution. They draw the line between what the machine settles and what a human must weigh.
The third is the Decision Owner. This is the named person answerable for every consequential choice the system surfaces. They give a rationale, which makes the AI auditable. When a senior leader asks who decided, there is a name.
The fix Murali proposes is cheap. Pick one decision-heavy process. Map who currently owns each judgment in it. Often the honest answer is that nobody does. Then assign the three roles as hats people wear, before they harden into headcount you have to justify.
His before-and-after makes the point. Picture an AI that triages 340 exceptions overnight. Without the roles, they land on a planner as an undifferentiated pile, or get waved through on an acceptance rate that measures nothing useful. With the roles assigned, only 12 reach the planner, each with full context. She can override a buy recommendation because she knows it is tied to an upcoming packaging change the model cannot see. The manager's dashboard tracks decision quality, not how often the AI's advice was taken. When an SVP asks who decided, there is a name and a reason.
The lesson lands wider than supply chains. Most organisations are pouring money into models while leaving the accountability layer empty. They are automating decisions faster than they are deciding who owns them. The tools arrive first, the accountability never follows. That is how you end up with a system that is confident, fast and answerable to no one.
Murali's three roles are not a technology upgrade. They put a human back at the point where a choice is made. The cost is a few hours of mapping. The alternative is learning, after the fact, that no one was ever in charge.
