By James Proctor, Co-Founder and Managing Director, The Inteq Group
A decision inventory is a structured catalog of the decisions that drive a critical business process: what each decision is, the outcome it serves, who owns it, what information it genuinely requires, and where the boundaries of authority sit. Built before any technology is selected, it becomes the blueprint for an agentic AI program, showing where agents create value, where they create risk, and where design attention must concentrate first. It is the least glamorous artifact in the entire AI conversation, and in my experience it is the one that most reliably separates programs that compound from programs that stall.
What Actually Goes Into a Decision Inventory?
Five fields carry most of the weight. The decision itself, stated as a genuine choice rather than a task. The outcome it exists to serve, which forces clarity about why the decision matters. The current owner, meaning the person or role actually accountable, not the box the org chart implies. The information the decision requires, listed against where that information actually lives. And the authority boundary: what may be committed at this decision point and what must move up. Most critical processes resolve to somewhere between eight and twenty decisions once you strip out the tasks masquerading as decisions.
The inventory is rarely long. It is always revealing.
“The inventory is rarely long. It is always revealing.”
How Do You Build One?
Not from documentation, and not in a conference room. You trace real cases. Take a sample of recent items through the process end to end and interrogate every point where the case could have gone more than one way: who chose, on what basis, with what information in front of them, and under what authority. Then reconcile what the cases show against what the process owner believed, because the two will differ. The work is measured in weeks, not quarters, and it requires disciplined analysis rather than committees. What it cannot survive is delegation to people too junior to ask who really owns this and get an honest answer.
Two disciplines keep the inventory honest. First, reject tasks masquerading as decisions: 'review the application' is not a decision, while 'determine whether this application proceeds without additional verification' is, and the difference is whether the entry names a genuine choice with consequences. Second, treat the inventory as a governed, living artifact rather than a workshop souvenir. Decisions change owners, authority moves, and an inventory nobody maintains quietly becomes fiction of a second kind.
What Does the Inventory Reveal?
Here is the finding I can nearly guarantee, and it is the part organizations find uncomfortable: your org chart is a work of fiction about decision rights. On paper, every decision has a home. Traced through real cases, a substantial share of consequential decisions have no explicit owner at all. They are made by whoever touches the case at the moment the choice becomes unavoidable, under authority nobody ever granted.
Medical device complaint handling makes the stakes concrete. Between intake and closure sit heavily consequential decisions: whether an event is reportable to regulators, how deep an investigation must go, whether a pattern warrants corrective action. Regulation assumes each has a clear owner. Inventory the process at a typical manufacturer and you find the reportability determination, a decision with direct regulatory exposure, is shaped in practice by whoever triages the intake queue that day, exercising judgment the quality system never assigned to them. No one decided that. It accumulated. Handing an agent a process in that condition does not create the governance gap. It inherits one that was already there, and makes it run faster.
Constructing the decision inventory, and reconciling it against real case behavior, is foundational work inside our agentic AI consulting services. It is also the single cheapest risk reduction available to an AI program, because every downstream design choice stands on it.
Before you evaluate a single platform, know your decisions. Everything else in agent-enabled design is built on that catalog, or built on sand.
