By James Proctor, Co-Founder and Managing Director, The Inteq Group
Identifying which processes to redesign for AI agents starts with a change in what you look for. Task-era opportunity hunting looked for effort: high volumes, repetitive steps, headcount concentrations. Agent-era opportunity hunting looks for judgment under strain: places where work queues in front of decisions, where exceptions concentrate, where a handful of experienced people are consulted constantly because only they can resolve what the process cannot. Read your operation for decisions rather than steps and the opportunity map redraws itself.
Why Won't Your Process Documentation Show You?
Here is the assertion that reliably annoys process teams, and I will stand behind it: for agent design purposes, your SOPs and swim-lane diagrams are close to useless, and treating them as the discovery source is how organizations automate the wrong things. Documentation describes the happy path, the portion of the work that never needed help. The judgment that actually runs the operation, what makes a case suspicious, when a rule should bend, who to call before committing, is almost never written down, because it was never a step. It lives in the heads of your most experienced people, exercised invisibly a hundred times a day. The process that creates your value is not the one in your documentation, and an agent program pointed at the documentation will faithfully transform the part of the operation that was already fine.
“The process that creates your value is not the one in your documentation.”
What Signals Reveal a Decision-Flow Opportunity?
Four signals do most of the work. Queues that form in front of people rather than systems, because a queue in front of a person is a decision waiting to happen. Exception rates, because exceptions are cases the designed path could not decide. The consultation pattern, the specific individuals everyone calls when something is off, because their phone traffic is a map of undocumented decisions. And context assembly, people gathering information from four systems before they can act, because that gathering is the tax a decision pays before it can be made. Where these signals cluster, you are looking at a process whose constraint is judgment, and that is agent territory.
Weight the signals by consequence, not just frequency. A decision made a thousand times a day matters, but so does one made weekly whose delay holds a seven-figure commitment hostage. The strongest candidates sit where the signals converge: heavy consultation, high exception rates, and expensive waiting in the same process.
What Questions Replace the Old Discovery Questions?
Classic process discovery asks what happens next. Decision-centric discovery asks what are you deciding, what do you need to know to decide it, what happens when you cannot, and who gets involved then. Law firm client intake shows how different the answers are. The documented procedure is tidy: run the conflicts search, circulate results, obtain sign-offs. The real process is dense with judgment the procedure never mentions: whether a corporate family relationship is a business conflict rather than a legal one, how a lateral partner's history changes exposure, when a waiver is worth requesting and from whom. Ask what-happens-next and you will automate search and routing, shaving hours off a cycle whose real cost is elsewhere. Ask what-are-you-deciding and the redesign target becomes the days of partner deliberation, which is where matters are actually won, lost, and delayed.
This way of reading an operation is a learnable analytical discipline, not a talent. Training business analysts and process owners in decision-centric discovery is one of the core aims of our agentic AI training courses, and analysts tell us it permanently changes what they see when they walk a process.
Steps tell you where the effort went. Decisions tell you where the value waits. Learn to read for the second, and your agent roadmap will stop resembling your old automation backlog.
