By James Proctor, Co-Founder and Managing Director, The Inteq Group
Humans belong in an AI-enabled workflow wherever at least one of two signals fires: the data behind a case cannot support confident handling, or the decision the case requires carries consequence worth a person's deliberate judgment. Everywhere neither signal fires, human touch points add latency without adding control, and everywhere both fire, the process should deliver a person the case with full context and real time to think. That is the entire placement discipline in one paragraph. The rest is applying it honestly to workflows that were never designed this way.
What Do the Two Signals Actually Measure?
Confidence is a property of the case, not the model: given the completeness, consistency, and corroboration of the information behind this specific item, how safely does standard handling apply? Risk is a property of the decision: what does this case commit the organization to if handled wrongly, and how recoverable is the error? The signals are independent, which is what makes them useful. A case can be perfectly documented and still carry consequence that warrants judgment. A trivial case can arrive with data too thin to trust. Hierarchy-based review designs collapse both dimensions into one crude proxy, usually dollar value or reporting line, and that collapse is why traditional workflows put experts in the wrong places with such consistency.
What Does the Confidence-Risk Grid Produce?
Four handling patterns, each with a clear human role. High confidence, low risk: flow untouched, which should be the volume majority in most processes. Low confidence, low risk: the agent resolves what it can or applies a cheap human touch, because the cost of an error is bounded. High confidence, high risk: a person confirms efficiently, with the case's full support assembled, because consequence, not ambiguity, is the reason they are there. Low confidence, high risk: the full human judgment pattern, deliberate review with complete context, and the only quadrant where deep expertise is genuinely irreplaceable. The grid is simple. Its power is that it gives every human touch point a stated reason to exist, which is precisely what inherited review structures cannot do.
“Put judgment where the case needs it, not where the org chart left it.”
How Do You Migrate From Where People Sit Today?
Inventory the current touch points, classify a sample of real cases against the grid, and compare. One more discipline keeps the design honest over time: placement is versioned, not permanent. As evidence accumulates that a case class is handled well, its touch point can move, deliberately and with a named decision, toward lighter human involvement, and a class that starts producing surprises can move the other way. A placement map with a revision history is a governed design. A placement map nobody has touched since go-live is on its way to becoming the next generation's inherited structure.
Wealth management compliance review shows the typical migration finding at its sharpest. Registered principals in many firms review trade and account activity in near-uniform fashion: hundreds of items a day, seconds apiece, the same glance for a routine rebalance as for a concentrated position change in a retiree's account. These are among the most experienced, most expensive professionals in the firm, and the placement design has converted them into its most expensive rubber stamps. Regrid the same workload and the picture inverts: in-policy routine activity flows with monitoring, thin-documentation cases surface for completion before review, and the genuinely consequential patterns, liquidity events, vulnerable-client changes, concentration risk, receive minutes of principal attention instead of seconds. Nothing about the people changed. What changed is that the design finally routed judgment to where cases needed it.
Running the touch-point inventory and the regridding analysis is a standard component of our agentic AI consulting services, and the before-and-after placement map is usually the single most persuasive artifact in the entire redesign.
Placement is a design decision with a right answer per case class. Treat it as a preference, or a perk of seniority, and you will keep paying expert salaries for signature work.
