Escalation Protocol¶

Protocol — instantiates Principal–Agent Alignment

Defines the threshold at which an agent's ordinary discretion runs out and the decision must be routed up to a higher or different authority.

Delegation works only if the agent knows where their own authority stops. Escalation Protocol draws that stopping line as a trigger, not a map: it names the conditions — a size limit, a risk level, a class of exception, a novel situation the mandate never anticipated — under which the agent must hand the decision up rather than decide it alone. Its defining move is that it fires on an event. A decision that sat comfortably inside the agent's discretion a moment ago crosses a threshold and must now travel to a named higher, peer, or emergency authority, on a defined path, within a defined time. Where a static authority map says who normally decides what, the escalation protocol says when normal is no longer good enough and where the decision goes instead.

Example¶

A B2B software company lets each regional sales manager approve discounts on their own up to 15% off list. The escalation protocol governs everything past that line. A manager negotiating a large renewal wants to offer 28% to save the account. The protocol's threshold — any discount above 15%, or any term longer than three years — trips, and the decision is no longer the manager's to make. The protocol specifies the route: the request goes to the VP of Sales with a one-paragraph justification and the deal's margin math attached, and the VP has 48 hours to approve, counter, or decline before the quote expires.

The point is not to slow the manager down on ordinary deals — 90% of quotes never cross the line — but to guarantee that the unusual ones reach someone with the authority and the wider view to own them. When the 28% deal turns out to set a precedent three other customers later demand, the decision is already on the VP's desk where the consequences belong, not buried in a field manager's inbox.

How it works¶

The protocol is built from a trigger, a route, and a clock. The trigger is a bright-line condition an agent can recognize in the moment without judgment calls — a numeric limit, a named exception category, a checklist of "stop and ask" situations. The route names exactly who receives the escalated decision and in what form, so it lands on a specific desk rather than diffusing into a general appeal. The clock bounds how long the higher authority has to respond, and says what happens on silence (default-deny for risky actions, default-proceed for low-stakes ones). The discipline is keeping triggers observable and few: a protocol with fuzzy triggers gets ignored, and one with too many turns every decision into a referral.

Tuning parameters¶

Threshold height — how far ordinary discretion extends before escalation fires. Set it low and safety rises but the higher authority drowns in referrals; set it high and the agent is left owning decisions above their competence.
Trigger sharpness — bright numeric line versus judgment-based "when in doubt." Sharp triggers are reliably obeyed but miss novel risks that don't cross a number; soft triggers catch more but get rationalized away.
Route breadth — a single named authority versus a branching tree by exception type. More branches route each case to the right expert but add lookup friction under time pressure.
Response clock and default — how long the authority has, and whether silence means proceed or halt. A tight clock keeps the agent unblocked; a fail-safe default decides who eats the cost of no answer.

When it helps, and when it misleads¶

Its strength is that it lets a principal grant wide everyday autonomy and still catch the rare high-consequence decision, without pre-reviewing everything. It is the mechanism that makes "management by exception" actually work — routine flows untouched, and only the exceptions climb.^[1]

It misleads when the threshold is set for the principal's comfort rather than the real risk gradient, so the higher authority becomes a rubber-stamp bottleneck and the agent learns to structure decisions to stay just under the line — splitting one 28% deal into two 14% ones. It also fails silently when triggers depend on the agent recognizing they are in an escalation-worthy situation, which is exactly the recognition a stretched or motivated agent skips. The guard is to audit near-boundary behavior and after-the-fact for decisions that should have escalated but didn't, and to keep the trigger list short enough to be reflexive.

How it implements the components¶

escalation_threshold — its core output: the bright line at which the agent's ordinary discretion ends and a higher or different authority must decide.
information_reporting_channel — the escalation route is the interrupt channel by which an out-of-bounds situation, with its justification, reaches the principal in time to act.

It does not draw the standing map of who-decides-what — that is Decision-Rights Matrix; it does not carry routine status on a cadence — that is Reporting Requirement; and it sets no incentives, metrics, or goals.

Instantiates: Principal–Agent Alignment — supplies the boundary-sensitive control that fires when ordinary discretion is insufficient.
Consumes: Decision-Rights Matrix — the escalation trigger presupposes a defined discretion boundary to exceed.
Sibling mechanisms: Decision-Rights Matrix · Reporting Requirement · Governance Board · Audit or Review Cycle · Reputation System · Service-Level Agreement · Fiduciary Duty Rule · Incentive Compensation Plan · Performance Contract · Clawback Clause

References¶

[1] Management by exception — the practice of leaving routine work to the agent's own judgment and involving higher authority only when a variable strays outside a preset band. The escalation protocol is the concrete rule that defines that band and the response. ↩