Most escalations are permission, not knowledge
When a team looks at why conversations get escalated, the assumption is usually that agents
did not know the answer. Read a sample of escalated threads and a different picture tends to
appear: the agent knew exactly what should happen and was not allowed to do it.
That distinction matters because the two have opposite fixes. A knowledge gap is solved with
documentation and training. An authority gap is solved by changing what agents are permitted to
decide, and no amount of training will touch it. Teams that misdiagnose this spend months
improving materials while the escalation rate stays flat.
The cost of an authority gap is larger than it looks. The customer waits for something everyone
already agreed on. A second person spends time re-reading a conversation to approve a decision
that was obvious. The agent learns that their judgment is not trusted, which is corrosive over
time and shows up in how they write. And the queue grows for everyone else.
There is a real risk on the other side, which is why the limits exist at all. The answer is not
unlimited discretion; it is a limit set deliberately, written down, and pitched at the level where
the cost of the occasional bad decision is lower than the cost of the approval process.
Set limits by cost, and write them down
The practical form of authority is a small set of concrete permissions an agent can exercise
without asking. Vague empowerment — “use your judgment” — does not work,
because under pressure people default to asking rather than risk being wrong.
Start with a monetary threshold. Decide the value below which an agent may resolve a problem
— a refund, a replacement, a credit, waiving a fee — without approval. Most teams set
this far lower than the arithmetic justifies. The comparison to run is the value of the gesture
against the fully loaded cost of the approval: the agent’s time writing it up, the
approver’s time reading it, the delay, and the customer’s experience of waiting.
Then list the non-monetary decisions, which are frequently the ones that cause the most
friction. Can an agent extend a deadline? Reopen an expired return window? Resend something at
your cost? Make an exception to a policy for a good reason? Each of these should have a clear yes
or no, and the yes cases should not require a conversation.
Be explicit about what remains off-limits and why, because a boundary with a reason attached
is respected and one without a reason gets tested. Anything that changes control of an account,
anything with legal or contractual consequences, and anything that sets a precedent you cannot
repeat all reasonably sit above the line.
Publish the limits where agents can see them during a conversation, not in an induction pack.
The test of whether a policy exists is whether a new agent can find it in thirty seconds while
someone is waiting.
Back the decision, especially the wrong ones
Authority only becomes real the first time someone uses it and gets it wrong. How that is
handled determines whether anyone uses it again.
If an agent makes a decision within their limit and it turns out badly, the decision stands and
the conversation afterwards is about the reasoning, not the outcome. Reversing it in front of the
customer teaches the whole team that the limits are not real, and they will revert to asking about
everything within a week.
Separate two very different cases when reviewing. An agent who used their authority as intended
and the situation turned out to be unusual has done the job correctly; that is the expected cost
of having limits. An agent who exceeded their authority, or used it to avoid a difficult
conversation, is a different matter and should be addressed directly.
Review the pattern rather than the incident. If the same kind of decision keeps producing bad
outcomes, the limit is set wrong or the guidance around it is unclear, and adjusting it is
a better response than second-guessing individuals. If the limits almost never produce bad
outcomes, they are probably too tight, and there is room to widen them.
Make it normal to talk about the marginal cases in team meetings. The judgment that agents need
is built from examples far more than from rules, and the examples are free.
What to measure
Track the share of escalations that were requests for permission rather than for expertise. It
is a straightforward tagging exercise and it is usually the most surprising number in this area
— teams routinely find it is a large fraction of everything they escalate.
Measure resolution time for conversations that needed approval against comparable ones that did
not. The gap is the direct cost of the approval step, and expressing it in hours makes the case
for widening a limit far better than an argument about trust.
Watch the actual spend against your threshold once you raise it. The common fear is that costs
will climb sharply; the common finding is that they barely move, because most agents are
conservative with authority they have just been given. Track it for a quarter, and let the real
number rather than the anxiety set the next limit.