An operations manager writes an escalation matrix. Four severity levels, six categories, two approval tiers and a flow diagram.
It is circulated, praised and filed. Six weeks later a serious issue sits with a junior member of staff for two days because they could not work out which box it fell into and did not want to look foolish asking.
The document was thorough and correct. It was also unusable at the moment somebody needed it, which is the only moment that counts.
1. Write rules that fit on one screen
If it needs a document to interpret, it will not be used under pressure. The test is whether somebody can recall it during a difficult conversation without looking anything up.
2. Name the triggers concretely
Anything involving safety, anyone threatening legal action, anything above a stated value, any request from a named account. Concrete triggers are recognised and abstract severity levels are argued about.
3. Make escalating explicitly safe
The single largest cause of late escalation is people fearing it reflects on them. Saying repeatedly that escalating early is correct behaviour is what changes it, and it has to come from managers.
4. Give a time based trigger as well
Anything unresolved after a stated period escalates automatically, regardless of category. This catches the cases where nobody was sure and everybody waited.
5. Name a person, not a department
Escalating to support management is a route to nobody. A named person, with a named deputy, is what makes an escalation actually arrive somewhere.
6. State what happens after escalation
Who owns it, what the timescale is and whether the original person stays involved. Without this, escalation feels like losing a case rather than getting help.
7. Cover the out of hours version separately
The route at three in the morning is different from the route at three in the afternoon, and people need to know it before they need it rather than working it out while something is going wrong.
8. Distinguish escalation from handover
Passing something to a colleague of the same seniority is not escalation. Confusing the two produces both unnecessary escalations and things that circulate sideways forever.
9. Make the automated escalation rules match the human ones
An assistant handing over on the same triggers that a person would use keeps the whole system coherent, which is the boundary in when to let a chatbot answer and when to.
10. Review the escalations monthly
What was escalated, what should have been and what should not. This review is what keeps the rules calibrated and it is almost never scheduled.
11. Treat a pattern of escalations as a content gap
Where the same question escalates repeatedly, the answer is missing rather than difficult, which is the ground in building a single source of truth for answers.
12. Measure how long escalated items wait
An escalation route with no timescale simply moves a delay somewhere less visible, which is worse than not escalating at all.
13. Escalate the prevention too
Where escalations cluster around one cause, somebody needs to fix the cause, which is the same principle as prevention advice customers actually follow.
14. Start from what people are actually asked
Rules written from the real contact mix rather than from imagination, which is the method in mapping where repetitive enquiries actually come from. See features, start at try, and the sector view sits under small business.