Blog

Escalation without a tier system

Support escalation on a team of five is a handoff, not a promotion. The five things that get escalated, the four it needs, and the two channels you have.

A customer writes in at 9am about a charge they do not recognise, and by lunchtime the thread has stopped moving. The only person who can answer is in a code review. Escalation is the work moving from the person who received it to the person who can fix it. On a team of five that is the whole definition, because there are no levels for it to move up.

The tier vocabulary grew up somewhere else. Tiered support describes an arrangement with a front line and specialists behind it, a way of routing volume rather than handing over one hard question. Where the words stop fitting the mechanics still apply, and they are four requirements rather than a structure.

What the tier vocabulary assumes, and what you have instead

A tier system is a routing rule, and it needs a queue deep enough that sending work sideways costs less than answering it. Neither that nor a front line describes five people who answer their own customers.

What a tier system assumes What a team of five has
A front line that answers first Whoever is at their desk, often the person who wrote the code
A specialist group behind it The same five people, at a different point in the day
Enough escalations to justify a written rule Few enough that a name is faster than a status change
A queue deep enough to route A handful of threads, mixed

Escalation is not triage, and a small team needs the second one first. Triage decides the order of the queue. An escalation happens to work that has already passed that test, meaning it is real and the person holding it cannot finish it alone.

What actually gets escalated

“Difficulty” describes how the first person feels, and it tells the second person nothing about what to do. The escalations a software team actually raises fall into five categories, and each needs a different second person.

Category The decision it needs The context the second person will ask for
A bug only an engineer can confirm Is this our bug, and when will it be fixed? The exact request, the account, the timestamp and the steps that reproduce it
A billing question Which charge is right, and do we owe a credit? The invoice, the plan and what the customer expected to pay
A refund above the threshold one person can approve Does this leave the account inside policy? The amount, the reason and the history on the account
A security report Who triages it, and what do we tell the reporter? The full report, what was tested and whether anything is still exposed
A customer threatening to leave What are we willing to do, and who tells them? What they asked for, what we already tried, and their renewal date

None of the five is really about difficulty. A refund is easy work that needs authority, a security report needs discretion, and a customer threatening to leave is a decision with a date attached. The second sentence of an escalation is who it goes to, and the five rows have four different answers.

Security reports deserve their own line. The internet has a mailbox convention for them, security@ at your domain, written down in RFC 2142. The security.txt file exists for the same reason: researchers regularly have no channel to report through, according to RFC 9116. A report that lands in support@ at acme.example is not an escalation you approve or decline.

The four things an escalation needs to survive the handoff

A handoff succeeds when the second person can start without talking to the first one. Four things make that possible.

Requirement What it sounds like written down What its absence costs
An owner “Priya has this one” Two people assume the other is on it
A reason “The customer cannot log in, and the fix is in the auth service” The second person re-reads the whole thread to find the ask
The context they will ask for The request, the account, the reproduction, the attempt that failed A round trip that ends with the customer explaining again
A time by which someone will look “Priya picks it up before Thursday standup, or it goes back to me” The thread waits behind everything with a real appointment

Most escalations fail on the last two. An owner and a reason are natural to write, because the first person already knows both. The context is work, and the deadline is a promise that someone has to keep.

Incident response wrote the shape of a clean handoff down first. Google’s incident management guidance is blunt about it: the new owner is apprised of the situation, the handoff is stated explicitly, and the person handing over waits for acknowledgment before leaving the task (Managing Incidents). The same moves are all a five-person team needs: say who has it, say what it is, and hear the person agree.

The deadline is the one a small team skips. An escalation is a commitment the customer cannot see. A support SLA is the vocabulary for the outside promise, and this is the internal version of it: not a reply time, just a time by which somebody will have looked.

The two channels a small team actually has

By the end of the exercise there is a rule, an owner, a reason and a deadline, and those four things have to live somewhere. A small team has exactly two places to put them.

A written handoff in the same thread A second tool that holds the work
What it is A note or a forward on the customer’s own thread A Linear issue, or the tracker your engineers already use
What it carries well The customer’s words, the history, the reply that has to come back Status, owner, labels and a list of everything open
What it costs Nothing to set up, and one place to look A second place to look, and the customer’s context has to be pasted in
How it fails Nobody is assigned, so it becomes reading rather than work The issue is filed and the thread is forgotten, so the customer is never told

The same-thread version is cheap and keeps the customer’s own words in one place. Its weakness is that a thread has no status, so “is this handled” is answered by re-reading it. The second-tool version gives the escalation a state other people can see, at the cost of splitting the record in two.

On a team that already tracks its work in Linear, the second channel is close to free. The cost is the third requirement: somebody pastes the customer’s request and account details into the issue, and pastes the answer back into the thread when the issue closes. Skip it and the engineer fixes a bug without knowing who is waiting on it.

When the engineer is also a founder

On a team of five the person receiving the escalation is often the person who wrote the code, and sometimes the person who can approve the refund as well. That collapses the handoff into a decision about whether to stop what they are doing.

The categories where the founder is the only possible second person are the ones with no engineering in them. A refund above the threshold, a credit that breaks a pricing rule, a commitment about a date, and a customer who has asked for a call are all decisions about money or the roadmap. There is no reproduction to write, because there is no bug.

The failure is about visibility rather than authority. A founder has no queue and no shift, so an escalation that arrives as a chat message competes with the rest of the day. Naming the decision and the day it will be answered is the whole escalation.

Why “just ask in Slack” loses the record

A chat message is a request, not a record. It is written in a hurry, it has no status, and it stops being findable by the end of the day, which makes it the cheapest handoff to send and the most expensive one to inherit.

The escalation that lives only in chat has an owner for as long as the conversation is on someone’s screen. Once the day moves on, nothing says which escalations are still open, and the customer’s message is in one system while the work is in another. The next person to ask what happened searches twice, asks a human, and gets whatever that person remembers.

That gap is also what makes an unowned thread expensive later, since a thread nobody picked up is the one that shows up in first response time once the customer has stopped expecting an answer. Deciding where the record lives is the category question, and shared inbox vs help desk covers it.

Where Chirp fits

Chirp is a shared inbox for small software teams that track their work in Linear, so an escalation can be a thread linked to an issue. Press i in a thread to link it to an existing issue, create one, or search for the one that already exists, and a assigns the thread an owner. t tags it and s snoozes it.

The reason and the context stay in the thread, and the issue holds the state. When every linked issue is Done or Canceled, the thread returns with a reply drafted for you that names the issue. No rule in Chirp can send mail, and the API can send replies but cannot link issues or write drafts.

The Issues view (g i) sorts linked issues by how many customers are waiting, which is the escalation backlog in one list. Chirp connects to Gmail, Google Workspace or IMAP, sets access per inbox with the roles Admin, Agent and Viewer, and costs $9 per user per month.

FAQ

How do I escalate a support ticket without tiers? Write down four things and move the thread: who owns it, why it is being moved, what the second person needs to start, and when somebody will look at it. Nothing about that needs a level, a severity field or a matrix.

What is an escalation path on a small team? It is the route a request takes to the person who can finish it, and on five people it is one hop: from whoever received it to whoever can fix it. An escalation path needs a diagram only when there are several destinations.

Who should own an escalated thread? Whoever can finish it, named explicitly, with the person who handed it over staying responsible until the new owner agrees. An escalation with two possible owners has none. The owner is the person the customer’s next message goes to, even if the answer comes from someone else.

Does the customer need to hear about the escalation? They need to hear that a person has it and when to expect the next update. The handoff changes nothing about what the customer experiences: from their side there is one conversation, and anything internal is invisible until somebody writes back.

Next step

An escalation is an artifact, and it needs one place to live. What a shared inbox is covers the arrangement that gives one address a visible owner, which is the first of the four requirements and the one a forwarded thread never supplies.