Stop the constant escalations

here is how to bring some sense of calm to your cx function

Hi Team!

It’s been a hot minute; hope you are well!

This Saturday night, we head out on Air France business class to Sicily with a layover in CDG, spend the week at hotels that run roughly $700+ a night, take a boat to Malta for two days, then head back to NYC.

ALL WITH POINTS!!

Malta will be country #40, and I genuinely can’t remember the last time I paid cash for a hotel…

If you needed to hear it one more time… go get yourself some signup bonuses and see the world. PLEASE.

Now, back to regularly scheduled programming. In the last few weeks, I’ve seen several CX issues on social handled publicly by the CEO, and it got me thinking.

I’ve heard this story a million times, and even experienced it myself (I am a nightmare customer 🙃)

A customer spends two days arguing with support, gets increasingly pissed, asks for a manager, gets passed around a few times, and eventually lands with someone in the C-suite.

That person reads the thread, apologizes like an actual human, makes a decision, and the customer immediately calms down.

More often than not, the C-level exec does not even offer a discount or a refund; they validate and talk like a human.

So why does this happen? And what can brands do to avoid it?

Let’s get into it.

This week’s newsie is brought to you by my friends at Kim.cc.

Tickets closed by AI have an average re-open rate of 50%+

A closed ticket doesn’t necessarily mean a resolved issue.

When a customer has to come back because the first answer didn’t solve their problem, the brand pays for the same issue twice: once for the AI interaction, and again for the human cleanup.

That makes re-open rate one of the clearest signals of whether support AI is actually working.

The opportunity isn’t to automate more tickets. It’s to resolve more of them the first time.

But resolution requires more than retrieving the right answer from a knowledge base. It requires judgment — knowing how to handle edge cases, exceptions, ambiguity, and situations where the documented answer isn’t enough.

That’s the layer most AI support tools lack.

Kim.cc adds it.

Kim is an AI-native service for ecommerce customer support. AI handles the volume, while Sentinels, trained CX, QA, and prompt experts, implement, train, and continuously improve the system, stepping in where human judgment is needed.

And when a ticket goes wrong, Sentinels don’t just correct that one interaction. They teach the AI how to handle that situation the next time.

That’s how Kim works across 200+ ecommerce brands: continuously improving how tickets are handled so more issues are resolved the first time, and fewer customers have to come back.

Because the real measure of support AI isn’t how many tickets it closes.

It’s how many stay closed.

Sometimes they just want you to give a sh!t

I think companies misunderstand what an angry customer is actually asking for more often than we realize.

Sometimes they absolutely want their money back. If the product showed up broken or you charged them for something you shouldn’t have, you certainly should do the right thing.

But plenty of customers aren’t primarily fighting you over $20 or the 20% discount on the “next order.”

They’re fighting because something went wrong, they took the time to explain it, and every response since has made them feel like nobody at the company actually understands why they’re upset.

Someone writes four paragraphs explaining what happened and gets a $5 credit with “we apologize for any inconvenience this may have caused.” Depending on the situation, the $5 almost makes it worse. It’s insulting.

The customer wanted somebody to acknowledge that the experience sucked, and the response feels like you assigned a very small dollar value to making them go away.

Then the thread reaches somebody senior, who writes, “I read through everything, and I’m really sorry we handled this the way we did. I completely understand why you’re frustrated.”

The customer finally feels heard.

Maybe the resolution stays the same. Maybe they don’t get another dollar. Somehow the temperature of the entire conversation changes.

We spend a lot of time training CX teams on rules and policies, and considerably less time teaching people how to recognize what the customer is actually asking for.

Sometimes the right answer is money. Sometimes it’s fixing the product. Sometimes it’s simply having one person at the company say, credibly, “Yeah, I see why you’re pissed.”

We need to hire folks with EQ and let them cook.

You trained everyone to escalate

Escalations are useful when something genuinely requires more authority or expertise. Every company will have situations where the stakes are high enough, the request is unusual enough, or the risk is significant enough that somebody senior should make the call.

The problem starts when escalation becomes the default destination for anything that doesn’t fit perfectly inside the playbook.

An agent gets a request that feels reasonable but falls slightly outside policy, so they ask a manager. The manager has more authority, but the safest decision is still to follow the existing process.

The customer gets another version of the same answer, gets angrier, and by the time someone senior enters the thread they’ve spent two days trying to convince the company that the issue deserves an actual decision.

I’d pay very close attention to what changes once the executive responds.

In a lot of cases, the executive actually has less context than the original agent. They didn’t spend the last day inside the thread, and they probably know less about the day-to-day mechanics of the policy.

They simply feel comfortable deciding that this particular situation deserves an exception, or admitting that the company handled something badly without worrying about whether the wording is approved.

The agent closest to the problem can have the most context while holding the least authority to act on it.

As the issue moves upward, the person making the decision often knows less about the conversation but feels far more comfortable making the call.

That is how you end up with a CEO who looks weirdly brilliant at customer support.

And it should infuriate you.

You threw AI at every problem

AI is genuinely good at a lot of support work now. WISMO, return status, address changes, subscription updates, basic product questions, and all the repetitive work that used to eat huge chunks of a team’s day should increasingly be handled without a human.

I’ve said this in the past… but that changes what the human job looks like.

Once the easiest tickets disappear, a larger share of the conversations reaching your team are messy. Something went wrong twice. The customer already tried the obvious fix. The normal policy doesn’t quite fit. There’s history involved.

The technically correct answer can still make the situation dramatically worse.

Those conversations require better judgment than the average support ticket did a few years ago.

A lot of companies upgraded the technology while leaving everything around the human team basically untouched. They kept the same hiring profile, the same QA process, the same approval tree, the same bullshit macros, and the same fear of somebody making an exception leadership didn’t explicitly approve.

So software takes the questions that barely require judgment, while people inherit more of the conversations where judgment is the entire job.

Then we give those people even less room to use it.

I’m pretty sure we have that backwards.

Your whole CX team should not sound the same

I can’t say I hate a macro or canned response per sè. Nobody should spend their Monday typing the same return instructions 80 times because someone decided handcrafted support emails are more authentic.

But somewhere between “we should have a consistent brand voice” and “we should save agents some typing,” many companies decided that every person on the CX team should sound exactly the same.

I don’t think that’s a good thing.

You hired seven different humans. They have different personalities, different ways of explaining things, different senses of humor, and different instincts for dealing with people. Then you handed all seven the same paragraph and told them not to touch it too much.

Customers can feel that.

If I write three paragraphs about a very specific situation and get a reply that could have been sent to 5,000 other people that morning, I have no evidence anybody actually read what I wrote.

The macro should get the agent 80% of the way there. Let them do the remaining 20% themselves.

There should still be guardrails. I don’t need every agent improvising your warranty policy based on their mood that afternoon.

But there is a huge amount of space between letting everyone say whatever they want and making every person at the company sound like the same damn macro.

AI handling more of the routine work makes this even more important. If a customer has reached a human because the situation is complicated enough to need one, sounding like a human seems like a reasonable place to start.

Sometimes they have permission and still don’t give a sh!t

I also don’t want “empower your team” to become the explanation for every bad CX organization, because I’ve seen brands give people plenty of freedom and still get painfully mediocre support.

The team can make exceptions. There’s budget for surprise and delight. Leadership has explicitly told them to waive fees, send replacements, spend a little extra, and use their judgment when the situation deserves it.

Then you read the tickets, and everyone is still sending the same macro, nobody follows up, and a customer who has ordered 14 times gets treated exactly like someone placing their first order.

At some point, that stops being a permission problem.

Sometimes you hired people who are good at working a queue but don’t naturally care enough to look beyond the ticket in front of them.

The same goes for CX leadership. Someone can keep response times down, maintain the macros, run the weekly meeting, and send you a perfectly nice CSAT report while having almost no POV on whether the actual customer experience feels good.

That matters even more as AI takes over the repetitive work. The human job is increasingly made up of messy situations where the policy won’t tell you exactly what to do.

I’d rather hire someone who is curious enough to understand what happened, cares enough to want to fix it, and has the judgment to know what a reasonable fix looks like.

You can teach someone your policy and give them a bigger budget. Teaching them to actually give a shit is a lot harder.

Your customer has no idea who any of you are

Companies underestimate how absurd their internal processes can feel from the outside.

Inside the company, the ticket moved to Operations because Operations owns that workflow. Finance has to approve something, the manager is out today, the warehouse hasn’t responded yet, and the original agent went off shift.

To the customer, they emailed one company on Monday and have spent the rest of the week explaining the same problem to that company.

They don’t know which department owns replacements, and they definitely don’t care why somebody in Finance needs to approve $60. They want one person who knows what happened and stays attached to it until it’s over.

Executives tend to be good at this because they collapse all of that internal complexity. They take responsibility for the company as a whole and tell the customer what happens next.

Your frontline team should be able to create that same feeling without personally running Finance, Operations, and the warehouse.

Here are the 8 things I’d change

If escalations keep landing with leadership, I’d start here:

• Pull your last 20 to 30 executive escalations and compare the first replies with whatever finally resolved them. Look at what materially changed once someone senior got involved. Sometimes the remedy will be different. In other cases, the biggest change will be tone, clarity, ownership, or somebody finally feeling comfortable making the decision.

• Find the exact response where the customer became meaningfully more upset. Most ugly threads have a moment where the original problem becomes secondary and the customer starts getting angry about how the company is handling it. Figure out what your team said right before that happened.

• Stop automatically reaching for discounts when somebody gets angry. Figure out what they actually want first. If you owe them money, give them the money. If they’re looking for accountability, a tiny coupon in place of a real acknowledgment can feel more insulting than useful.

• Write down what agents can actually do without asking permission. Refund limits, replacements, shipping upgrades, fee waivers, credits, phone calls, and common policy exceptions should have clear boundaries. If your team has to guess where their authority ends, they’re going to send anything remotely unusual upstairs.

• Look for decisions managers approve almost every time. If managers keep rubber-stamping the same type of $50 or $75 exception, move the decision closer to the person who already has the customer context. Your approval history is giving you a pretty good map of authority you no longer need to centralize.

• Let people personalize the macros. Every meaningful response should contain enough of the agent’s own language that the customer knows somebody actually read the conversation. Consistency matters, but if every person on your team sounds interchangeable, you’ve probably taken it too far.

• QA judgment alongside compliance. An agent can follow every policy, use every approved phrase, and still make the customer much angrier. Review whether the response moved the situation forward, whether the agent understood what the customer needed, and whether a reasonable decision earlier in the thread could have prevented the escalation.

• Turn good judgment into training. When an agent handles something beautifully or an executive saves a difficult conversation, break down what worked. Show the team how the person thought about the situation, which details mattered, what they were willing to do, and why. The same type of issue should require less seniority the next time it happens.

I think senior leaders should keep reading customer emails. They should answer some too. There are few better ways to understand what your company actually feels like from the outside.

But customers should feel those standards long before anyone with “Chief” in their title has to enter the thread.

If the apology gets warmer, the ownership gets stronger, and people become much more willing to make a reasonable decision every time another title gets added to the email chain, I’d look pretty hard at what the organization taught the first person who answered.

Your customer probably didn’t need the CEO. They needed somebody who was allowed to solve the problem.

That’s it for this week!

Any topics you’d like to see me cover in the future?

Just shoot me a DM or an email!

Cheers,

Eli 💛