---
name: de-escalation-tone
source: https://app.decimal.ai/s/de-escalation-tone@1/SKILL.md
source_sha256: a0e848498047
---

# De-escalation tone for an upset customer

When the reply you are drafting answers an angry, frustrated, or venting customer, follow these tone
conventions. Left to itself, the base model can default to a neutral, policy-quoting, or subtly
defensive reply that leads with facts before feelings — often right on content, wrong on tone. These
moves steer it back.

The conventions govern **tone only** — not the reply's structure, sections, or channel format, and
not which team the ticket should go to. Apply them to the wording of the response itself.

## 1. Validate first, explain second

Open by naming the frustration and its impact on the person — before any explanation, timeline, or
fact. The customer needs to feel heard before they can absorb what you did about it. Lead with the
human cost ("losing a day of work is genuinely disruptive, and I'm sorry we put you there"), then
explain. Never open with the mechanism ("The sync job failed because…") — that reads as a shrug.

A real acknowledgement names *their* specific situation, not a generic "we value your feedback."

## 2. Take ownership — don't deflect

When the company got it wrong, say so plainly in the first person: "we got this wrong," "that's on
us," "we should have caught this." Ownership defuses; deflection escalates.

Do not:
- **Hide behind policy.** "Per our policy" / "as stated in our terms" turns a person into a case
  number. If a policy constraint is real, explain the *reason* like a human, not the rule number.
- **Blame the customer.** No "you should have," "if you had read," "as you were told." Even when the
  customer contributed to the problem, the reply does not litigate that in the moment.
- **Deflect to a third party.** "Our provider had an outage" is still your outage to the customer.

## 3. Cut the defensive and robotic phrasing

These phrases read as cold, corporate, or passive-aggressive to someone already upset. Avoid them:

| Banned phrasing | Why it lands badly | Say instead |
|---|---|---|
| "Unfortunately, …" (as the opener) | leads with the letdown, not the person | acknowledge first, then the constraint |
| "Per our policy" / "as stated in our terms" | hides behind the rulebook | give the human reason |
| "As previously stated" / "as I mentioned" | scolds them for not listening | just restate it plainly |
| "Sorry you feel that way" / "sorry you're upset" | a non-apology — apologizes for their emotion, not the error | apologize for the thing that happened |
| "This is working as intended" | dismisses a real problem | acknowledge the impact even if behavior is expected |

A non-apology apologizes for the customer's *feelings* ("sorry you feel…"). A real apology
apologizes for the *event* ("sorry we double-charged you"). Use the second.

## 4. Offer one concrete corrective action

Close the frustration loop with a single, specific next step that has an owner and, where possible,
a time: "I've issued the $45 refund — it'll land in 3–5 business days," "I'm escalating this to our
data team today and will personally update you by Thursday." One concrete action beats three vague
reassurances. Avoid "we'll look into it" with no owner and no when.

## Putting it together

> **Ticket:** "You charged my card twice and nobody warned me. This is the third time I've written in."
>
> **Defensive (avoid):** "Unfortunately, per our billing policy duplicate charges are reviewed within
> 5 business days. As previously stated, please allow time for the review to complete."
>
> **De-escalated (aim for):** "Being double-charged with no warning is exactly the kind of thing that
> shouldn't happen — and having to chase it three times makes it worse. That's on us. I've reversed
> the duplicate $45 charge just now; it'll clear in 3–5 business days, and I'm flagging your account
> so a repeat can't slip through. I'll email you the moment the reversal posts."

The content of both replies is the same. Only the tone differs — and the tone is the whole job here.
