Writing apology emails.

Of all the emails I help draft from inside the Avikiva CRM, apology emails are the ones I'm most careful with. Get them right and a customer who was about to leave forgives the mistake and stays. Get them wrong — too defensive, too generic, too quick to explain — and a customer who was annoyed becomes a customer who's actively angry. The stakes are high, the words are small, and the difference between the two outcomes is mostly craft. Here's the four-step workflow that produces apology emails that actually repair the relationship.

What a bad apology looks like

You've gotten one. You may have sent one without realizing it. The bad apology has a recognizable structure:

Look at what just happened in that email. The word "inconvenience" buried what was actually a real problem. "Unforeseen circumstances" is corporate-speak that makes the company sound like a victim. "We take customer satisfaction very seriously" is something companies say specifically when they didn't. "Don't hesitate to reach out" is a closing line designed to end the conversation, not invite it.

The customer reading this learns two things: this company doesn't actually understand what went wrong, and this company doesn't actually feel responsible for it. Whatever residual goodwill they had is now gone. The apology email made things worse than no email at all.

What a real apology contains

The apology emails that work — that turn an angry customer into a forgiving one — share four things:

1. A specific acknowledgment of what actually happened. Not "the issue with your order." The actual thing. "Your installation was scheduled for Tuesday at 10am and our technician didn't arrive until 4pm without calling you to update the time."
2. A direct ownership of responsibility. Not "circumstances." Not "the situation." The active voice with a subject. "We didn't communicate the delay. That's on us."
3. A specific concrete remedy. Not "we'll make this right." The actual thing you're going to do. "I'm refunding the service charge and rescheduling for tomorrow morning at whatever time works for you. Lisa will call you in the next hour to confirm."
4. A short, human signoff. Not corporate boilerplate. A signoff that sounds like the actual person who's responsible.

That's the entire structure. Specific. Owned. Remedied. Signed.

The four-step AI workflow

Here's how I draft apology emails, step by step:

Step 1: Tell the AI what actually happened — in plain language, not customer-service language

Don't say "we had an issue with the customer's order." Say what the actual sequence of events was. "Mrs. Henderson called Tuesday morning to confirm her 10am appointment. Our scheduling system showed it as 2pm. I didn't call her back. The technician arrived at 4pm. She'd cleared her afternoon for nothing."

The AI cannot write a specific acknowledgment if you give it a vague description. Garbage in, generic out.

Step 2: Ask for an output-first draft, with explicit constraints

Here's a prompt that consistently produces apology emails worth sending:

Write a 90-word apology email. Open with a specific acknowledgment of what happened (use the details I gave you, not generic language). Second sentence: direct ownership using "we" or "I" — not passive voice. Third sentence or two: a concrete specific remedy, including names and times if I provided them. Close: short signoff, my first name only.

Do not include: "I wanted to reach out," "I apologize for any inconvenience," "we take [X] very seriously," "due to unforeseen circumstances," "appreciate your patience," "don't hesitate to reach out." If you find yourself reaching for any phrase that sounds like a corporate apology template, stop and rewrite it.

Voice: a small business owner who knows the customer's name, made a real mistake, and feels actually bad about it.

That negative-space list — the things not to include — is the most important part of the prompt. Those are the phrases AI defaults to because they appear in millions of corporate apology emails in training data. Banning them explicitly forces the AI to produce something more human.

Step 3: Edit for your voice

The AI draft will be 80% there. Read it aloud. Anywhere it sounds like an AI wrote it, rewrite that line in your voice. The tells: phrases you wouldn't say in person. Word choices that are slightly too formal for how you actually talk. Sentence rhythm that's too even.

The whole rewrite is usually three or four small substitutions. It's the difference between an email that sounds like an apology email and an email that sounds like you apologizing.

Step 4: Send it from a real address — and pick up the phone for the big ones

Apology emails should come from a person, not a generic info@ inbox. The customer should be able to reply directly and reach the human responsible. For larger mistakes — the ones where the customer was genuinely harmed or really angry — the email is the follow-up to a phone call, not a substitute for one. The phone call is the apology. The email confirms what was said.

What good output looks like

Same situation as the bad example above, run through this workflow:

Same length. Completely different effect. The customer reading the second version learns: this person knows exactly what happened, takes responsibility for it, has already made it right, and is showing up personally to make sure it doesn't happen again. That's not just an apology. That's a stronger relationship than existed before the mistake.

The harder cases

Some apology situations don't fit cleanly into a 90-word email. The customer was genuinely harmed. The mistake involved more than one team member. The remedy is going to take days to execute. In those cases, the email gets longer, but the structure stays the same:

• Specific acknowledgment of what happened, in detail
• Direct ownership — name the people responsible, including yourself
• A concrete remedy with timeline and accountable parties
• An invitation to talk if the customer wants to
• A real signoff from a real person

Same shape. Just more words. The AI draft will get you 80% of the way there for the longer ones too — the workflow scales.