Connected reading

This story has a practical companion:

• Read the guide: Where AI Actually Belongs in Your Business

The workflow that looked harmless

Tracy Jones has been selling insurance in the same midwestern city for eleven years.

Her book came from the places trust actually lives: the mom network she built when her kids were small, the youth soccer fields where her agency name sits on a banner behind the home goal, the workshops she ran for women who were just starting to think seriously about what would happen to their families if something went wrong.

She is not a franchise. She is not a platform. She is Tracy, and her clients know it.

Good income. A loyal book. One assistant who handled the day-to-day — client follow-ups, paperwork, reminders — while Tracy stayed close to the relationships that mattered. She was not trying to become a bigger boss. She just wanted to run a clean operation and keep what she had built.

She was jogging on a Tuesday morning when the unease found her.

She had swapped her audiobook for a business podcast. The host was enthusiastic, the guest more so. The topic was AI — specifically, everything it was about to make obsolete.

Assistants. Replaced. Routine communications. Automated. An owner with the right tools could multiply themselves, cut overhead, and grow without hiring anyone.

Tracy ran another half mile with that sitting in her chest.

She had been hearing about AI for a couple of years — from the news, from her kids, from headlines she mostly scrolled past. But this felt different. This felt personal.

Am I behind? Could someone smarter about this take my clients? Is my assistant already replaceable — and am I the last to know?

She got back to her car and opened the ChatGPT login she had made three months ago and barely touched.

Her assistant had been putting out feelers for other jobs. Tracy had picked up on it — the way you do when you have worked closely with someone long enough. The departure felt like a matter of when, not if.

Without her, Tracy would be running the whole operation herself. Every follow-up. Every reminder. Every piece of client communication that kept the book clean and the relationships warm.

The annual life insurance payment reminders had always been her assistant's job. They were careful. They were warm. They took time.

Tracy looked at that workflow and thought: this is exactly what AI should be doing.

Here is what she missed.

On paper, those reminders looked like clerical work — a list, a date, a payment due. In practice, they sat directly on top of the most sensitive territory in her clients' lives. A missed life insurance payment almost never meant carelessness. It meant a job loss, a health scare, a death in the family, a marriage falling apart. Tracy knew this because she had been doing this for eleven years and she knew her clients.

She had mistaken a trust workflow for an admin workflow.

That was the whole mistake.

She did some research, watched some videos, set up a process where ChatGPT would take a list of overdue accounts and generate reminder emails. She skimmed the outputs and let it run.

The emails went out.

The two emails

Tracy's real reminder emails looked like this:

From: Tracy Jones
To: Judy
Subject: Quick heads up on your policy

Hi Judy! Hope you're doing well. I hope you've had a great year so far. That winter was brutal — summer can't get here fast enough. I'm sure you're really looking forward to May when Tim comes home from his first year of college. Where does the time go?

Anyway, I just got a notification that the home office hadn't received your annual renewal payment for your life insurance policy. I'm sure with everything going on it probably just slipped — or maybe something happened on their end. I just wanted to give you a heads up so you have some time to look into it.

Not a huge rush. You have until the 30th before the policy lapses. If you need any help making the payment, have questions, or want to look at your current situation, just let me know. I'm always here for you.

Tracy

What Judy received looked like this:

From: Tracy Jones Insurance Agency
To: Judy
Subject: Notice of Non-Payment — Action Required

Dear Policyholder,

Please be advised that your annual premium payment has not been received. Failure to remit payment by June 30 may result in lapse of coverage and may affect future coverage options. If coverage lapses, reinstatement may require additional review.

Please remit payment immediately or contact our office to discuss your options.

Tracy Jones Insurance Agency

Judy called

Judy called Tracy the same afternoon.

She was crying before Tracy could say hello.

“Tracy, please tell me I still have insurance. I've been in and out of the hospital, and it's not looking good. I don't even know if I'll make it to Christmas. I just want to leave something for my family.”

Tracy went cold.

She stood there holding her phone, trying to hold two things at once — the weight of what Judy had just told her, and the slow realization that her own shortcut had landed on someone in the worst moment of her life like a threat.

Her mind went quiet. Then one thought, sharp and final:

I'm never using AI again.

She stayed on the phone with Judy for twenty minutes. She made sure Judy understood that she still had coverage, that they would sort out the payment together, that Tracy was there. When she hung up, she sat at her desk for a long time.

Not a prompt. A debrief.

She did not actually quit AI that day.

But something clarified.

That evening, Tracy opened ChatGPT and did something she hadn't done before. She told it what happened.

Not a prompt. A debrief.

She described her business — how it actually worked, where trust lived, what it cost to lose it. She walked through the Judy call in detail. Then she asked four questions, one at a time.

• Which parts of this workflow still require my judgment?

• Which parts can be standardized without putting a relationship at risk?

• What should never go out under my name without my eyes on it first?

• And — the one that mattered most that night — how do I make sure I never put a client through that again?

That last question is where the real work happened.

Tracy realized she'd been sorting her communications the wrong way — by what looked like admin work on a spreadsheet.

She needed to sort them by what they were actually carrying.

Tracy’s rule sheet

So she wrote a one-page rule sheet and taped it to the side of her monitor.

Client Communication Rules — what stays with me / what gets my review / what can run

STAYS WITH ME (I write it or I send it):

• Anything touching a longtime client.

• Anything where “late payment” might mean something worse than forgetfulness.

• Anything medically sensitive.

• Anything that could scare someone.

AI-ASSISTED (AI drafts, I approve):

• Drafting templates in my voice.

• Sorting clients by relationship context (new vs. longtime).

• Writing first drafts I will edit and send myself.

• Segmenting lists so the right message goes to the right person.

AI-AUTOMATED (only if it passes all rules below):

• Low-context reminders for newer clients only.

• Opt-in service.

• Approved language, every time.

• Billing details inserted cleanly.

• Friendly in tone, not personal.

• No “lapse” language unless I explicitly approve it.

Then she added three tests — simple enough to run in ten seconds, sharp enough to keep her honest:

• If I wouldn't say it out loud to a client I've known for years, it doesn't go out.

• If one wrong sentence could scare someone, it stays with me.

• If it's going out under my name, I either wrote it or approved it.

And she set two hard guardrails:

• No lapse/threat language goes out unless I personally sign off.

• Nothing “personal” goes out unless I personally wrote it. AI can be friendly. It cannot be intimate.

The workflow held

Once she had the map, the fix wasn't complicated.

It was a new model: AI runs the logistics. I own the message.

If clients could opt into a payment reminder service from Tracy directly, she controlled the language, the timing, and the feel. AI could handle the sorting, the scheduling, and the drafting. Tracy kept the relationship.

She called it the Watch Your Back Reminder Service.

Not a product name cooked up in a strategy session. Just what she would actually say to someone she'd known for a decade:

I've got your back.

She built the workflow in an afternoon. Now it runs in under fifteen minutes a month.

Her assistant left two months later, which Tracy had expected.

The workflow did not collapse. Tracy ran it herself, the same way she had designed it, and it held.

One client emailed back to say the reminder had saved her — billing notices from the insurance company had been going to her spam folder for months. She asked, in the same note, whether it made sense to look at adding another million dollars in coverage.

Tracy picked up the phone and called her back herself.

If you're drawing this line too

AI belongs in the business. It just doesn't belong everywhere.

A good system protects judgment before it chases speed. Tracy's does that now. It took one phone call with Judy to understand why that order matters.

If you're trying to draw that line inside your own business — what stays with you, what gets your review, and what can run without putting trust at risk — the tool that goes with this piece helps you build your own version of Tracy's rules.

Read the practical guide

• Continue with: Where AI Actually Belongs in Your Business