How to make every customer feel seen with AI (without the creepy part)

TEG is an official member of the Anthropic Claude Partner Network. We build Claude Skills, Projects, and MCP integrations for B2B ops teams. See our full Claude services.

In 2022, a woman named Anna Brose tried to return an unopened bag of dog food after her dog Gus died. Chewy refunded it, suggested she donate the food to a shelter, and sent flowers with a note signed by the person she had talked to. Her tweet got more than 600,000 likes. The coverage is in Fortune, Newsweek and the Boston Globe.

Nothing in that story is clever, and none of it was AI. A person noticed what was actually happening to a customer and acted on it. The hard part is doing that when you have a thousand customers and a team of ten, because no one can remember that much. AI does not replace the caring. It gives a small team the memory that caring depends on.

Statisticians call a study of one person an "n of 1." It is a good name for the feeling you want a customer to have. Here is how to give it to thousands of customers, and where it goes wrong.

What does it take for someone to feel seen?

We think of it as four things. A customer feels seen when your message shows that you:

1. Remember them. You know what has been said, sold, promised or left unanswered.
2. Understand their situation. You know their role, their size and what is probably on their plate.
3. Noticed something specific. You mention a real detail instead of a category.
4. Heard what they told you. The message responds to their words, not your campaign calendar.

All four are recall and research problems, which is where Claude is strong.

Why does it break at scale?

It is rarely a lack of caring. The details are scattered across a CRM note, an old thread and one person's head, and nobody has time to read all of it before every message. So it goes out generic, and the customer can tell.

Mail merge puts a first name in a template and calls it personal. Most people spot that in a second. Much of what you will read about "segment of one" personalization is about software that picks products to show. This is a different job: writing to a person the way someone who knows them would.

What does the difference look like?

Here is an illustrative example, not a real client. A customer is three weeks into onboarding. The templated message looks like this:

Hi Sam, just checking in to see how onboarding is going! Let us know if you have any questions.

Before drafting the second version, Claude builds a short brief from the CRM, the kickoff call notes and the email thread: the last contact, what is still open, what changed, and the one thing worth saying. The draft that comes from it looks like this:

Hi Sam, on the kickoff call you said the Friday reporting was eating your team's week. The report template we set up on the 9th is live, but I do not see that anyone has opened it. Want me to walk Priya through it this week?

The second message is not better writing. It is better preparation, and that is the part that scales.

How do you build this?

Start with one group that deserves more attention than it gets, such as customers in their first 30 days or accounts six months from renewal. Then:

1. Connect the sources. Claude reads your CRM, email and meeting notes through connectors, so each person's history is in one place.
2. Write down what your best person checks. Save the questions your best account manager asks before reaching out as a Claude skill.
3. Have Claude brief before it writes. The brief is what makes the draft specific.
4. Check before you contact. Has anyone already talked to this person?
5. Review every draft at first. A person reads and approves each message until you trust the output.

Run it on ten customers and read the drafts. Would you be glad to receive them? If yes, widen the group.

Where does it turn creepy?

There is a line between feeling seen and feeling watched. A useful test is whether the person would expect you to know the thing you are referencing. Their last email to you, their order history and the goal they told you about are fair. A detail that exists only because you tracked their behavior is not.

Say you have a workflow that analyzes who visits your website. Do not reach out with "I saw you were on our website." That is proof you were watching, and it helps no one. Write about the things it makes sense for you to remember: the conversation you had, what they asked for, what is still open.

If you want help finding the first group worth doing this for, our free AI Value Assessment is built for that. We listen to how your team works and show you where Claude can take some of the load.

Frequently asked questions

What does it mean to make a customer feel like an n of 1?

It means the customer feels like the only one you are talking to. The message reflects their history, their situation and what they actually said, instead of a template with a name swapped in.

How is this different from mail merge or personalization software?

Mail merge inserts fields like a first name. Personalization software mostly picks products or content to show. This is closer to what a good account manager does: read the person's history across your own CRM, email and notes, then say the one thing worth saying.

Does Claude send these messages automatically?

It does not have to. In the workflows we build, Claude researches and drafts, and a person reads every message before it reaches a customer. Sending automatically is a choice you can make later, once you trust the drafts.

How do you keep personalization from feeling creepy?

Only reference what the person would expect you to know, such as their past emails, orders and stated goals. Skip anything that reveals you were tracking their behavior. If a workflow flags who visited your website, do not open with "I saw you were on our website."