Atiba de Souza.Start a conversation
WritingAI

The Customer You've Served for Fifteen Years Might Be the Wrong One to Talk To

By Atiba de Souza

An editorial cover image for a piece on The Customer You've Served for Fifteen Years Might Be the Wrong One to Talk To. The cover reads: AI The Customer You've Served for Fifteen Years Might Be the Wrong One to Talk To

The call she almost ended wrong

Ariana Demers got on a call to cancel. She'd emailed asking to pause her subscription. That's the whole context you need going in: someone reaching for the off switch.

Atiba didn't answer the cancellation. He walked her through the emotion first, then where she actually stood, then what was actually stopping her. And then he put seven pages of LinkedIn prospects in front of her, generated every month by his team, that she had never once looked at.

She believed the only way she could sell was from a stage. She was waiting on a brand deal that might put twelve people a month in front of her. Meanwhile roughly ninety people a month were already arriving, and nothing was being done with any of them.

The call she opened by asking to pause ended with her asking to be helped more, and a Monday follow-up with a proposal for how to keep paying.

That's not a story about a save. It's a story about what happens when you finally look at who's actually showing up, instead of who you've decided is supposed to show up.

The Win, the steps, and the narrow — and why you can't skip to the last one

The 7 Practices of Effective Marketing start with The Win, a measurable outcome everyone agrees defines success before a single tactic gets discussed. If you don't have that number written down, stop reading here. Everything below this point is a way of aiming more precisely at a target, and there's no aiming without a target.

Then comes Thinking Steps Not Tactics, the ordered path from where you are to that Win, mapped before anyone touches an ad, a funnel, or a launch. This matters here specifically: if you haven't mapped the steps, you don't yet know whether "who we're talking to" is even one of the steps that's broken. You might be about to sharpen a message aimed at a problem you don't actually have.

Only once both exist does Narrow the Focus do real work. It exists because at any given moment there's more than one important thing you could say, and you can only aim a message at one target at a time. A message written to cover everyone you've ever served reaches no one.

Nobody skips this third practice on purpose. It gets skipped because it's invisible. You think you've done it. You have a persona doc, a customer avatar, a slide with a name and an age range on it. What you haven't done is test it against reality.

Real customer versus ideal customer

Here's the distinction that dissolves the confusion: a customer you've served loyally for fifteen years and an ideal customer are two different things. You can serve someone well for a decade and still not want your entire message built around getting more of them.

The test is simple: would you be happy if only that kind of person walked in, forever?

Most businesses have never run that test against their actual customer list. They've run it against their memory of their favorite customer, or their fear of turning anyone away. That's not the same test.

Ideal Customer: not who you can serve, and not who you have served — the person you'd be glad to see arrive if every future customer looked exactly like them.

Steve came in certain, and he was wrong

Steve Derrington came to Atiba wanting an audit of his ads agency. He'd already decided what the problem was: the ads. He wanted someone to confirm it and then take the account over.

The data said the opposite. Steve already had more than enough lead flow, from the patients walking in today, to beat his own goal. His ads were never the constraint.

Steve was closing 38% of his consults. He needed 50% to hit his goal. Nobody in the practice had ever written either number down.

That's the whole story once you see it. He was ready to pay someone to rebuild something that had nothing to do with his goal, because the thing that was actually broken never showed up on a dashboard, it was invisible the same way an untested ICP is invisible. He wasn't tracking it, so he couldn't see it, so he assumed the visible thing, the ads, must be the problem.

If Atiba had taken the job Steve asked for, Steve would be a year further along and no closer to the number he actually needed. Instead the engagement he got wasn't the one he requested. It was the one his own numbers proved he needed.

The carry-over to ICP work is direct: before you interrogate who your marketing is aimed at, confirm that audience is genuinely your constraint. Sometimes the problem isn't who's showing up. It's what's happening to them once they're in the room, which was also, underneath it, Ariana's problem.

Where AI earns a seat, and what it costs to ask

Once you've confirmed the audience really is the constraint, here's the pressure test. And here's the part most people underplay: the moment you run it is the moment you risk finding out the story you tell about your own business is wrong.

Pull your actual customer list, the last twelve months, the ones who paid, stayed, and referred. Not the persona doc. The real names and the real behavior behind them. Feed that to AI alongside your written ICP. Ask it one direct question: does the pattern in this data match the person I said I was writing for, or have I been describing the customer I remember instead of the one who's actually buying?

That question is not neutral. If the answer comes back a mismatch, one of two uncomfortable things is now sitting in front of you: either the people you've built your identity around serving aren't the ones keeping the business alive, or the ones keeping it alive aren't the ones you've been proud to talk about. Steve thought his ads were failing him. Some of you are about to find out your ICP has been failing you the same way, quietly, for years, while you pointed at something else.

That's the actual use case, and it's why AI belongs in this step specifically. It's fast and cheap at the thing you're slow and defensive about doing yourself: pattern-matching across a hundred real data points instead of the three anecdotes you keep in your head, without flinching at what it finds. It won't tell you who your ideal customer is. It will tell you, plainly, where your story about them and the evidence stop agreeing, and then you have to decide what to do with that.

Inconsistent applicationConsistent application
Message written to cover every customer you've ever servedMessage aimed at the one ICP you defined and tested
ICP written once, filed, never recheckedICP checked against the last twelve months of real behavior
"Am I serving this person well?""Would I be happy if only this person walked in?"
The mismatch surfaces eventually, after money's spentThe mismatch surfaces this week, before a dollar goes out

When not to do this

Said plainly, twice, because it's not a footnote: if you're too new to have twelve months of real customers to test against, don't run this yet. You'll just be formalizing a guess and calling it data. Get to real transactions first.

And if you haven't defined a measurable Win, skip this entirely. Narrowing your focus onto the wrong target with more confidence is worse than not narrowing at all. This practice matters most of the seven, but it's third in the order for a reason.

Monday

Yes

No

The Win
defined and measurable

Thinking Steps
mapped, in order

Narrow the Focus
one ICP written down

Test it
AI vs. 12 months of real customers

Match?

Sharpen the message,
spend with confidence

Rebuild the ICP
around who actually buys

Pull the list. Twelve months, real customers, who paid, who stayed, who sent you someone else. Put it next to your written ICP and ask AI where they diverge. Then ask yourself the only question that matters: if only that kind of person, the one the data actually shows you, ever walked through your door again, would you be happy?

If the answer is no, you don't have a marketing problem yet. You have a decision to make about who you're actually building this for.