Connected reading

This is Part 1 of Erin’s case.

• Read the practical companion: Why Your Best-Looking Number May Be Lying to You

• Continue to Part 2: Knowing the Right Number Didn’t Fix the Business

• Continue to Part 2 guide: How to Build a Recovery Plan for a Small Business Using AI

• Continue to Part 3: She Had a Recovery Plan. Then Sales Slowed Down.

• Continue to Part 3 guide: How to Use AI to Review a Plan Before Drift Gets Expensive

Part 1 - The month that made no sense

It was 6:53 on a Monday morning in early June, and Erin Walsh was exactly where she wanted to be.

Back right corner of the coffee shop three blocks from her office. Window seat. Oat milk latte. Forty-five minutes of silence before Slack, clients, and employees started asking things from her.

May was supposed to feel good.

She opened the income statement expecting the strongest month in the history of the agency.

The first number delivered.

Revenue: $59,500

Best month ever.

Then she looked lower.

Net Income: ($2,300)

She stared at it.

Then back at the revenue line.

Then down again.

The air went out of her lungs.

“What just happened?”

Five months earlier, the move had looked smart.

Erin ran a seven-person boutique marketing agency in a mid-sized Southeastern city. She had been in marketing for seventeen years and running her own shop for eight. The agency she worked for before had been sold. She got let go in the transition. There was no noncompete. A few clients followed her. More came later through referrals, results, and the kind of trust that compounds when you do strong work long enough that people start handing your name around.

She was careful. She kept reserves. She paid above market to keep good staff. She didn’t overextend. She remembered what happened when businesses dismissed social media too long and spent the next five years trying to catch up. She had no interest in repeating that mistake with AI.

So at the beginning of the year, while planning the months ahead, she did what a smart operator would do.

She pulled her numbers, opened ChatGPT, and asked what looked like a perfectly reasonable question.

Her agency had three core offers:

• Monthly content retainers — $3,000/month
  Blogs, newsletters, social copy, revisions, check-ins

• Website refresh projects — $8,500/project
  Messaging cleanup, website copy, conversion improvements

• Strategy intensives — $3,000/half day
  Offer direction, messaging, planning, and a written plan the client could use right away

The data she handed AI looked like this:

Current offer mix

• Legacy retainers: 7 clients

• Website refresh projects: 2 active projects

• Strategy intensives: 4 booked sessions

Recent billing trend

• Retainers were the largest and most stable revenue line

• Website work was healthy but irregular

• Strategy sessions were profitable, but smaller and one-off

And the prompt she gave it was this:

“I’m planning the year ahead and want to grow the agency intelligently. Here’s my revenue mix and active client count by offer. Analyze which offer is performing best and where I should focus growth over the next 6–12 months. Prioritize stability, growth potential, and revenue contribution.”

The answer came back exactly the way smart-looking bad advice usually does: clean, organized, and hard to argue with.

“Monthly content retainers appear to be the strongest growth engine in the business.

• They contribute the highest recurring revenue

• They improve predictability

• They create better long-term visibility than one-off work

• They offer expansion opportunities over time

Recommendation: Focus growth efforts on acquiring more retainer clients over the next 6–12 months.”

It looked disciplined. It sounded strategic. And it gave her exactly what every operator secretly wants at the end of a planning session:

a clean answer.

So she followed it.

She leaned harder into retainers.

And without fully meaning to, she started demoting one of the healthiest parts of the business in the process.

Her strategy intensives used to be a real offer. Profitable. Clean. High-judgment work. A half day, a clear plan, a defined end. Clients walked away with direction. Erin walked away with good money and almost no drag.

Then little by little, strategy stopped being a product and started becoming a sales mechanism.

A front-end conversation designed to lead into ongoing retainer work.

By the end of February, strategy as a real line of business was basically gone.

Website refreshes began thinning out too. The business was becoming more and more organized around the one thing AI had told her was “winning.”

New monthly retainers.

For a while, the revenue line rewarded her for it.

By May, the agency looked like this:

May revenue by offer

• Legacy retainers: 7 clients = $21,000

• New retainers: 9 clients = $27,000

• Website refreshes: 1 project = $8,500

• Strategy intensives: 1 session = $3,000

Total revenue: $59,500

And the monthly financial picture looked like this:

At first, she explained it away.

March was thin, but manageable.

April was uncomfortable, but maybe that was just transition pain.

She told herself the same story a lot of founders tell themselves when they’re trying to scale recurring revenue:

The new clients are always heavier at first. The systems will settle. The team will get faster. Revenue is still moving in the right direction. This is just what growth feels like before it smooths out.

Then May turned negative.

And the story stopped working.

Sitting in the coffee shop, Erin finally saw the real mistake.

She didn’t close the laptop.

She opened ChatGPT again.

Not with a clean, polished prompt this time.

More like a smart person trying not to panic in public.

She typed:

“I think I optimized around the wrong thing.

Revenue went up, but profit turned negative. I followed your recommendation to push retainers, and now I’m looking at the biggest month in agency history and the business still lost money.

Here’s what changed:

• I sold more new retainers

• strategy sessions mostly turned into sales calls for retainers

• website work dropped off

• new retainer clients took way more time than I expected

• revisions went up

• contractor usage went up

• some of the newer clients are slow to pay

Help me figure out what is actually going wrong.

I want a rough, useful view of what I actually keep by offer.

Separate legacy retainers from new retainers.

I do not need accounting perfection. I need decision-making clarity.”

That was the turn.

Not:

Tell me what’s winning.

But:

Help me figure out what I should be measuring before I decide what to push.

This answer came back differently.

Not because AI had changed.

Because the question had.

“You optimized around revenue growth, not business health.

The earlier answer was accurate for the question you asked. But revenue and client count are visibility metrics. They show what is happening at the surface. They do not show what you keep.

For this decision, compare each offer using three things:

1. What it brings in

2. What it really costs

3. What’s left after that

For new retainers, ‘what it really costs’ includes more than the monthly fee and payroll line. It includes senior review time, contractor overflow, revision burden, onboarding drag, and payment delay.

Also: legacy retainers and new retainers should be separated. They are functioning like two different offers with two different economics.”

Erin wrote back:

“Okay. Show me.”

So GPT did.

Erin’s Offer Health Map

These weren’t audited numbers. They were rough enough to show the shape of the problem — and that was all she needed that morning.

She stared at the two retainer lines first.

Same monthly price.

Completely different economics.

GPT kept going:

“Your mistake wasn’t just that you sold more of the wrong thing.

You also demoted one of the healthiest offers in the business.

Strategy intensives were producing strong contribution with low drag. New retainers were producing weak or negative contribution with high drag.

You optimized for revenue and recurring billing. That hid the real economics.”

That was the whole lesson.

Not all revenue is good revenue.

Not all recurring revenue is healthy revenue.

And the thing that looks most scalable can still quietly weaken the business underneath if you never ask what it actually leaves behind.

She had not just sold more of the wrong thing.

She had demoted one of the right things.

A visible metric had flattered her.

A useful one would have helped her decide.

She sat there another minute, staring at the page that now meant something completely different than it had thirty minutes earlier.

She knew where she had gone wrong.

She knew which offer had looked best and which one had actually made the most money.

She knew the business hadn’t drifted into this mess by accident. She had steered it there with a smart-looking question and a clean-looking answer.

She also knew she was late.

The office was waiting. The team was waiting. The day had already started without her.

She closed the laptop, finished the coffee anyway, and before she stood up, she blocked Friday on her calendar.

She knew what Friday was for.

Figuring out how to take what she had just learned and get herself out of this mess.

Here is what Erin saw too late.

She didn’t make a careless mistake. She made a careful one. She pulled her numbers, asked a reasonable question, and got a clean answer that looked like strategy. She followed it for five months. The revenue line confirmed her the whole way.

The problem wasn’t that she used AI. The problem was that she handed it a billing report and asked which number was biggest — when what she actually needed to know was which offer left something real behind after the work was done.

Those are different questions. They get different answers.

A visible metric shows you what’s happening at the surface. A useful metric helps you decide what to do next.

The dashboard was accurate. It just wasn’t useful.

Before you ask AI what to grow, make sure you know what health looks like.

Otherwise, it may help you scale the very thing that is quietly draining you.

Erin didn’t fix the business that morning. She only saw it clearly for the first time. In the next part of this case, we’ll follow what happened next — how she used that clarity to start unwinding the mistake, rebuild the offer mix, and work with AI on a plan to get the business healthy again.

Download the Offer Health Map. It walks you through the same process Erin used: name the decision, separate visible metrics from useful ones, and compare what each offer brings in, what it really costs, and what is actually left.

Continue Erin’s case

• Read the guide for Part 1: Why Your Best-Looking Number May Be Lying to You

• See what Erin did next in Part 2: Knowing the Right Number Didn’t Fix the Business

• Read the Part 2 guide: How to Build a Recovery Plan for a Small Business Using AI

• Continue to the finale: She Had a Recovery Plan. Then Sales Slowed Down.

• Read the final guide: How to Use AI to Review a Plan Before Drift Gets Expensive