The Messy Middle of AI Adoption

When Marketing Has to Let Go of the Script

‍

Most service businesses don’t struggle with AI because the model “isn’t good enough.”

‍

They struggle because AI forces a deeper change: moving from scripted marketing sequences to adaptive conversations with customers.

‍

And that’s where things get uncomfortable.

‍

Right now we’re onboarding an AI system for a company that handles a high volume of inbound and outbound leads. The team was already stretched thin trying to respond to new inquiries while keeping warm prospects engaged.

‍

The goal seemed straightforward:
let the AI handle early conversations, qualify interest, and book meetings when someone is ready.

‍

But once we got into implementation, we hit what I think of as the messy middle of AI adoption.

‍

Where the real work actually happens

‍

The first step wasn’t technology.

‍

It was mapping the workflow.

‍

We had to define:

• How the AI should interact with leads
• When to use email vs text
• The tone of the messages (human, informal, slightly messy)
• When a lead is “ready” for a human conversation
• The meeting workflow — booking, confirmations, reminders

‍

A surprising amount of effort went into the micro-details.

‍

For example, if someone books a meeting, the system should:

• confirm the meeting
• remind them the day before
• send a check-in the morning of
• send a final reminder 10 minutes before

‍

Not because automation is cool — but because show rates improve dramatically when this happens consistently.

‍

Then we had another problem.

‍

The company had hundreds of existing leads that had gone quiet over the past few months.

‍

So we uploaded those leads into the system along with their conversation histories.

‍

Now the AI could read the prior exchanges and decide how to re-engage.

‍

The first message looked something like:

“Hey — it’s been a little while since we last spoke. I wanted to check in and see if continuing the conversation still makes sense.”

‍

Simple. Human. Not a campaign.

‍

And that’s when the real tension appeared.

‍

The control problem

‍

The marketing team had a completely reasonable instinct.

‍

They wanted to design drip sequences.

Message 1
Message 2
Message 3
‍

Send every X days.

‍

Predictable. Controlled. Approved.

‍

But the whole point of the system was the opposite.

‍

‍

Instead of predetermined messages, the AI would:

• read the conversation history
• interpret the lead’s situation
• ask relevant follow-up questions
• respond naturally

‍

In other words, have a conversation.

‍

And that’s where teams start to feel uneasy.

‍

Because scripted marketing sequences are not just tactics.

‍

They are risk controls.

They ensure the brand voice is consistent.
They prevent off-message communication.
They protect the founder from surprises.

‍

When AI moves to dynamic conversations, those controls disappear — unless you replace them with something better.

‍

AI changes the unit of work

‍

Traditional marketing automation is built around campaigns.

‍

AI systems are built around conversations.

‍

That sounds like a small shift.

‍

It isn’t.

‍

Instead of deciding:

“Which email gets sent next?”

‍

You’re deciding:

“How should the system think about this lead?”

‍

The AI has to decide things like:

• Is this person curious or serious?
• Do they have an urgent need?
• Are they a good fit for the business?
• Should we push for a meeting, or keep nurturing?

‍

Good salespeople do this instinctively.

‍

Now the system is trying to do it too.

‍

And that means it will say things you didn’t explicitly script.

‍

Guardrails beat scripts

‍

The breakthrough moment with teams usually comes when they realize:

‍

You don’t need to script every message.

‍

You need to define guardrails.

Things like:

• what the AI is allowed to promise
• what it should never say
• when it must escalate to a human
• what qualifies someone for a meeting

‍

For example:

‍

The AI can:

• answer questions
• clarify the problem
• offer a discovery call

‍

But it cannot:

 • commit to pricing
• guarantee timelines
• confirm scope

‍

Those rules let the system operate safely but flexibly.

‍

Marketing still protects the brand.

‍

Sales still controls qualification.

‍

Operations still protects delivery.

‍

The AI simply handles the volume of conversations in between.

‍

Why the messy middle matters

‍

Right now this system is still running human-in-the-loop.

‍

Every conversation can be reviewed.
The team can step in anytime.

‍

And something interesting is happening.

‍

The same people who initially wanted strict drip campaigns are starting to see the advantage of dynamic conversations.

‍

Leads respond more naturally.

‍

The exchanges feel more human.

‍

And the team spends less time sending follow-ups and more time talking to people who are actually ready.

‍

The technology was never the hard part.

‍

The hard part was learning to trust a different operating model.

‍

AI adoption isn’t a rollout. It’s a learning loop.

‍

Most SMBs think AI adoption looks like:

1. Pick a tool
2. Turn it on
3. Watch it work

‍

In reality it looks more like:

1. Map the workflow
2. Define guardrails
3. Run with humans supervising
4. Review conversations
5. refine the system
   
   

‍

Over time, the AI behaves less like a script and more like a junior team member with very high throughput.

‍

And that’s when the leverage starts to appear.

‍

The real question founders should ask

‍

If you’re thinking about AI in customer-facing workflows, the question isn’t:

“Which tool should we use?”

‍

It’s:

“What rules would make us comfortable letting a system talk to our leads?”

‍

Because once you answer that, the rest of the system becomes much easier to build.