Episode 96 - You don't have an AI problem, you have a maturity problem

Welcome to episode 96 of the Retention Blueprint.

I've written a lot about agentic CRM lately. So today I want to come at it from a different angle.

Not "what it is." But "where are you, really, and why does everyone want to skip ahead?"

Because the most expensive mistake I'm watching right now is a team starting an agentic retention project on Monday and trying to finish by Friday. On top of workflows they designed in 2019.

There are five levels of retention AI maturity. Not marketing tiers. Actual, sequential capability.

Level 1 is rules and reporting. "Email them if they haven't logged in for 7 days." Static segments, lagging indicators. Generative AI lives here too, drafting your subject lines and images. It makes the old paradigm faster. It doesn't change it.

Level 2 is predictive targeting. Churn scores, propensity models. Useful, but reactive by design, and the signal is always late, because a churn score only fires once the customer is already halfway out the door. What has changed at Level 2 is execution. Tools like Claude Code let you ingest entire journeys into your ESP in hours or days, not months. Ping me if you want to learn how.

Level 3 is where the paradigm actually shifts. Embeddings. Instead of a rule or a score, you build a behavioural fingerprint for every customer, and you detect drift weeks before a churn signal ever appears. Then you intervene before churn becomes conscious.

Level 4 is assisted agents. AI that recommends the right intervention, human in the loop where it matters: high-value customers, regulatory edge cases, complex pricing. This is where one-to-one CRM starts to emerge, but at this stage it's nascent. You're working in one life stage, with one agent, checking everything.

Level 5 is autonomous agents. Moment-specific, self-improving, operating inside pre-approved guardrails. A fleet that scales your execution while protecting margin and LTV. Even here, your agents are still selecting from pre-approved content and pre-approved rules. The difference is that the human is now only in the loop on edge cases, and your agentic capability is rolled out across every life stage.

Here's the trap.

Thinking there's an AI magic switch (or tool you can buy) that lets you jump from Level 1 to Level 5 by Friday.

Because when you jump from Level 1 or 2 straight to Level 5, you build autonomous agents on late signals and old workflows. They break trust. They alienate the customers you were trying to keep.

So what does good actually look like?

Netflix and Uber. They already run this, and neither uses a visual journey builder. No canvas. No dragging boxes around a flow.

Instead they build a big menu of approved content, and let agents read real-time signals and pick the right thing for each person, every time.

Two principles make it governable:

The agent selects, it does not author. It generates and scores variants, then chooses from what's already been approved. It never free-writes to your customer.

It carries pointers, not payloads. The decision is separated from the send. The content is assembled at the moment of send, from a library you've signed off.

So here's the reframe I want to leave you with.

Agentic CRM is not a product you buy.

It's a level you climb to.

And the blocker is almost never the AI.

It's your data, and it's the level you're honestly at today.

The good news: you don't need perfect data to start.

You need to stop trying to leapfrog.

Secure the level below, then build the one above.

So, genuinely, where are you? Level 1 dressed up as Level 4? Solid at 2 and eyeing 3?

Hit reply and tell me the number. I read every one, and I'll tell you the fastest honest route to the next level.

And if you want to go deeper: in September I'm running a few agentic CRM boot camps, following the invite-only group I ran in July. Five or six brands per session, max, and we go deep on agentic CRM in your specific context. They're complimentary for Heads of CRM, CRM Directors, or their bosses (CMO, CCO). If you'd like a spot, hit reply and I’ll reserve one for you.

Until next time,

Tom