Episode 100 - Mass personalisation in CRM is both a quality and throughput problem

Welcome to episode 100 of the Retention Blueprint! Wow big milestone, lots of writing, heres to the next 100! 🎉

Today I want to talk about something I've watched happen in dozens of CRM teams over the years.

Someone pulls up the campaign plan.

They look down the segment list.

And they say: "what we're sending is ok, but if we had the time, another version for that cohort would do a lot better."

Everyone in the room nods, because everyone knows it's true.

And then they send the ok one.

The reason = team capacity and production effort.

You can do more and want to do more, but there are only so many hours in the day.

So you decide what goes in and what gets dropped.

(Even if what gets dropped is costing you impact).

That's the production ceiling and most mature CRM teams have hit. It's why brands like DAZN and SKY have 50 and 100 people working in Retention and CRM, but even then they don't do everything they wish they could.

The problem: the average customer doesn't exist

So if the average customer doesn't exist and you send a communication for the average, it hits no one.

Take two of your customers.

One opens the app twice a week and saves one thing.

The other opens five times a week and saves nothing at all.

They might sit in the same segment.

They might be in the same lifecycle stage.

They might be on the same journey.

They are probably getting the same ‘in-life’ or ‘lapsing’ program.

But one of them is quietly deciding the subscription isn't worth the price, and the other is engaged but behaviourally stuck.

Different problem. Different timing. Completely different message.

So you deliver the version that is half right.

And at scale, approximation means masses of lost opportunity.

There are two ceilings, not one

When people talk about not being able to personalise at scale, they're usually describing one problem. There are two, and they are solved completely differently.

The quality ceiling. You can't produce content that is actually good at volume. So the variants you do ship are thin (Hello $Firstname), because that's all the capacity your campaign or creative team have.

The operational ceiling. And even if you had all the variants, you don't have capacity to execute them all in a quick enough window.

Most teams believe they only have the second issue, but the first is real.

Why generic AI didn't fix the quality ceiling

Almost everyone reading this has used AI to create CRM marketing content.

You put your email brief into ChatGPT, you read what came back, and honestly it is just not good enough.

Or you use a tool which has built-in content generation. Basically, you're using an LLM, and again, the output isn't good enough.

The reason is that LLMs are trained on all of the world's data on the internet, and most advice about email and CRM marketing misses the mark. So when an LLM creates an email, the output is poor.

That's because good CRM copy is specific, and a generic model has none of the specifics.

It doesn't know your lifecycle stages. What counts as good in onboarding is actively wrong in-life.

It doesn't know your behavioural drivers. It has no idea which reason for leaving this particular customer is drifting toward.

It doesn't know your voice, your constraints, or your customer.

It doesnt know behavioural science or how to apply it to your copy in way that is contextual and not pushy.

So it writes the average of everyone else's marketing.

Generic AI gives you infinite mediocrity.

"Good" is not one thing.

This is the bit that gets skipped. Good is different at every stage of the lifecycle for every customer, and the differences are knowable.

Registered to paid. One step, one job. Registration captures data and starts personalising. It does not also try to sell. Most journeys fail here because they attempt three conversions at one touchpoint.

Trial and activation. Get a result fast. Ninety percent of customers who don't use the product in the first ninety days have roughly a ten per cent chance of becoming long-term subscribers. Everything in that window should centre on one first meaningful outcome.

Onboarding. Drive them towards the activity threshold that predicts long tenure. And don't waste the highest open rate you will ever get on account admin. The receipt should orient, not confirm purchase.

In-life. Usage is retention. Show the value visibly, reduce novelty, watch for engagement drops or where perceived value drifts below actual value.

These are moments when leaving becomes the rational choice.

Customer love. "I don't feel valued" is one of the core drivers of churn, and it's the one almost nobody builds a programme for. Even loyalty programs are typically built from a perspective of extraction, not reaffirming how much the brand values its customers.

Lapsing. Signals appear sixty to ninety days before someone leaves. Lighter, earlier interventions always cost less than heavier, later ones.

Underneath it all sits the principle I keep coming back to: companies don't create value; customers realise it.

So if your goal is to help customers realise value. An average communication means you're leaving masses of opportunity on the table.

Each stage asks a different question.

Most teams manage frequency across the base. A contact policy, a cap, an inactivity rule.

And every one of those rules is wrong, because there is no correct base-level frequency. Each customer has their own normal.

A "seven days inactive" trigger fires identically on a customer who has always watched once a week and a customer who has just fallen off a cliff. It is right for neither of them.

Instead, measure time to next meaningful action against that customer's own baseline.

Intervene when their gap widens. Stay away when their behaviour is normal.

Churn is a lagging indicator of emotional detachment.

Drift is the signal.

Time to next meaningful action measures drift and how you intervene before intent to leave hardens.

But for TNMA interventions to be successful, they must be relevant.

So quality is a knowledge problem

Read back over the last two sections. Everything in them is knowable and teachable.

Which is exactly the kind of thing you can give an AI model.

That is what I mean when I talk about a second brain. A knowledge base the AI works from: your retention principles, your strategy and constraints, your brand values, the brand's voice, real-world context like sports results or the weather.

The output is good because of what went in.

And good at volume, not just a few variants, hundreds of variants now tied to a customer's preferences, behaviour, and profile, even tailored to external content like sports results or the weather.

And then it hits the assembly line.

Solve quality, and you hit the second ceiling immediately.

Brief. Copy. Build. QA. Configure.

A CRM ops lead once told me it would take at least two weeks to configure four hundred emails.

That was an honest answer; my agency did it in 8 hours.

Fix throughput without fixing quality, and you don't get better marketing.

You get four hundred thin variants instead of four.

You have industrialised mediocrity, faster.

Fix quality without fixing throughput, and the best campaign of your career sits in a build queue until the moment it was written for has passed, and then it doesn't get sent.

You need both quality and throughput.

And the two are not solved by the same thing, which is why so many teams buy a new tool, feel briefly optimistic, and end up back where they started.

Three questions for your next team meeting

Between the idea and the send, which step takes longest? That's your throughput ceiling, and most people have never actually measured it.

Who decides what good looks like? If the answer lives in one person's head, it cannot scale, with AI or without it.

Is your contact cadence based on each customer's normal, or on a rule for everyone? If you have a fixed inactivity rule, it is wrong for almost everyone it fires on.

One last thing

If you want to keep seeing this in your inbox, just click a link or hit reply with a hello or a question. I read and reply to every message, and it helps signal to your email provider that this is worth delivering. If it isn't for you, unsubscribe with my blessing. I only want to show up where it helps.

Until next time,

Tom

Tom Burrell is a retention strategy consultant with 27 years driving the retention P&L, formerly SVP at DAZN with senior roles at Manchester United and Publicis Groupe. He works with ExCo teams on structural retention strategy and agentic execution across streaming, iGaming, sports and subscription apps. Tom is founder of Retention AI, the Retention Agency for the Age of AI. Visit https://retention-ai.net/ to find out more or book a call.

PS One quick favour, and it genuinely helps: I've just moved The Retention Blueprint to a new sending home. If this landed anywhere but your main inbox, drag it across and mark it as not spam. This will genuinely help the next one reach you.

---