Ogres, Onions, and AI: There Is More To This Than People Think

Ogres, Onions, and AI: There Is More To This Than People Think

"For your information, there's a lot more to ogres than people think. Ogres are like onions. Layers. Onions have layers. Ogres have layers."

No, this quote was not justification for posting an algorithm-beating picture of Shrek on LinkedIn although if it gets the engagement, we are not judging. It is here because it is a more honest description of how AI actually works in a CRM context than most of what gets written about it.

Most of what gets written is either breathless enthusiasm or existential concern. Neither is particularly useful if you are trying to decide how to deploy it in a real programme, with real data, and real consequences when something goes wrong.

So here is my actual view, after using it as a working tool rather than a talking point.

The Promise Was Always There. The Access Was Not.

Ten years ago, the capability that AI now puts within reach of a mid-market DTC brand existed. It was just ring-fenced inside organisations with data science teams, seven-figure technology budgets, and the runway to build proprietary models over multi-year timelines.

Behavioural segmentation at genuine scale. Predictive churn modelling that updated in real time. Personalisation that went beyond first name and last category browsed. These were not theoretical concepts. They were operational realities for a small number of businesses that could afford to build them. Everyone else was running batch-and-blast and calling it a lifecycle programme.

That gap has closed. Not entirely, and not without caveats, but meaningfully. A lean CRM team with the right foundations in place can now access analytical capability and creative velocity that would have been genuinely out of reach five years ago. That is a structural shift in what is competitively possible, and most brands have not yet worked out what to do with it.

Where It Actually Earns Its Keep

I want to be specific about this, because the generic answer - "AI helps with personalisation and efficiency" - is not wrong, it is just too vague to be actionable.

Interrogating large data sets. What previously required a data analyst, a working knowledge of SQL, and several days of back-and-forth can now be surfaced in a conversation. Behavioural patterns across hundreds of thousands of customers, cohort performance trends, churn indicators, product affinity clusters - the speed at which you can move from a question to an answer has changed fundamentally. That alone has genuine commercial value for any team trying to make faster, better-informed decisions without adding headcount.

Prototyping before committing resource. This is where I find it most consistently useful. Testing a new segmentation approach before building it. Modelling what a revised lifecycle might look like before touching a live flow. Generating ten creative directions for a campaign in the time it would previously have taken to develop two. AI compresses the distance between an idea and something you can actually evaluate - and that compression is worth a significant amount in team time and strategic velocity.

Surfacing where the opportunity sits. Rather than guessing which levers to pull, AI can identify where the genuine gaps are. Which customer cohorts are underperforming relative to their behavioural profile. Which sequences have the weakest conversion and why. Where lifecycle gaps are actively costing retention. The analysis that used to take weeks can now take hours - which means the time saved goes into acting on what it finds, rather than just finding it.

The Part That Does Not Make It Into the Case Studies

AI gets things wrong. Not occasionally, and not always obviously. It gets things wrong confidently, fluently, and sometimes in ways that only become apparent once something has already been sent to 200k customers.

In a CRM context, the stakes of activation are categorically different from the stakes of prototyping. A mis-calibrated segment is not a draft that needs another pass - it is a deliverability problem, a brand experience problem, and depending on what was in it, potentially a compliance problem. These are not theoretical risks. They are the predictable consequence of removing human judgement from the final step.

This is why I would argue that quality control is not a drag on an AI-enabled programme. It is the architecture. Every output that moves toward activation needs a competent set of eyes that understands what the data is actually saying, what the downstream implications are, and what good looks like in the context of that specific brand and that specific audience.

The machine finds patterns in what it has been given. It does not understand your customer relationships, your commercial context, or the difference between a technically correct output and a strategically sound one. That distinction sits with the person operating it - and it always will.

You Still Have To Drive The Car

There is a version of the AI conversation that treats it as something that happens to a business. A capability that arrives, removes friction, and delivers outcomes without requiring the business to change how it thinks or what it knows.

That version is not accurate, and betting on it is expensive.

The teams getting the most from AI in CRM are not the ones who adopted it earliest. They are the ones whose people understand segmentation logic, attribution methodology, lifecycle design, and what a healthy programme actually looks like - and are now using AI to operate at a level that would previously have required a significantly larger team. The tool accelerates the capable. It does not replace the capability.

A driver who has never sat behind the wheel does not become one because the car has GPS. The navigation is useful. It does not teach you what to do when the road ends.

The Foundation Still Determines Everything

There is a layer that sits between your customer data and whatever AI tooling you are considering, and it is the layer most brands have not built properly. Programme architecture. Flow logic. Suppression rules. Segment definitions. Attribution methodology. A coherent view of what retained looks like in your data versus reactivatable versus gone.

If that layer is broken, AI makes it worse faster. You are not solving the problem - you are scaling it, with better copy.

Before asking what AI can do for your programme, I would ask these questions instead;

• What percentage of your list is genuinely active?

• Do your flows reflect your actual customer lifecycle, or is it the default configuration you set up and never revisited?

• Do you have a clean view of your retention rate by cohort, or are you working from blended figures that hide more than they reveal?

• Can you produce those numbers on demand, or does it take a week and several spreadsheets?

Most brands cannot answer these cleanly. That is not a reason to layer technology on top of the gap. It is a reason to close it first.

What This Actually Changes

The honest case for AI in CRM is not that it solves the hard problems. It is that it gives teams with the right foundations access to a level of analytical depth and operational speed that used to require far more resource to reach.

That is not a small thing. For a DTC brand operating with a lean team, the ability to move from question to insight to activation in hours rather than weeks is a genuine competitive advantage - provided the judgement sitting above it is sound, and the programme underneath it is built.

Used that way, it is the closest thing to enterprise-level CRM capability that most mid-market brands have ever had access to. The dreams we had ten years ago about what a properly data-driven programme would look like - they are largely buildable now. The constraint is no longer the technology. It is whether the business has the foundations and the expertise to direct it.

Shrek was right. The layers matter. And in this case, they have to be in the right order.

VALIX works with DTC and subscription brands to build the CRM foundations that make tools like AI actually perform — and provides the strategic and technical expertise to make sure what gets activated is worth activating. If you want to understand what your programme looks like underneath the surface, start with an audit.