What subscriptions actually teach us about AI, data and CRM

We sat down recently with Craig Niven, who runs FILDI, a subscriptions growth consultancy working with Shopify DTC brands globally. Craig has 15 years of deep experience at Sky, BT and News UK. He knows the subscription model from the inside.

We covered a lot of ground. What came out of it was less a conversation about subscriptions specifically and more a shared view on what it actually takes to make data and AI work in a commercial context. The subscription model just surfaces these tensions more clearly than most.

Six themes stood out.

1. Most businesses are solving the wrong problem

Craig described something we both encounter constantly: clients who arrive with a solution already formed. A platform they want to implement. A channel they want to activate. A tool they’ve been sold.

The first job, before anything else, is to qualify or disqualify that solution. And to do that, you have to actually understand the problem underneath it.

“By the time a metric becomes visible in a dashboard, the damage is often already done. The focus should be on the predictors of those outcomes — not the outcomes themselves."

This is a discipline issue as much as a capability issue. Organisations under commercial pressure reach for the nearest available fix. The nearest fix is usually a communication, a discount, a campaign. None of those solve the underlying problem and just defer it.

In practice, most businesses already have the data they need to understand their real problems. It is usually poorly organised, fragmented across systems, or simply not being looked at in the right way. That is where the diagnosis starts.

2. Inertia isn't a strategy (but there are two very different things being called by that name)

The inertia question in subscriptions is genuinely interesting. Most businesses know they rely on it but few are honest about what that actually means.

Craig made a distinction worth noting. There are two types of inertia operating simultaneously in any subscriber base:

• Purposeful non-disruption: the subscriber is satisfied, receiving what they want and does not need to hear from you. Contacting them unnecessarily prompts a cancellation review.
• Passive inertia: the subscriber has forgotten they are paying or can't be bothered to cancel. This is fragile by definition.

Most brands can't tell the difference between the two and that's the problem. A communication strategy built on the assumption that all inertia is passive will over-communicate to the satisfied and under-serve the disengaged. Both errors are costly.

“The goal is not to remind people they are subscribed. It is to make them glad they are. Those two things require completely different decisions.”

Regulatory change is also coming. The DMCC Act — delayed from spring to autumn 2026 — will require greater ongoing transparency for subscribers. Brands that have been relying on passive inertia as a model need a different strategy, and they need it before the legislation forces the issue.

3. Personalisation is a proxy for a relationship you don't actually have

Personalisation is one of the most overused and least examined terms in marketing. Craig raised the right question: what data should actually inform how you treat a subscriber?

My view is that personalisation in its current form is mostly a proxy for relevance. It is what brands do when they do not have a real relationship. “Hi [first name]” isn't personalisation, it's a mail merge.

The subscriber context is different. Someone paying you every month has already made a financial commitment. That commitment generates behavioural signals continuously — what they use, how often, when they engage, when they go quiet. That signal is more valuable than almost anything a brand could buy externally.

“The question for a subscription brand is not whether to personalise. It is whether you are using the data the relationship is already generating, and whether you are acting on it in a way that reflects what you actually know about that individual.”

Most brands are not doing this. They collect subscriber data and then treat those customers identically to cold prospects. The commitment has been made and then ignored. Agentic AI changes what is possible here — not because it replaces the relationship, but because it can operationalise what you know about a subscriber at a scale no human team could manage.

4. AI isn't the replacement. It's the enabler. But that only matters if the brief is right.

Craig put this clearly: AI is only as good as the question you ask it. If you don't have the domain expertise to evaluate the output, you will take a poor answer at face value and act on it. The tool does not protect you from a bad brief.

This applies to both LLMs used for analysis and to agentic systems used for execution. An agent pointed at reducing churn will reduce churn. It may do so by eroding trust, suppressing communication in ways that accelerate disengagement or optimising a metric that doesn't actually reflect relationship health.

Unless you have defined what a good relationship looks like — not just what a retained customer looks like — the agent is flying blind.

“The brief is the critical skill. It always was. Before AI, you briefed a member of your team. You had the experience to evaluate whether they had done the right thing. AI accelerates the loop significantly. It does not change what the loop requires.”

Speed is the real benefit. The ability to go through a diagnostic cycle, a content iteration, a modelling exercise — in hours rather than weeks — means more value can be delivered within any given engagement. That is only true if the direction of travel is correct from the start.

5. Most retention programmes are treating the symptom, not the problem

Craig was direct on this: there are agencies claiming to do retention work that are simply sending emails. That is a symptom response, not a solution.

Churn has root causes that vary by subscriber cohort and business model: 

• Product overwhelm — subscribers receiving more than they can consume before the next delivery.

• Payment failures — which may reflect broader financial pressure on the subscriber, not a preference change.

• Pricing sensitivity at renewal when introductory rates expire.

• Boredom.

Each of these requires a different intervention.

Treating all churn with a win-back communication is the equivalent of prescribing the same medication for every diagnosis. It will work for some and miss entirely for others. The aggregate metric may look stable while the underlying problem compounds.

“Retention is 5-7x cheaper than acquisition. Which makes it worth asking honestly: is your retention programme actually retaining people, or is it measuring that some came back?”

The harder work is identifying the leading indicators of churn before it appears as a number. What changes in subscriber behaviour precede cancellation in your specific model? That varies. Most brands have not done the diagnostic work to find out. They are tracking the outcome and reacting to it, not tracking the drivers and preventing it.

6. You can't automate a process you don't fully understand

The automation conversation in subscriptions tends to focus on individual touchpoints: onboarding sequences, win-back flows, payment failure journeys. Each one is automated in isolation and, in isolation, it works.

The problem is that the subscriber experiences all of them simultaneously. They do not know which team or system sent which message. They experience the relationship as a whole. When automations are designed in silos, the cumulative effect on that experience is invisible until it becomes a churn number.

Craig described a project where senior leadership wanted to automate parts of a process, but the people actually running the process had entirely different views of what it involved. Neither was wrong. They were looking at different parts of the same system. Without an end-to-end view, automation entrenches the existing fragmentation rather than resolving it.

“The best approach is to design around the customer experience — what is the subscription journey from the subscriber’s perspective? That should define the processes, not the internal business structure. Otherwise you are imposing your organisational logic on the customer relationship.”

A unified data layer is the prerequisite for this. You can't make intelligent decisions across a subscriber journey if your data is siloed by channel, team or platform. A Single Customer View isn't a data project, it'ss the foundation on which coherent automation becomes possible.

The thread running through all of this

Subscriptions surface something that is true across most of marketing but is harder to ignore in a recurring revenue model: the intelligence is only as good as the foundation underneath it.

That means:

• Data that's connected and accessible.

• Processes designed around the customer.

• Objectives that define what a good outcome actually looks like — not just what a retention metric looks like.

• And questions that are precise enough to generate answers worth acting on.