Episode 89 - Agentic CRM 3.0: The Architecture

Did someone forward this to you? If so, click to subscribe

📰 Top Story: Agentic CRM 3.0 - The architecture 

Lately, I’ve been writing about the future of CRM, or really, the now of CRM: Agentic CRM 3.0.

Instead of designing a journey for a segment, you build a system that watches each customer individually: their usage patterns, engagement signals, and the rate at which they’re drifting away from the behaviours that correlate with retention or towards ones that indicate value growth. 

Then the agent intervenes before drift becomes intent.

Not because of a trigger, but because their individual behavioural fingerprint is changing.

The system doesn’t wait for a blunt rule like “they haven’t opened the app in seven days.” It notices that this specific customer is opening less often than they were last week, that session depth is shrinking, that they’ve stopped saving content, and it acts when the drift starts, not when it’s already too late.

The action space is broader, too.

Not just which message to send.

But, which channel.
Which offer.
Whether to re-engage first and discount later.
Whether to flag the account for human review.
Whether to outbound.
Whether multiple messages together are better than a daily sequence.

And it learns 

What works for similar customers informs the next decision. 

The system updates as new outcomes come in and gets better over time.

So what’s the GOAT tool for agentic CRM?

When I dropped this LinkedIn post about this a few weeks ago, that was one of the questions I got in the comments - what’s the GOAT tool for agentic CRM?

Since then, I have had lots of conversations about this exact topic.

And the answer is: there isn’t one single winner.

It’s more multifaceted than that.

Let’s break it down.

Start at the last mile: your execution platforms.

Any CRM or customer engagement platform worth its salt can deliver email, push notifications, in-app messages, and similar customer touches.

Some tools, like Braze, publicly describe AI decisioning built on reinforcement learning and contextual bandits, so the system can choose between options like channel, message, timing, and frequency, then learn from outcomes over time.

Other tools, like Airship, have launched an agentic goal-optimisation layer for campaigns and customer experience.

Optimove also sits in this category, with AI-driven orchestration that helps move customers through journeys in real time as their behaviours change.

These are all operating at the execution layer.

Decisioning outside the Customer Engagement Platform

What if your customer engagement platform doesn’t have reinforcement learning built in, but you still want to move toward agentic CRM?

That’s where tools like Hightouch and Aampe come in.

Hightouch publicly describes its AI decisioning as reinforcement-learning-based, using warehouse data to learn which messages, channels, and timings work best for each customer, while also picking up patterns that generalise across segments.

Because Hightouch is warehouse-native (not a traditional CDP), it makes sense if your data already lives in Snowflake/BigQuery and you want decisioning without data movement.

Aampe is best thought of as a real-time personalisation and learning layer. It observes customer behaviour as events happen, runs experiments in parallel, and updates what content or message should come next based on the latest response signals.

It is used alongside downstream engagement tools rather than replacing them.

The key difference between them is this: Hightouch is batch/warehouse-first (great for comprehensive customer context). Aampe is event-stream-first (great for sub-second reactivity). However, both are used to instruct the execution layer. 

Upstream capability

Then there’s the layer upstream of all that.

Most enterprise brands already have their core data stored in a cloud data platform such as AWS, Azure, or Google Cloud.

That’s where a lot of the raw behavioural data lives, and it’s often where the most important signals start to appear first.

Those platforms also give you the infrastructure to build AI-powered systems on top of that data, including embeddings, model hosting, and custom agents.

In other words, if you feed behavioural events into the cloud layer, you can start detecting drift, value growth, and churn risk earlier in the lifecycle, sometimes before the customer has consciously decided to leave, downgrade, upgrade or deposit. 

That has a significant impact on bonus spend in iGaming, churn prevention in subscription products, and retention orchestration in streaming and media.

So the real question isn’t, “Which tool is the goat?” 

It’s: how do I orchestrate the best outcome for my customers and business? 

In most businesses, that means: 

1. Agents built for your specific context on your cloud data stack  

2. A reinforcement learning layer on top  

What I do 

At my core, I am a retention strategy consultant. I have spent 26 years in the field, with the last decade focused on high-frequency, engagement-led, recurring-revenue businesses where retention & LTV depend on usage. I have been successful at it, generating over $200m incremental value for my clients. 

I now help enterprise and recurring-revenue brands define the strategic architecture for Agentic CRM 3.0: identifying the moments of truth that materially affect retention, translating customer behaviour into actionable signals, and designing the decision framework an agent can use to intervene earlier and more effectively. 

The focus is on improving customer lifetime value, reducing avoidable churn, and creating a more adaptive CRM operating model.

Depending on the organisation, agents are delivered through three models: enabling your internal team, partnering with your existing implementation partner, or leading a fully managed end-to-end program (as part of my agency model: retention-ai).

Want to learn more? 

Message me back, and I’ll share a recording of a session I ran on this recently. 

Work in iGaming? 

If you work in iGaming CRM,  I'm running a short agentic AI for iGaming retention session in late May on Google Meet. I'll walk through how agentic AI can help iGaming teams:

1. Detect player behavioural drift/growth before defection or deposit signals appear, reducing bonus dependency and increasing customer value.

2. Specific iGaming technical build and use cases

Reply to this email to apply to join. 

Until next time, 

Tom