This session had a very TMM energy, and we loved it: folks arriving from gardening, hot water bottles in hand, and a chat that was somehow both deeply driven by nuanced questioning and briefly derailed by gherkins (watch the replay for that one).
Underneath the warmth, the topic was a serious one: most marketers feel the pace of AI change is hard to keep up with — and yet it’s getting harder to opt out. Kirsty and Rich’s job today was to turn “AI is huge and scary” into something you can actually use.
They did that by doing three things very well:
- demystifying the language (so you know what people mean when they say “agentic”)
- putting strategy before tools (so you don’t end up automating nonsense faster)
- showing what “personalisation” should look like when it’s not creepy
A note from the speakers: you didn’t need to have done the School of AI programme to follow along – today stood on its own, with a condensed set of lessons and actions.
Table of Contents
- 1) Where most marketers are right now
- 2) The AI landscape in three layers
- 3) The ladder from chatbot to strategic teammate
- 4) Where AI drives ROI
- 5) The segmentation trap (and what “true personalisation” actually means)
- 6) Case study: Currys and Black Friday
- 7) What happens to marketers in all this?
- 8) Governance and privacy
- 9) Five actions to take this week
- The core shift (the bit to write on a post-it)
1) Where most marketers are right now
They kicked off with a simple pulse-check: are you a sceptic, explorer, builder, or pro?
Most people landed in explorer or builder territory — using tools a bit, testing, trying to work out what’s real vs hype. The reassurance was clear: you’re not behind. The pace is genuinely mad, and most organisations are still early-stage.
The key message here wasn’t “catch up”. It was: start where you are, and build literacy steadily.
2) The AI landscape in three layers
Rich offered a very usable way to think about AI capability — three layers, each with a different job.
1) Machine learning = the optimiser
Not creating from scratch. Improving performance by spotting patterns in data.
Example: predicting the best send time for an email based on behaviour.
2) Agentic AI = the creator
Generates content and takes actions toward an outcome, but needs guardrails.
Example: creating personalised push notifications at scale within brand rules.
3) Autonomous marketing = the manager
Runs and adapts campaigns end-to-end in real time (content, schedule, targeting).
Example: adjusting a journey as performance shifts, without waiting for your next report.
A useful line to steal: “Use machine learning to optimise, use agents to build.”
3) The ladder from chatbot to strategic teammate
Most of us still picture AI as a text box. They argued we’re moving beyond that — and your role shifts as AI matures.
Step 1: Reactive assistance
Drafting, summarising, first-pass thinking. You prompt; it responds.
You’re the instructor.
Step 2: Rule-based automation
“If this, then that” workflows. Efficient, but rigid.
You’re the architect.
Step 3: Agentic AI
You set a goal (“reduce churn in this segment”), and the system coordinates actions to achieve it, adjusting as it learns.
You’re the strategist.
The bit they kept coming back to: even at the agent level, humans set the North Star. AI can scale execution, but it can’t choose what should matter.
4) Where AI drives ROI
The claim was simple: AI is most valuable where generic feels like spam.
They used a clear contrast:
- A generic newsletter is often fine.
- A generic offer at a high-intent moment (right product, right time, right context) is a missed opportunity.
AI can bridge the gap between:
- real-time signals (behaviour, location, inventory, intent)
- and real-time content decisions (message, product, timing, channel)
The point wasn’t “send more”. It was: react faster and more accurately than humans can at scale.
5) The segmentation trap (and what “true personalisation” actually means)
Kirsty’s framing here was strong: we treat people like averages, but nobody is average.
Segmentation is useful, but it’s static. Life isn’t.
A segment can’t tell you someone is:
- buying a gift
- having a bad service moment
- suddenly price-sensitive
- browsing with high intent right now
They made a helpful distinction:
- Automation without empathy leads to classic “why are you advertising what I just bought?” moments.
- The goal isn’t automated marketing.
- The goal is augmented marketing: AI finds patterns; marketers keep it human.
A good mental model they used:
AI personalisation should feel like service, not surveillance.
6) Case study: Currys and Black Friday
Rich shared an example from Currys using Movable Ink’s AI (DaVinci) during Black Friday.
The challenge: high volume, high noise, and a classic risk — generic emails don’t just get ignored, they create fatigue.
The shift: away from manual guessing and batch-and-blast, towards AI making lots of individual decisions at scale: message, product, subject line, timing — based on real-time signals.
The outcomes they highlighted:
- 49% lift in click-through rate
- 30% increase in conversion rate
- and the kicker: reduced send volume while making more money (because targeting focused on higher intent)
The underlying lesson: relevance can beat volume.
7) What happens to marketers in all this?
They didn’t pretend job anxiety doesn’t exist. They handled it with a grounded take:
AI is a world-class sous chef.
It can prep at scale. It can’t decide what’s on the menu. It doesn’t know what “on brand” tastes like.
So the marketer’s value becomes more concentrated in:
- strategy and prioritisation
- brand integrity and tone
- cultural context (especially market nuance)
- relationship building and trust
- innovation (AI iterates; humans invent)
A line worth keeping: “The future of AI is more human, not less.”
8) Governance and privacy
The Q&A landed on a practical truth: we’re not at “set and forget”.
If you’re building your own agents or systems, the advice was:
- set clear guardrails (what “good” looks like, and what’s off-limits)
- give feedback, iteratively, as outputs appear
- expect maintenance (tools and models change quickly)
On privacy, the most practical guidance was:
- If in doubt, don’t put it in.
- Don’t put sensitive data into free tools.
- Make sure consent and trust stay central: “helpful vs creepy” is a useful test.
9) Five actions to take this week
They closed with a simple “don’t panic, do this” list.
1) Start small
Pick one problem. One workflow. One win.
2) Audit your existing tech stack
A lot of tools already have AI features baked in (subject line generators, summarisation, automation helpers). Start there.
3) Set guardrails
Define what “on brand” looks like, and what isn’t. Tell AI what you want and what you don’t.
4) Keep the customer test running
For every personalisation effort: helpful or creepy?
5) Build AI literacy
Learn what’s real value vs AI-for-AI’s-sake. Stay curious, keep it practical, and use education as your unfair advantage.
The core shift (the bit to write on a post-it)
Stop asking: “How many people signed up?”
Start asking: “What did they do — and what happened next?”
Same energy applies beyond webinars too. As Joe said at the end: the most useful outcome of a session like this is that it gives you a new way to think — and a foundation for better decisions afterwards.