Applied AI and the new marketing revolution: optimise and prophesise.

Google and Meta didn’t win the last 15 years of marketing because their household data was better than AOL, or because their creative options were better than Yahoo, or because their pockets were deeper than Microsoft. They won because of the AI tech they invented.

Finding consumer insights not observable to humans, while testing messaging at scale — deep-learning neural networks and reinforcement learning form an intelligent data-driven system where the results of everything make the system smarter.

As I discussed in Part 1, marketing has always been about two questions: Who are we selling what to? What should we tell them about it?

The platforms won by inventing and industrialising the AI to solve both ends of this equation — and added the ever-important performance economics for “how much is this worth?” Meta, as the follower, hired from Google to hammer these auctionomics.

Neural networks and deep learning

Deep learning handles the who / what by finding structure in billions of messy signals: engagement, searches, time, events, behaviours, transactions, profiles, performance metrics, and formulas.

Not blunt SQL rules like customers who spent over $500 on jeans.

Not the cardboard personas agencies still trot out like Suzy Moonshine.

Segments derived from advanced GPU computation across thousands of dimensions, discovering affinities no human could observe. They look and perform like nothing our two-dimensional monkey brain can derive.

That’s how a casual shopper in New Jersey and a heavy buyer in Texas both end up in a hidden seasonal splurges segment, or how late-night weekday conversions reveal an at-risk impulse buyer.

Deep learning is segmentation reborn: from rules to exploration, from data minimisation to data maximisation.

Once the who / what is clear, reinforcement learning takes over the creative message and the response / delivery optimisation.

Reinforcement learning

Target a user.

Test an offer.

Adjust bids.

Swap creative.

Every impression you make can be a test with measurable results. If it works, exploit it; if it doesn’t, keep exploring and learning. That’s called optimisation.

RL is not especially new. Multi-arm bandits have been a backbone of adtech ad servers for 15 years, but RL has had a lot of focus in AI the past few years with agentic development that enables serverless RL, multi-agents, transfer learning, and improved exploration.

And here’s the kicker: putting deep learning and reinforcement learning together creates closed-loop optimisation.

It is a beautiful thing how these different applications of AI work together, because it relates directly to ROI. The sharper the who / what, the faster RL converges on the right message and offer. Knowing ahead of time that this optimisation is going to happen within thresholds related to volume and performance is how the platforms generate profits. They control the efficiency. By controlling their margins, they control your CPA.

Performance Max and Advantage+ are deep learning and reinforcement learning running at scale, faster than any human team could ever iterate. They’ve fuelled every moment of relevance you’ve ever felt with their content — like getting a cashmere sweater from a brand you’d never heard of in your feed, that you order seconds after being served the ad. NNs and RL power a combined $5 trillion in valuation for Google and Meta. Imagine what these technologies will do for your little billion-dollar brand.

The new martech: data plus AI in your cloud

While Google and Meta were locking advertisers inside black boxes, they handed out ways to build your own boxes. Google open-sourced TensorFlow. Meta gave away PyTorch. NVIDIA released CUDA and RAPIDS. These are the pipes and processing that make all of this run. Open-sourcing wasn’t entirely an act of altruism — it’s how the platforms ensure another ten years of growth — but the building blocks of their dominance ended up on GitHub.

We can bristle at Amazon, Meta, and Google and debate if AI will make people irrelevant, but the takeaway shouldn’t be resentment. It should be recognition. And action.

Their moats may be impenetrable, but new castles can now be built in the clouds.

The data warehouse is yours. Now, for the first time, the AI can be too.

A decade ago, Snowflake and Databricks did not exist. Five years ago, you didn’t have elastic GPUs. Two years ago, you could barely evaluate how deep-learning models were chewing on data. Just a year ago, it was impossibly hard to smooth the rough edges of data workflows.

Today, marketers and data teams have it all available to them. And every day it is improving, getting bigger, faster, and smarter. You can build or buy marketing AI products that use the same DNA but are tuned for your business, your data, your KPIs. Not Google’s or Meta’s.

It’s already happening:

• A casino matched mid-tier players with relevant offer strategies that drove more revenue than some VIPs.
• A streaming service cut churn by discovering then targeting segments of cross-device, cross-genre subscribers that were at risk, with relevant content.
• A QSR chain instantly boosted weekday lunch profits by targeting relevant coupons to only those who needed the nudge.

Optimising outcomes everywhere

It’s no accident that this revolution fuels data sharing. The platforms are starting to share more data back to your Snowflake, Databricks, or BigQuery data warehouse. This data can fuel NN and RL applications that keep you spending — ideally because you are hitting your cost, volume, and objective goals. Data sharing is the choke point the platforms keep for the price of inventing this stuff. But every brand has an ROI number where their marketing spend will always be on, and that’s a very desirable metric for everyone.

As this new marketing revolution becomes clearer in 2026 to your board and CEO, your first-party data becomes the most valuable asset of your company. There is no doubt more companies than ever will achieve value and profit from their data. It’s computationally impossible not to improve with this newfound intelligence. More companies will also start their own data businesses, as more data becomes a competitive advantage in every market, and will make partners, vendors, and services more revenues and profits as well.

Which leads to the last question about marketing AI I want to talk about. If every brand can spin up its own closed-loop matching machine, how does that change the very shape of customer strategy, creativity, analytics, and competition? That’s where we’ll go in Part 3.