Question 1

Is AI in marketing just building 'faster horses'?

Accepted Answer

Subx CEO Jonathan Harris argues most marketing AI speeds up existing processes rather than replacing them, like building faster horses while someone else builds a car. Dynata's Steven Millman adds that you can only build the car once the equivalent of the internal combustion engine exists, and that many companies use AI as a performative 'machine that goes bing'. Both say the goal should be the outcome, such as the right ad to the right person at the right time, not optimising each waypoint.

Question 2

What is the difference between artificial general intelligence and superintelligence?

Accepted Answer

Artificial general intelligence (AGI) would match or exceed humans across the broad range of cognitive tasks, with perfect recall and far more data, and could be told to build better versions of itself. Artificial superintelligence would be thousands or millions of times smarter than humans and think in ways we cannot comprehend. Steven Millman expects AGI within about ten years, though he stresses it would be a tool with no feelings, and ethics lie with the people who build it.

Question 3

What is a 'walled garden' large language model and why would a business want one?

Accepted Answer

Dynata's Steven Millman describes two meanings. One is limiting the data a model is trained on, such as your own technical documentation, so it is less likely to stray, hallucinate or repeat bias from the open internet. The other is locking down your own instance so outside changes to the model cannot break your applications and your private or personal data never goes back to the model's owner.

Question 4

What would a personal AI assistant mean for marketers?

Accepted Answer

Steven Millman predicts people will build AI personas of themselves that follow them across devices, recommend news and products, and filter what reaches them. Dom notes this flips marketing from push to pull. Both raise the question of how content will make money, since answer engines and assistants would bypass the sites that currently earn from search and advertising.

Question 5

What is quantum computing and why does it matter for AI?

Accepted Answer

Steven Millman explains that ordinary computers use binary transistors that are either on or off, while quantum computers use qubits at the atomic scale that can exist in more than two states, potentially making them many orders of magnitude faster. Experts he speaks to expect real quantum computers in three to five years, probably rented like cloud services. They would let AI run massive computations fast enough to make artificial general intelligence practical.

Question 6

How could quantum-powered AI change marketing?

Accepted Answer

Steven expects AI to move from generating rough creative prompts, which make up about 40% of current agency AI use, to producing usable creative, possibly generated live around themes humans set. Real-time processing could enable hyper-personalisation and continuous rather than periodic market research. Dom notes quantum computing may also be far more energy efficient, but stresses that marketers must protect customer privacy and trust.

Question 7

Will regulation slow down AI's disruption of marketing?

Accepted Answer

Steven thinks not in time: laws are written by people who do not fully understand the technology and lag years behind it. He says the EU AI Act largely addresses threats from three years earlier, such as facial recognition. Jonathan uses GDPR as an example of a well-meant law that has not made people feel their data is protected, and predicts the data will ultimately win over the law.

Question 8

Could AI take over most marketing work?

Accepted Answer

Steven says Sam Altman's claim that AI could do most of what marketers do sounds high but is not out of the ballpark once AGI arrives, because most time in knowledge businesses goes on editing, admin and other non-creative tasks. He worries safety work is being cut as AI companies race each other, and that AI's lack of transparency will help bad actors get around regulation. He sees the film WALL-E, where AI takes away human agency, as a more realistic risk than Terminator.

Question 9

Why do data inputs matter so much for AI in marketing?

Accepted Answer

Jonathan says AI is essentially maths that transforms data, so poor inputs produce poor outputs. Marketing data comes from many channels and teams with competing objectives, no single point of decision and no view of cause and effect, so its quality is inconsistent. Steven warns about 'synthetic data' tools that generate extra survey responses resembling the ones you have, which add no real representativeness.

Question 10

Are large language models good at prediction?

Accepted Answer

No. Steven's rule is 'it's a language model, stupid': LLMs are not maths models and are poor at forecasting. The AI doing most of the practical marketing work today is predictive analytics, recommendation systems, marketing automation and machine learning clustering, which use historical data to find patterns humans might miss.

Question 11

Could AI make websites and apps irrelevant?

Accepted Answer

Jonathan is investing in the idea that within about five years websites and apps may not exist in their current form. Chat-style interfaces let people describe the outcome they want, such as the right barbecue for their family and location, making the presentation layer irrelevant. This would also give marketers a far deeper understanding of customers than inferring intent from page visits, where typical conversion rates are only 1% to 2%.

Question 12

How should companies build teams to deliver AI projects in marketing?

Accepted Answer

Steven recommends clear objectives that every team member can state in one sentence, genuinely cross-functional teams rather than blind handoffs, shared data access and mutual education between data scientists and marketers. Build prototypes and get feedback in loops rather than building the whole thing first, keep privacy and ethics in view throughout, and celebrate progress along the way.

Question 13

How should businesses prepare their data and tools for AI-driven marketing?

Accepted Answer

Jonathan advises removing subjective, business-centred barriers from the customer journey, such as arbitrarily gated content, so data reflects real customer behaviour rather than internal bias. Steven says to choose the right tool rather than the newest: machine learning is well suited to testable problems like conversion optimisation, while generative AI is harder to validate. Jonathan's team chose reinforcement learning for recommendations because it is dynamic and has a feedback loop.

Question 14

Will AI favour big agency networks or small independents?

Accepted Answer

Steven thinks it will look more like the dotcom boom than a wave of consolidation. Big players like Publicis are investing heavily, mostly to find new uses for off-the-shelf tools rather than build their own models, but AI also lets small firms do what big firms can at low cost. Dom describes a CMO who gave her team five hours to find useful AI tools, leading to 17 adoptions, one of which saved £40,000 from her annual budget in the first week.

Question 15

How will advanced AI change market research?

Accepted Answer

Steven says AI is good at 'columns, not rows': predicting how rich, real personas would answer new questions, but poor at inventing entirely synthetic respondents. That could let researchers run shorter surveys, get higher response rates and extrapolate the rest, or estimate answers to questions they forgot to ask. Real survey work will still be needed, because AI cannot read minds and depends on recent, high-quality data.

AI and marketing technology

11 episodes on ai and marketing technology

Questions about ai and marketing technology