Recapping the Marketecture Podcast: John Hoctor on Agentic Media Analytics

In a recent appearance on Episode 168 of the Marketecture podcast, Newton Research CEO & co-founder John Hoctor joined hosts Ari Paparo and Eric Franchi to discuss where marketing technology is headed in the age of agentic AI.

At Newton Research, we've been building AI agents for three years and like to think of ourselves as "agent-native." From the beginning, our co-founders had a clear mission: to build a virtual data scientist that could take on complex marketing analytics tasks. Here's a look at what we're building and how we're giving marketing teams unlimited access to powerful analytics.

‍Beyond the Chatbot

Most AI tools today give you a chat interface and call it a day. Our approach is different. Newton Research deploys a contained system that lives inside a customer's own infrastructure — AWS, GCP, Azure, Snowflake, or Databricks — rather than pulling their data out of it.

A few principles guide how we've built this:

Data stays where it lives. Our agents connect directly to a client's existing data, which eliminates the security risks and delays that come with moving sensitive enterprise data around.

No guessing games. Our agents show every step they took to reach an answer — think of it like a spreadsheet where you can click any cell and see the formula behind it. You always know exactly how a result was calculated, which matters a lot in enterprise settings.

Context from day one. By connecting to a brand's internal data from the start, our agents quickly develop a clear, accurate picture of that specific data environment.

Giving AI a "Master's Degree" in Marketing Science

General-purpose AI isn't enough for enterprise work. When a standard AI model tackles a complex marketing problem, it tends to make inconsistent choices in its methods — the kind that can quietly lead to costly budget decisions.

We train our agents the way you'd train a junior analyst: not just on how to run the numbers, but on why certain approaches are used and when. The result is agents that follow consistent, enterprise-grade methods rather than improvising their way through an analysis.

That specialized training shows up in areas like Marketing Mix Modeling (MMM), Incrementality Testing, and Lookalike Modeling.

Multi-Agent Workflows: Moving Past the Blank Prompt

Starting every analysis from scratch is already feeling outdated. We're moving toward structured, multi-step workflows — where users can kick off sophisticated, multi-step analyses without writing a single line of code.

The practical upside is significant. Tasks that used to take analysts days — pulling data from multiple sources, combining tables, building a model — now take minutes. That frees human analysts to focus on what they're actually good at: applying strategic judgment and making the calls that move budgets and shift bids.

Making Incrementality Testing Less of a Headache

Incrementality testing has always been one of those things that's valuable in theory but painful in practice. The setup is complex, the measurement is messy, and most teams run it far less often than they should as a result.

Our agents simplify the entire process — from identifying test and control groups to measuring final results. When the friction drops, teams run more tests. And when teams run tests continuously rather than once or twice a year, they start building a real informational edge over competitors.

What's Next: Causal Modeling and "Primordial Soup"

Traditional MMM relies on mathematical frameworks that are, frankly, about 30 years old. We're moving toward causal modeling — using modern AI architecture and GPU processing to handle datasets that were previously too expensive and slow to analyze at scale.

We sometimes describe this new frontier as "primordial soup." Take customer journey data, call center logs, and brand awareness signals, feed them into a powerful neural network together, and you start finding cause-and-effect patterns that were completely invisible when those datasets lived in separate departmental silos.

We're also addressing a gap in the market with "MMM Light" — a flexible approach that builds on existing models and feeds in fresh data like incrementality results. Brands can reallocate budgets in real-time without paying for a full model rebuild every time.

The Bottom Line

A lot of organizations are still dabbling with AI — burning money on generic prompts and calling it a strategy. That's not a path to reliable analytics.

The brands and agencies that will pull ahead are the ones building on trusted, intelligent platforms that actually understand marketing science. The technology is ready. The question is whether the industry is willing to move past the prompt.

‍