Ask any marketing team what it wants to do with AI and you’ll get a long list. Ask what it has actually shipped, and the list gets short very quickly. The reason usually isn’t the model. It’s the plumbing underneath it.
That was the core argument of a September MarTech Conference panel titled “Built for yesterday: Why your data architecture can’t keep up with AI,” which looked at how legacy enterprise data stacks quietly cap the speed AI is supposed to unlock — and why buying another platform rarely fixes it.
The session featured Courtney Adams, head of content and product marketing at MessageGears; Jacqueline Freedman, CEO and founder of Monarch Advisory Partners; and Mike Maynard, chairman of Napier Partnership Limited, moderated by Kevin Haag, senior vice president of data strategy at Qualify Digital.
Old data means reactive marketing
Adams framed the problem in blunt operational terms: if your data is an hour or more old, she said, your execution is reactive by default. You can hire strong talent and design elegant workflows, but a slow pipeline neutralises both.
The tell, she noted, is what happens when someone asks to activate a new customer attribute. If capturing one data point needs a data science sprint, a custom integration and complex SQL, a simple request becomes a multi-month project.
And swapping marketing automation vendors doesn’t help. If the architecture beneath stays rigid, the same latency simply moves to the new logo on the invoice. Freedman’s line on this was direct: a shiny new tool doesn’t fix problems that live outside it.
The gap is access, not ambition
Most marketing teams can reach only a sliver of the customer data their organisation actually holds. That constraint limits everything downstream — real-time journey triggers, granular segmentation, multi-touch orchestration. Teams design a brilliant campaign, then discover the stack can’t execute it.
AI widens the gap rather than closing it, because it expands what marketers can imagine without expanding what the infrastructure can support. Maynard summed it up with a phrase he uses at his agency: ideas are easy, execution is difficult.
In B2B, the panel argued, the challenge is feeding models enough context to speak intelligently to a buying committee. Starve them of context and AI just produces more generic messaging, faster.
Freedman’s advice to leadership: audit your motivation before your budget. Are you solving a business problem, streamlining a process, or answering board pressure to “do AI”? Maynard was equally unsentimental — throwing AI at something for its own sake, he said, is a waste of time.
Feedback loops are the forgotten half
Adams’ point that your AI is only as strong as the information it has has an underrated corollary. Teams routinely pull data out of the warehouse and never push campaign interaction data back in. The result: marketing, BI and data science teams working from conflicting records instead of sharpening one shared view of the customer.
Composable or monolith? It depends on your headcount
Freedman made the case for modular stacks, comparing a monolithic marketing cloud to an old house where every small renovation uncovers a new structural problem. Her framing: do you want a best-in-class stack, or a movable monolith? Modularity lets you replace one component as AI vendors evolve, without taking the whole system offline.
But she was careful about the limits. AI can’t fix bad wiring, she warned — it will just make bad processes go wrong much faster.
Maynard offered the counterweight. Composable suits enterprises with real engineering capacity. Smaller B2B teams often can’t maintain dozens of point-solution integrations, and an all-in-one suite with “good enough” features is the more practical call.
What to do this quarter
The panel converged on an audit-first approach that doesn’t require a multi-year migration:
- Map the data footprint. Every repository holding customer data, who owns it, and whether a genuine single customer view exists.
- Find the broken loops. Duplicate records and unlinked systems where campaign responses never return to the warehouse.
- Prioritise high-value signals. Maynard’s example: knowing how long a buyer intends to keep a vehicle beats dozens of low-value behavioural metrics.
Their closing advice was similarly grounded. Adams: don’t boil the ocean — prove AI activation on one campaign first. Freedman: walk your own customer journey end to end, from signup to post-purchase support. Maynard: ignore shiny capabilities and stay fixed on customer needs and the data required to serve them.
Source: MarTech
