Building a stack for the dominant forces of data and AI

At the September MarTech Conference, martech contributor Milton Hwang moderated a panel on what it takes to modernize the stack for a world reshaped by data and AI. Joining Hwang were: 

• Florian Delval, product marketing principal, AI Data Cloud for Marketing, at Snowflake
• Natalie Jackson, director of demand generation at CBIZ
• Angela Vega, director, capabilities and operations,  Expedia Group. 

The group tackled data quality, identity, consent and whether a “center of the stack” even exists anymore.

The hardest problem first: complexity at the speed of change

For Delval, the biggest issue is pace and complexity. Privacy rules are shifting, AI is accelerating and stacks are under strain. “The complexity comes from the volume of change happening right now,” he said. Success requires reorganizing data and integrations, not just bolting on tools.

Jackson faces the realities of a diversified B2B firm. CBIZ markets more than 300 services to wildly different audiences, from global tax buyers to “Kyle the kayak guide.” With data spread across systems, “a lot of the work is data transformation — how it looks in System A versus System B, and whether those systems can talk.” Her question: “Is anybody’s data good enough for AI to deliver the magic it promises?”

At Expedia Group, Vega argued that perfect data is a myth but, “we should be able to get to a level of good enough.” In a two-sided marketplace, granularity matters: a girls’ trip is different from a family trip. Vega stressed the need for data about our data — metadata, tagging and a semantic layer that AI (and marketers) can use without hand-built connections.

Data quality is mostly a work in progress

When asked to describe the state of their data in an audience poll, most viewers picked “work in progress.”

Jackson offered a familiar failure: getting a “sorry you couldn’t make it” email after she had attended an event. Her advice: when you know your inputs are shaky, soften the claims. “Say, ‘Thanks for your interest — here’s a recap,’” instead of doubling down on an error.

Dig deeper: 10 of the best insights from the September MarTech Conference

Delval warned against chasing purity for its own sake. “At some point the cost of more cleanliness won’t move the business.” Instead, look to unstructured data — like calls or reviews — that AI can finally mine. Vega agreed: consumers increasingly tell brands what they want in free text, and marketers must learn to capture and act on that intent.

What’s the ‘center’ of the stack?

The second audience poll of the session asked what sits at the stack’s center: CRM, MAP, data warehouse or CDP. CRM led, with warehouses close behind.

Delval sees a shift: cloud data platforms now run operational workloads, not just analytics. Many teams are leaving data in the warehouse and layering tools on top. Smaller firms, he noted, may still anchor on CRM.

Jackson highlighted the identity gap in B2B. CRMs tie everything to a work email, but when buyers change jobs they look like new people. “The best next customer is the one you already had who moved,” she said. Warehouses and CDPs can unify profiles beyond a single inbox.

Vega went contrarian: “There is no center. Think puzzle pieces — if I’m missing some, the picture’s incomplete.” Centralization once dominated, but modular, productized data domains are now healthier. Still, she agreed pipelines shouldn’t sprawl: clear ownership and canonical entities are key.

Delval summarized the balance: centralize governance, not necessarily tools.

Identity, duplicates and consent

Duplicates plague every enterprise. Jackson called them “a very real problem” that CRMs and MAPs can’t fix alone. Warehouses and CDPs can help, but only if teams agree on common denominators like domains or addresses.

Vega described the consumer version: Expedia unified logins across brands, but channel fragmentation still creates headaches (app vs. browser, logged-in vs. anonymous). Relevance requires connection, she said, but connection must respect privacy.

Dig deeper: How AI and data activation deliver unforgettable customer experiences

That’s where consent belongs “by design,” Delval argued. “Collect consent alongside the data point — understand what you’re allowed to do with it now and later.” Jackson, speaking as a consumer, was blunt: “Don’t text my cell while I’m at soccer practice.” If marketers want deeper access, they must provide reciprocal value.

AI’s near-term reality: Value with a clear ‘why’

The panel was optimistic about AI’s near-term role — if scoped correctly. Vega sees wins in semantic layers that let AI interpret intent without brittle schemas. Jackson wants AI to help marketers, not hallucinate for them. Delval pointed to unstructured sources as the new frontier. All agreed: don’t chase new use cases until you’ve nailed consent, identity, and measurement.

What they’ve added that mattered

In a lightning round, panelists shared stack additions that made a difference:

• Jackson: “Account-based marketing software is a game changer for B2B.”
• Vega: Adding Salesforce Data Cloud with Marketing Cloud has been transformative.
• Delval: Snowflake recently added Cortex/Intelligence for natural-language access to data — “we’re starting to see real value.”
  

Different tools, same principle: each solved a targeted business problem.

Five moves the panelists say to make now

1. Define “good enough” by use case. Write acceptance criteria for quality and consent tied to specific outcomes.
2. Collect consent at capture. Store permissions with the data point.
3. Establish a canonical identity. Decide which keys (login, email, loyalty ID) matter and who owns them.
4. Productize metadata. Build a semantic layer for content and events so both humans and AI can use it.
5. Pilot unstructured-to-action. Mine transcripts or reviews for themes and route insights weekly.

The bottom line

Modern stacks aren’t monuments; they’re modular systems that evolve as fast as customers. The winners will govern data centrally while activating it flexibly — with consent captured up front, identity resolved enough to be useful, and semantics rich enough for both humans and AI.

As the panelists showed, unforgettable experiences don’t come from AI alone. They come from relevance plus respect — and while AI can scale relevance, only culture and consent can guarantee respect.

Listen to an audio recap of the September 2025 MarTech Conference

Use the the player below to listen to an AI-generated recap.