Estimated reading time: 5 minutes · Last updated:
AI is changing how marketing and technology teams collaborate by making shared data and clear definitions the operational precondition for useful machine learning. Leaders from Monks, Zoom, Saatva, Ipsos and Christie’s International Real Estate told an ADWEEK House Advertising HQ group chat that teams must consolidate data, align terms and treat schema as an axis of distribution to keep models reliable and customers well served. This account is based on an ADWEEK House Advertising HQ group chat, as first reported by ADWEEK. The result: faster iteration and more predictable campaigns where data is joined across engineering, product and marketing — and where human judgment preserves brand voice.
“I think there’s a lot more acceptance and willingness to go out and do something that would be unexpected for a brand in your specific category or industry,”
Whitney Magnuson, head of brand and media at Zoom
Key takeaways
- Data and definitions: James Stephens, EVP of North America at Monks, says consolidating data and a shared dictionary is the first step toward predictable marketing.
- Discovery shifts: Whitney Magnuson, head of brand and media at Zoom, explained that consumer research often no longer lands on brand sites, changing how demand is modelled.
- Distribution control: Joe McCambley, CMO at Saatva, said large language models now recommend products directly and that accurate schema is critical to how they surface offers.
- Human oversight: Nikolett Vilmos, head of marketing and AI transformation at Christie’s International Real Estate for Dubai, warned that personalization must remain on‑brand and that judgment is essential.
Table of contents
Why shared data and a common dictionary are now infrastructure
Marketing leaders in the discussion framed data consolidation as an operational prerequisite for using AI beyond point experiments. When teams keep separate sources and define core outcomes differently — for example, what counts as “conversion” or “revenue” — models cannot learn in a way that supports cross‑department decisions. Joe McCambley described this as the first barrier: agree terms, map sources, then let AI iterate.
Practical work here is twofold. First, engineering and product teams must expose the datasets that drive customer experience — transaction, catalogue, pricing and performance data — so they can be joined. Second, stakeholders must document how each team uses key terms and metrics. McCambley said that process can take a year or two, but it is what turns isolated signals into repeatable forecasts.
Making that infrastructure visible changes project rhythms: short, technical release cycles meet longer commercial cadences, so overcommunication and shared dashboards replace stove‑pipe reporting. Hiring criteria shift too; Eric Matisoff argued for candidates who combine data fluency with “taste” — the ability to recognise quality in analysis and creative work.
How AI is reshaping discovery, affiliates and distribution
Panelists described a concrete change in how consumers find products: machine learning and LLMs increasingly surface recommendations without routing users to brand pages. Whitney Magnuson noted that brands must explain lower site traffic to stakeholders because consumer research often arrives already shaped by non‑brand sources. That shift forces marketers to rethink funnel attribution and where they measure intent.
Joe McCambley gave a specific distribution example: when LLMs recommend products directly, affiliate referrals can be bypassed unless product metadata is complete and consistent. He said brands should publish comprehensive schema — product details, pricing, FAQ and reviews — and ensure it matches site content, because mismatches reduce trust in downstream models.
Manny Flores of Ipsos pointed out that reach matters less than relevance in this environment. Algorithms build audiences from real‑time behaviour, collapsing insight, media planning and distribution into a single feedback loop. That makes fast iteration and reliable ground truth data more valuable than ever.
Where human judgment still decides brand advantage
Despite efficiency gains from AI, every speaker emphasised the continuing role of people in steering brands. Nikolett Vilmos noted that luxury and heritage brands need curated memories: models can generate thousands of personalised versions but a human must keep them on brand. She argued that personalization is a tool, not a substitute for brand stewardship.
Manny Flores framed predictability as bounded: faster learning tightens forecasts, yet human behaviour contains qualitative shifts models fail to foresee. That humility matters in planning and in setting expectations for executive stakeholders who want precise answers from AI. Teams therefore need not just data engineers but decision owners who can interpret model outputs against market context.
Zoom’s focus on investing in memorable experiences illustrates the trade: data and AI improve targeting and measurement, but experience and creative distinctiveness remain the durable advantage. Hiring for taste and cross‑discipline communication — product, engineering and brand — helps organisations convert AI capabilities into lasting customer relationships.
How this shift could play out
The case for
- Joined datasets and a shared dictionary shorten feedback loops, making forecasts more actionable and campaigns easier to scale across channels.
- Accurate, platform‑aligned schema and metadata reduce distribution friction when LLMs surface product recommendations, preserving sales volume even as clickthroughs decline.
The case against
- Organisations that fail to align definitions or expose datasets risk feeding models inconsistent inputs, producing unreliable forecasts and poorer decision making.
- An overreliance on automated personalisation without human review can dilute brand distinctiveness, especially for premium and heritage brands.
What to be careful about
- Data mismatch between published schema and page content can cause LLMs or recommendation engines to de‑prioritise or mistrust a brand’s offers.
- Siloed definitions of revenue or conversion prevent models from learning across departments and undermine prediction quality.
- Rushing to automate customer touchpoints without brand review can create off‑brand messaging at scale.
The bottom line
AI alters the ways marketing and technology teams must work together by making joined data and shared definitions the practical starting point for performance. The panel’s examples show that improved predictability follows from plumbing and governance — publishing reliable schema, documenting metric usage and hiring people who combine technical skill with taste. At the same time, human oversight remains the gatekeeper for brand memory and creative distinction. Organisations that align infrastructure, measurement and decision ownership will be able to use AI to speed iteration without losing control of who their customers think they are.
What to watch
- Watch for brands to publish or update product schema and metadata across platforms; no date has been set.
- Watch for cross‑team launches that publish a shared measurement dictionary and joined dashboards; no date has been set.
Frequently asked questions
What is the first operational step teams must take to make AI useful for marketing?
Start by joining datasets and agreeing definitions: Joe McCambley recommends creating a shared dictionary that maps terms like “conversion” and “revenue” across teams. He said that aligning terms and documenting how each department uses them can take a year or two but is necessary before models can deliver reliable forecasts.
How should brands adapt to lower site traffic while preserving demand measurement?
Accept alternate discovery paths and instrument intent elsewhere: Whitney Magnuson explained that consumers may research without arriving on brand pages, so teams must capture signals from partners, marketplaces and aggregated behaviour and adjust conversion models accordingly.
When does automation risk harming the brand?
Automation risks brand dilution when personalised outputs are not reviewed for tone and fit: Nikolett Vilmos warned that even if AI can generate thousands of tailored variations, a human must ensure they remain on brand, especially for luxury and heritage names.
Related reading
