In 2025, 38% of CIOs named “monetizing company data” as their top business initiative: more than double the share from the previous year, according to the State of the CIO survey. Yet for most companies, the reality of using data to generate insight, let alone revenue, remains elusive.
A recent interview with Tomás Puig, CEO of Alembic, on NVIDIA’s AI Podcast offers a window into how a business can tackle the problem with causal AI and proprietary data infrastructure.
Rethinking the Role of Data in Marketing
One of the central claims made by Puig is that the next wave of corporate profit will come not from new data sources or public models, but from proprietary datasets. “All the alpha, all the profit that will exist in corporations in the next while will all come from private data sets,” Puig said.
There is reason to take this view seriously. As large models converge in performance, unique data becomes one of the few remaining levers for competitive advantage. Puig argues that brands with rich behavioral data, such as Disney, with its mix of streaming, theme park, and merchandise touchpoints, are better positioned to generate net-new insights. Recent moves by Disney to integrate data across its platforms via tools like Disney Compass reinforce this direction, aiming to reduce fragmentation and improve measurement.
That said, turning internal data into a reliable source of business intelligence is far from straightforward. Many firms lack the infrastructure, governance, or modeling sophistication to operationalize their own data at scale. The mere presence of unique data does not guarantee insight, and the risk of overfitting to niche datasets, especially in high-variance consumer environments, should not be ignored.
Causal Inference and the Limits of Retrospective Analysis
One of Alembic’s technical differentiators is its emphasis on causal inference: an attempt to move from correlation to cause-effect relationships in marketing.
The company uses a mix of spiking neural networks (SNNs) and custom-built simulations to identify how sequences of events (e.g., watching an Olympic broadcast, then searching for flights, then booking a ticket) contribute to business outcomes. According to Puig, this structure allows them to map chain reactions and calculate “optimal lag” between media exposure and conversion.
But while the theory is sound, causal inference remains difficult to apply in noisy, real-world marketing environments. Issues like unobserved confounding variables, poor data quality, and the complexity of user journeys often limit how precise these models can be. Businesses considering this approach should pair any causal modeling effort with rigorous testing and cross-validation, rather than assuming inference is equivalent to certainty.
There’s also a question of accessibility. Alembic’s system is built on custom hardware and proprietary algorithms.
While technically impressive, this raises questions about whether such methods are replicable, or even necessary, for most organizations. In many cases, simpler tools, such as uplift modeling or Bayesian A/B testing, may be more pragmatic for teams without specialized infrastructure or budgets.
Moving Past the Dashboard, and the Hype
Puig critiques the industry’s reliance on dashboards, arguing that they too often serve as endpoints rather than catalysts for action. “Nobody wakes up in the morning and says, ‘you know what I want in life? Another dashboard,’” he said. Instead, Alembic delivers daily “intelligence briefings” designed to contextualize data around business goals, using LLMs to summarize insights while reserving decision logic for more robust statistical models.
The takeaway is not necessarily to emulate Alembic’s approach wholesale, but to interrogate the usefulness of how insights are delivered. Are they actionable? Do they reduce uncertainty? Do they map to decisions someone is actually empowered to make? If not, even the most advanced models risk becoming another layer of noise.
A Measured View on Personalization and Creative Automation
Perhaps most refreshingly, Puig pushes back against the tech industry’s fixation on personalization. While AI can personalize content down to the individual, he suggests that marketing value often lies in shared experiences. like watching a gold medal ceremony or attending a live event. “Human beings love building tribes and community,” he said. “The most memorable experiences are usually collective ones.”
This runs counter to the dominant narrative in MarTech, which tends to frame personalization as a universal good. It’s worth taking seriously. Over-targeting can erode brand coherence and miss the emotional resonance that comes from cultural moments or shared narratives. That said, personalization does remain critical for downstream optimization—particularly in commerce or service flows. Puig’s own view concedes this balance, noting that brands should focus on personalization only after engagement has been established.
For marketers evaluating AI tools, Puig offers some accessible guidance. Beginners should use tools like ChatGPT to plan interactively, not just to generate copy. More advanced users might train AI systems to track technical papers or expert voices. While the advice is sensible, it’s also common across the field—and raises a larger point: Alembic’s most differentiated contributions likely won’t scale to most companies as-is. The real value lies in understanding the direction, not copying the stack.
