Why Fanatics Stopped Defining Audiences and Let Machine Learning Take Over
For years, the playbook for performance advertising was simple: define your most valuable customer, figure out where that person spends time, and buy media accordingly. It sounded logical. It was also, as Fanatics Sportsbook recently discovered, fundamentally limited.
The sports betting operator decided to try something different. Instead of starting with a hypothesis about who their audience should be, they let machine learning figure it out. The results were striking: a 19% increase in projected customer lifetime value (LTV) from campaigns run using ad tech company Cognitiv’s full-funnel CTV solution.
This isn’t just an incremental improvement. It represents a fundamental shift in how advertisers should think about performance marketing in an era of machine intelligence.
The Old Way: Guessing the Audience
Andy Magnes, director of paid advertising at Fanatics Betting & Gaming, described the old approach candidly. Marketers would define audience segments based on demographics, behavior, or assumptions about who their customers were. Then they’d optimize campaigns for efficiency — lowest cost per acquisition, highest click-through rates — within those predefined boxes.
The problem? Those boxes were built on assumptions. Marketers were effectively narrowing their own view of the addressable market based on what they thought was true, not what the data actually showed.
“You can definitely get ahead of yourself on targeting,” Magnes told Digiday, reflecting on the shift.
The New Way: Optimizing for Outcomes
Fanatics’ work with Cognitiv changed the equation entirely. Instead of defining audiences and optimizing for efficiency within those constraints, they defined the business outcome they wanted — higher customer lifetime value — and let machine learning find the people who delivered it.
Cognitiv’s approach trains custom machine-learning models toward advertiser-defined outcomes. The models don’t start with assumptions about who the customer is. They start with patterns in the data — betting behaviors, engagement signals, response patterns — and build behavioral segments from the ground up.
This is where AudienceGPT comes in. Cognitiv’s product lets marketers describe their target audience in natural language — “high-value sports bettors who engage during live events” — and the machine learning systems build the behavioral segments that match those descriptions. These segments can then be activated across multiple channels, including online video and social platforms.
“We just started modeling on value,” Magnes explained. The new approach allowed Fanatics to establish a clearer link between ad spend and the actual customer value being generated — a connection that’s notoriously difficult to make in traditional performance advertising.
Letting Algorithms Define the Audience
Jeremy Fain, CEO of Cognitiv, described the paradigm shift succinctly: “Fanatics tells us who the high-value people are, and we go out and find more of those people.”
The crucial insight here is that algorithms can identify traits associated with valuable customers that human marketers would never think to look for. Patterns emerge in the data — combinations of behaviors, timing signals, content preferences — that point to high-LTV segments that no demographic profile would capture.
Of course, Fain also noted that “at the top of the marketing funnel, you do need a hypothesis as to the target customer-base, and these prompts end up being that hypothesis.” AudienceGPT’s natural language prompts serve as the starting point — but it’s the machine learning that refines and expands the picture.
CTV: The Channel That Demands a Better Approach
The shift was particularly significant for connected TV (CTV), which has become an increasingly important channel for sports betting advertising. CTV presents unique challenges for traditional audience targeting — it’s a lean-back medium, measurement is fragmented, and the line between brand awareness and performance is blurry.
By optimizing directly for customer value signals rather than predefined audience segments, Fanatics was able to treat CTV as a true performance channel rather than a branding exercise. The 19% LTV improvement came specifically from Cognitiv’s full-funnel CTV solution, demonstrating that outcome-based optimization works even in traditionally “upper-funnel” channels.
The Broader Lesson for Performance Marketers
Fanatics’ experience points to a broader evolution in digital advertising. Machine learning programs are increasingly being tasked not just with finding audiences, but with determining who those audiences should be in the first place. The marketer’s role shifts from audience definer to outcome definer — from “I know who my customer is” to “I know what a valuable customer looks like to my business.”
That’s a fundamentally different skill set. It requires letting go of assumptions, trusting data over intuition, and being willing to discover that your best customers might not look anything like you expected them to.
For Fanatics, that leap of faith paid off. And the 19% LTV improvement is just the beginning.
Lessons for Every Performance Marketer
The Fanatics case study offers several actionable takeaways for advertisers across verticals. First, outcome-based optimization works. Defining the business result you want — whether that’s LTV, retention rate, or average order value — and letting machine learning find the audiences that deliver it is demonstrably more effective than starting with audience assumptions.
Second, natural language interfaces for audience definition (like AudienceGPT) lower the barrier to entry. Marketers don’t need to become data scientists to benefit from ML-powered audience discovery. They just need to describe what a good customer looks like.
Third, CTV and other “upper funnel” channels can be treated as performance channels when the right optimization framework is in place. The assumption that CTV is purely a branding medium is outdated.
Finally, the key insight is that machine learning doesn’t replace marketer expertise — it augments it. The marketer still defines the outcome. The machine finds the path. Together, they outperform either working alone.
