Every August, preseason football returns and I do something my family finds mildly obsessive: I start rebuilding my predictive model for the coming season.
I've been doing some version of this for years. I call the current iteration the Four-Pass Model — a framework that evaluates NFL matchups across four variables: defensive depth sustainability, offensive line durability, coaching conservatism under pressure, and explosive play vulnerability. Layered on top is what I call a Fragility Index — a composite score that identifies teams statistically likely to blow leads late in games — and a Red Flag Companion Analysis that surfaces hidden risk factors traditional power rankings completely miss.
People ask me: why not just use AI to pick games?
I'll admit — my relationship with football goes deeper than the analytics. I played in high school and into college, until a Dean delivered a more urgent prediction than anything my model has ever produced: bring your grades up, or the game ends for you. I did. And given that the NFL draft was never seriously considering my file, it was probably the right call.
But playing the game gave me something the data alone never could — an inside understanding of how football actually feels under pressure: what changes in a huddle when a team that should win starts to feel the weight of almost winning, what a tired offensive lineman looks like in the fourth quarter before the stats show it, how a coaching staff's psychology shifts when a comfortable lead starts to feel fragile.
That foundation — not the spreadsheets, not the algorithms — is what the Four-Pass Model is built on.
And the answer to why I don't just "use AI to pick games" reveals something important about how AI actually works. It's the same lessons I applied when helping architect IntellEmotion™, Montage Labs' multimodal AI platform that solves for the missing link in marketing - and the key driver in buying decisions - emotion.
The Framework Came First. The AI Came Second.
When I built the Four-Pass Model, I wasn't starting with data. I was starting with decades of watching football through the eyes of someone who thinks analytically about performance under pressure — and who once stood in that huddle himself.
The insight that a team with a Fragility Index of 6 or higher — combined with high reversibility scores — is statistically entering upset territory in a way that traditional power rankings miss? That didn't come from a spreadsheet. It came from the field.
It came from watching coaching decisions change in the fourth quarter when a team that should win starts to feel the pressure of almost winning. From recognizing that offensive line fatigue is a late-game multiplier that nobody's box score captures. From noticing that certain head coaches, when ahead by 10 in the third quarter, call plays as if they're behind.
That pattern recognition is what gets encoded into the model. The model then executes the framework at a speed and consistency no human could match. But the framework itself? That's human. That's experience. That's the thing AI cannot manufacture on its own.
Now replace "football" with "marketing." The story is identical.
Thirty Years in Marketing Taught Me What Algorithms Miss
I spent the better part of three decades in corporate marketing — as a C-level consultant, as an executive, across industries and company sizes. I was an early practitioner of predictive analytics applied to marketing decisions back when "data-driven marketing" was still a novel phrase. I was a two-time top-10 finalist in the Direct Marketing Association's annual Analytic Challenge — an international competition — during an era when applied analytics was genuinely pioneering work.
What did three decades in the trenches teach me that no algorithm inherently knows?
It taught me that emotion is not a soft variable. It is the variable. The entire architecture of consumer behavior — why someone clicks, buys, returns, refers — runs on an emotional substrate that rational analysis consistently underweights. Every great marketer I've ever known understood this intuitively. Most analytics frameworks ignored it entirely because it was hard to measure. That doesn't make it less real. It makes the measurement problem more important.
It taught me that context is not background noise. Context is signal. The same advertisement placed against content that creates tension, unease, or dissonance lands differently than the same advertisement placed against content that creates warmth, excitement, or aspiration. Brand safety isn't just about avoiding offensive content — it's about understanding the emotional environment your brand is entering and whether that environment amplifies or undermines what you're trying to communicate. This is a fundamentally different — and far more sophisticated — problem than most "brand safety" solutions are designed to solve.
It taught me that the primary question is never "how much did we spend?" It's always "what did it produce?" CPM is a pricing metric. Business lift — new customers acquired, revenue generated, basket size increased, trial conversion achieved — is the only metric that actually matters to a CMO who has to answer to a CFO. Any analytics system that leads with impressions and ends with clicks has fundamentally misunderstood what marketing is for.
These aren't opinions. These are thirty years of hard-won conclusions.
And they are encoded into the architecture of IntellEmotion.
What IntellEmotion Actually Is
IntellEmotion is a multimodal AI platform that analyzes streaming video content in real time — not just what's being said, but what's being felt — to determine the optimal contextual environment for brand placement. It scores content across multiple emotional and contextual dimensions to produce the IE Score: a composite signal that tells advertisers not just what a piece of content is about, but what emotional state it creates in the viewer.
This is not a technology problem that someone figured out and then looked for a marketing application. It's a marketing problem — one I've understood for decades — that we then built the technology to solve.
The decision to weight emotion in the IE Score composite didn't come from a model. It came from thirty years of knowing that emotional resonance is the most underweighted and most impactful variable in advertising effectiveness. The decision to make business lift the primary output — not CPM, not viewability — came from thirty years of sitting across the table from CMOs and watching what questions they actually ask when the campaign is over. The decision to build multilingual and model-agnostic capability into the core architecture came from understanding that the streaming market is global, that no single AI provider will dominate it, and that a platform that locks clients into one model becomes obsolete before it scales.
None of those decisions are in any AI training dataset. They're in my head, accumulated over a career. They're now in the architecture of the platform.
The Parallel That Keeps Me Honest
Here's why I find the NFL model useful as a reference point — not just as a hobby, but as a professional discipline.
When my NFL framework misses — when a team I predicted to win loses by 20 — I don't blame the model. I examine the assumptions encoded in the framework. Was the defensive depth variable properly weighted for a dome team playing outdoors in January? Was the coaching conservatism score capturing this specific coordinator's tendencies or just generic conservative-coach behavior?
This is exactly the analytical discipline required when an AI-powered marketing application produces a result that doesn't match expectation. Not "the AI was wrong" — but "what assumption in the framework needs refinement? Where did the signal fail to capture the reality?"
The AI doesn't know when to question itself. You have to build that habit into the system — and into the humans overseeing it.
That discipline — the habit of examining the framework rather than blaming the output — is what separates someone who has built and iterated models over decades from someone who deployed a platform last quarter.
The Question Every Brand Should Be Asking
If you're a brand, a media company, or an agency evaluating AI-powered advertising technology right now, here's the single most important question you can ask any vendor:
What domain expertise is encoded in your model's decision logic, and who put it there?
Not "which LLM do you use." Not "what's your CPM." Not "how many publishers are in your network."
Who built the framework that tells the system what matters? What is their evidence? How many years of domain experience shaped the weighting of those variables? And — critically — how does the system get smarter when it's wrong?
The answers to those questions will tell you more about whether a platform will actually work for your brand than any demo ever could.
The best AI marketing tools aren't built by AI companies who learned marketing. They're built by marketers who learned AI.
The same way the best NFL prediction models aren't built by data scientists who watched a few games. They're built by people who played them.