I'm AI Practice Leader at OmniMedia Solutions Group. My work spans 30+ years and four technology revolutions across media, manufacturing, distribution, retail, financial services, and pharma, plus a couple of startups and exits. What I bring to a room is pattern recognition. I've watched the same AI adoption challenges play out across all of those industries. I wrote a book about it called The AI Alchemist. The short version: the technology isn't the hard part. The human and operational layer is.

The reality is that most pilots are scoped against the part of the work that demos well, not the part that actually drives the outcome. And that's what many of the technology vendors are famous for: build a great demo, close you quickly, and leave the operational complexity for your team to figure out. The data underneath is often messy. The decision rights are unclear. Agreement on what success looks like is usually informal. So the pilot ships, the demo is impressive, and then nothing moves. Six months later it gets relabeled or quietly shelved.

I see this pattern in every industry, not just M&E, but media has a particular flavor of it. Your data is fragmented across decades of acquisitions, your audience is moving faster than your measurement frameworks, and your business model is being disrupted by new entrants. That combination makes for a hard environment to run successful pilots in if you haven't first done the foundational work underneath.

Most AI conversations collapse very different things into one phrase. It helps to separate four distinct layers.

The data work is still underway at most M&E companies. Three years of investment in unifying systems, identity resolution, content metadata, and audience modeling, and that work is still far from finished, largely due to business and IT funding constraints. But what I see playing out in parallel is that even where the data is getting better, organizations still can't make decisions any faster. The data layer is improving while the decision layer is standing still.

Three things didn't keep pace with the data work: how fast a model output becomes a business action; who owns the outcome when the model is wrong; and who breaks the tie when AI and people disagree. We've built a Ferrari engine and bolted it to a tractor steering column. The bottleneck moved.

There's also an organizational bottleneck that's just as important. In most M&E companies, the work AI touches cuts across Sales, Rights, Ops, Finance, Legal, and IT, and each function carries its own decision rights, accountability framework, and budget. Rarely does someone own the cross-functional view, so even a good AI recommendation has nowhere clean to land. The companies escaping pilot purgatory are addressing this by building a unified operating model across those six functions, with a single budget for AI-driven decisions and clear executive accountability.

Three tells you can spot in any meeting.

First, who runs the meeting? In a dashboard culture, the senior leader presents the numbers and the team reacts. In a decision-system culture, the system runs the meeting, the team comes in with predictions and recommendations already generated, and the meeting is about what action to take.

Second, what actually gets reviewed in that meeting? If the answer is last week's numbers rather than next week's recommended actions, that's a dashboard culture.

Third, and this is the diagnostic question, does anyone in your organization have the formal authority to overrule a model, in writing, with documented reasoning? If yes, you have a decision-system culture. If no, you have a dashboard culture pretending to be one.

Three places where the value is real and measurable, and I see this pattern across all the industries I work in, not just media.

It's not a panacea. AI doesn't fix a broken data foundation, it doesn't resolve accountability questions when a model is wrong, and it doesn't replace human judgment in irreversible decisions.

Here's the trap I see most often: every executive wants conversational AI on top of their business. That's fine, but conversational AI is the interface to a system of intelligence, not a substitute for one. A natural-language layer over fragmented or ungoverned data produces confidently wrong answers, which is worse than no answer at all in any high-stakes workflow.

AI is the engine. Humans are the steering wheel and the brakes. The organizations that win build both.

There are three specific areas where humans must stay in the loop. First, anywhere the action is irreversible, such as contracts, regulatory filings, or terminations, because the conversational layer is well-suited for retrieval and synthesis but unreliable for irrevocable acts. Second, anywhere two models disagree: someone has to break the tie in writing with reasoning that gets audited, otherwise the model confirming what leadership already believed gets cited while the contradicting one gets explained away, which is AI-flavored confirmation bias rather than genuine AI-driven decision-making. Third, anywhere model confidence is high but consequences are large, because a model can be right ninety-five percent of the time and still be catastrophically wrong in the single case that costs the relationship.

The truth is that escaping pilot purgatory isn't primarily a budget question, it's a sequencing question. The companies that get there consistently follow the same pattern, and it doesn't require a massive program, just a small number of moves executed in the right order.

Start with one commercial workflow where the outcome is measurable and the cross-functional pain is real, whether that's yield optimization, pitch-to-pay, or make-goods management. Spend the first sixty days documenting it clearly: who decides today, with what authority, against what data, and where Sales, Rights, Ops, Finance, Legal, and IT each touch the process. The deliverable at the end of that phase is an operating cadence and a written set of decision rights, not a model.

This is one of the most miscalibrated conversations in the industry. Most organizations measure AI ROI use case by use case, tracking how many hours a chatbot saved or what percentage a model improved, and while those numbers are real, they miss the strategic asset entirely. The thing that compounds over time isn't any individual model; it's the platform underneath all of them.

Without a defined AI platform for use cases to roll up into, every pilot becomes a one-off with its own data, governance, and integration overhead. When the platform is genuinely defined first, the marginal cost of adding the next use case drops by roughly 70 percent and time-to-value shrinks from quarters to weeks. The right ROI question isn't what did this model save us, but rather what is the compounding value of the platform across the next twenty use cases we'll run on top of it.

Platform discipline also protects your people. When pilots get scoped without a platform foundation, AI projects tend to get quietly appended to existing job descriptions, so the seller, the analyst, and the ops manager are expected to maintain a model and respond to its recommendations on top of everything they were already doing. Most don't have the bandwidth, which means the model either gets ignored or, worse, used carelessly.