This post was adapted from an article by Tereza Fukátková, Head of AI at TV Nova, data scientist, instructor and storyteller.
AI initiatives are sprouting like mushrooms after rain. As someone whose passion is AI, I am
on cloud 9, the demand is through the roof! Even cautious industries like banks or media (my domain) are taking it seriously. McKinsey's The State of AI report shows that in 2024, 78% of the surveyed companies used AI at least in one department, a huge jump from 55% in 2023. Mushrooms, right?
But some of those mushrooms dry out, rot, or grow on the wrong substrate. A true PoC
graveyard. Gartner would tell you that’s perfectly normal: most AI technologies have just
tumbled off the Peak of Inflated Expectations and are barreling toward the Trough of
Disillusionment - or, as I prefer to call it, towards enlightenment. The plunge can be harsh,
but it plants you on solid ground from which you can start climbing the slope of productivity
again.
At the base of that slope, often unseen yet absolutely vital, lie data pipelines - the bedrock of
every successful AI initiative.
It all starts with strategy
I would be lying if I claimed data pipelines were the only ingredient in the recipe for AI
success. At the very core sits a sound AI strategy, one that grows naturally out of the
broader business strategy. It clarifies your ambition and spotlights the key domains you’ll
double-down on to get there.
Yet the deeper you dig into what truly powers that ambition, the trail always loops back to data and how to deliver it to your solutions reliably and at scale.
Any credible AI plan has this at its bedrock. The strategy I’m shaping for TV Nova, with the
not-so-modest goal of making us the media AI forerunner, is no exception.
Data is energy
Why are pipelines such a big deal? Let’s start with the basics—what is data? You’ve
probably heard the classic comparison that “data is the new oil.”
I prefer the metaphor Carruthers and Jackson (both seasoned CDOs) use in Halo Data: Understanding and Leveraging the Value of your Data: data is energy.
And just as your company taps electricity or gas to drive machines, light up stores or keep laptops running, it can tap data to spark brand-new value. Small caveat: it only works when the energy actually flows. Guess what keeps it flowing? Pipelines!
Let’s peel the onion one layer deeper: if you’re not collecting data today, start … now.
Otherwise, your AI ideas will never make it off the slide deck. Buying a license for a chatbot
can lift your productivity, sure, but that’s only the tip of the AI iceberg. Anyone can swipe a
credit card and get the same tool. What they can’t buy is the data only you own: what
customers search for on your e-shop, last week’s unexpected best-seller, the most returned products, the patterns buried in your logs.
AI on its own isn’t your secret sauce. AI combined with the energy of your data - is.
Data pipelines - the real differentiator
Once we picture data as energy, pipelines are simply the grid that carries that energy to
every place it is needed. If this still feels abstract, imagine a modern, elegant house—beautiful and ready to impress—but disconnected from electricity and running water. You could live there, but the excitement would fade fast.
The same applies to AI initiatives: without data pipelines, the architecture remains a showroom, not a home.
W. Liang and the co-authors underline this in a study published in Nature Machine Intelligence. They show that the real differentiator isn’t the model you pick, but how you treat the data
at every step of its journey: “Choices made in each step of the data pipeline can greatly
affect the generalizability and the reliability of the AI model trained on these data, sometimes
more than the choice of the model.” In the end, it’s the data and the pipelines behind it that drive everything.
To make this concrete, take the personalisation project we're building for our VOD platform.
The key to success is knowing both sides of the equation: each viewer on one side, our
library of films and series on the other. Using on the out-of-the-box metadata would deliver an okayish result, but okayish isn’t our ambition. The surest way to boost any machine-learning model is to feed it richer, cleaner data. That became mission one.
I can’t spill the entire secret sauce behind our approach, but to give you one ingredient: we mine metadata from genAI speech-to-text transcripts.
Liang’s paper further notes that: “Data scientists spend nearly twice as much time on data
loading, cleansing and visualization than on model training, selection and deployment.” I’d add: data scientists don’t exactly cheer for that part, but it is exactly where data engineers shine. In our personalisation project, the engineering team's pipeline delivers high-quality input, sharpens model performance, and spares the data-science team from tedious manual interventions.
Shine a light on the pipelines
The case for solid data and well-built pipelines is clear - not just to me, but to most data and
AI leaders I speak with. The theme pops up at nearly every meet-up or conference. Yet if
everyone understands the stakes, why do pipelines so often slip down the priority
list?
Because, just like the wiring and plumbing in that showcase house you imagined, you don’t notice them at first glance. They live below the surface, invisible until the moment you flick the
switch or turn the tap. They soak up the budget early and only pay you back later, when the lights come on or, in our world, when the model makes it to production and starts delivering real
value.
It’s on us, the AI leaders, to give data and pipelines the credit they deserve —to write them
into the AI strategy and, crucially, act on it. Sure, the models and their shiny results will
always grab the brighter spotlight. But explaining what makes those results
possible doesn’t kill the magic, it will enhance it.
When people see the solid ground the projects stand on - the data and the pipelines that carry it - AI becomes more tangible, more approachable and ultimately more impactful.
Turning Ambition into Reality
Three takeaways for AI leaders ready to move from vision to outcomes:
-
Power AI with your own energy—your data. AI alone isn’t magic; the magic happens when it’s fuelled by high-quality, metadata-rich, proprietary data.
-
Invest in pipelines early. They’re the infrastructure that moves your data-energy to where it creates value. Build PoCs with pipelines in mind so you’re ready for production success.
-
Make the invisible visible. Pipelines may be hidden like wiring in a house, but they’re vital. As AI leaders, it’s our job to shine a light on them and make sure their value is recognized.
The AI landscape will keep evolving at breakneck speed. New models will emerge, new use cases will inspire, and the temptation to chase the next shiny thing will always be there. But the organizations that actually deliver on their AI promises will be the ones with rock-solid data foundations—pipelines that keep the energy flowing.
So, before you leap to the next big idea, ask yourself: Is my data ready to deliver? If the answer is “not yet,” now is the perfect time to start building the grid that will power everything to come.
Comments