Most broadcasters are still debating whether AI will steal their jobs. I’m more interested in whether it can help them make better decisions in the next ten minutes — and whether that’s actually worth the effort.

By Ben Anchor — Wednesday, 18 February 2026 · Listen to the podcast episode

I’ve spent the last few months wrestling with a question that sounds deceptively simple: what if broadcasters stopped trying to predict the distant future and focused instead on improving what happens next?

That’s the premise behind nowcasting — a term borrowed from economics, refined during the UC Berkeley AI Strategy Programme, and now something I’m actively developing through Ancast Intelligence. The latest podcast episode unpacks the concept in detail, but what stuck with me wasn’t the elegance of the framework. It was the tension between what nowcasting promises and what it actually demands from organisations that aren’t built for it.

The Seductive Logic of Near-Term Optimisation

Nowcasting compresses the time horizon of prediction. Instead of asking what will happen next quarter based on historical patterns, it asks what is most likely to happen in the next few minutes or hours based on signals already available. Economists use it to estimate GDP before official figures are published, gathering high-frequency data like shipping movements and credit card transactions to understand the present more accurately.

When I first encountered that idea, I immediately thought about how broadcasters rely on yesterday’s ratings to inform tomorrow’s decisions. If economists can use near-term signals to improve economic estimates, why couldn’t broadcasters use behavioural signals to refine scheduling decisions?

The analogy is compelling. Traditional broadcast operations were built around predictability: programming grids set weeks in advance, promos rotated through fixed cycles, break structures predefined. That system worked when audience behaviour was stable and channel loyalty was high. But today, audiences switch platforms instantly, devices fragment engagement, and competing streaming services influence attention in real time.

In that environment, relying solely on lagging indicators leaves optimisation opportunities untapped. Forecasting still matters for strategic planning, but once strategic direction is defined, there are thousands of micro-decisions that shape actual performance: the order of promos, the sequencing within a break, the timing of a trailer, the transition between genres. Nowcasting focuses on improving those micro-decisions by interpreting behavioural signals in a structured way.

On paper, it’s elegant. In practice, it’s messier.

Where the Theory Meets the Messy Reality

The natural entry point is promos. They’re short-form, they appear frequently, and they carry lower risk than advertising contracts. Broadcasters invest heavily in creating promotional assets, yet placement decisions often rely on experience and historical norms rather than granular behavioural analysis. We know overall break retention, but we rarely isolate whether a specific promo in a specific context influenced tune-out behaviour.

Nowcasting enables a more precise question: given the signals present at that moment, was there a more effective option?

The process begins with data most broadcasters already possess. As-run logs provide second-by-second records of what aired. Audience measurement systems provide retention curves and tune-out points. Metadata describes genre, duration, tone, and targeting. When those datasets are aligned across thousands of historical breaks, correlations emerge. Shorter promos may consistently reduce drop-off during weekday evenings. Certain genre pairings may increase session continuation. Device type may influence tolerance for longer creative.

The model identifies these patterns and learns probabilistic relationships. It can then simulate alternative scenarios: if promo B replaced promo A in that context, what would the predicted retention curve have been? The output isn’t absolute certainty — it’s a probability estimate. But when applied systematically, even small percentage improvements can compound into meaningful commercial impact.

Here’s where I’ve become more cautious than I was six months ago: technical feasibility is only part of the equation. Commercial alignment is critical. Technology partners often prefer cost certainty, whereas exploratory initiatives involve uncertainty. Incentive alignment, governance clarity, and defined evaluation metrics must be established early. Nowcasting sits at the intersection of editorial expertise, data science capability, and commercial strategy. That intersection requires coordination and trust.

And trust, frankly, is harder to build than models.

Why Fast Channels Are the Proving Ground (and Why That Matters)

Fast channels operate in high-variance environments. They’re ad-funded, they experience high churn, and viewers can exit instantly. That volatility means even small improvements in retention can translate into revenue uplift. Fast ecosystems often provide richer device-level behavioural data and tend to have fewer legacy constraints than traditional linear broadcasters. That makes experimentation more feasible.

If you can demonstrate a one or two percent uplift at break-level across multiple channels, the aggregate commercial effect can be substantial. But here’s the debate I keep having with myself: is incremental optimisation the right framing, or does it undersell the potential?

I’ve started emphasising incremental gains over transformational promises because incremental gains compound. A 1% improvement in retention across hundreds of breaks accumulates quickly. Media markets are competitive, margins are tight, and transformational promises often fail because they aim too high too quickly. Nowcasting succeeds when positioned as disciplined optimisation rather than dramatic overhaul.

But I also wonder whether that positioning creates a ceiling. If the pitch is “we’ll help you eke out marginal improvements,” does that inspire the organisational energy needed to change ingrained workflows? Or does it consign nowcasting to the category of “nice to have” rather than “strategic imperative”?

The Cultural Shift That Nobody Talks About

One of the key insights from recent work is that nowcasting can begin as a contained desktop exercise. You analyse historical data offline. You test whether measurable optimisation signal exists. There’s no need to alter live operations or implement automation at the outset. The first milestone is evidence. If no meaningful signal appears, you stop. If signal is measurable, you build internal confidence and consider limited live testing later.

A realistic pilot might analyse a month of historical data for a specific channel — say, December, to include Christmas reruns. You’d isolate defined break types, simulate alternative content choices, and quantify predicted retention uplift. The outputs would include uplift curves, scenario comparisons, and confidence intervals. The objective isn’t perfection; it’s directional evidence.

Proper validation requires holdout datasets, cross-validation techniques, and clear separation between training and testing data. You avoid tuning the model purely to historical anomalies. You focus on consistent patterns across time periods. Statistical discipline is essential. Without it, optimisation claims lack credibility.

But the long-term implication is cultural, not technical. The shift is toward comfort with near-term optimisation. Organisations become more willing to experiment within controlled boundaries. They move from reactive reporting toward proactive refinement. Strategy remains important, but execution becomes smarter.

That shift requires leadership that values testing over certainty. And in my experience, that’s the rarest resource of all.

Where I’ve Landed (For Now)

If I had to summarise nowcasting in one sentence, it would be this: it’s a disciplined way to use real-time or near-term behavioural signals to improve the next decision, without disrupting long-term strategy.

The organisations that become comfortable refining the next decision consistently will outperform those that rely solely on retrospective analysis. In today’s media environment, standing still is still a decision. Nowcasting is about choosing better ones.

But it’s also a bet. A bet that broadcasters have the appetite to interrogate their own data, the patience to validate signal before scaling, and the humility to accept that editorial instinct doesn’t disappear — it just becomes more informed.

I’m making that bet. We’ll see if the industry is ready to join me.