The risk isn't waiting to fix your video strategy. It's that by then, the cost of entry will have tripled and the learning curve will be vertical.
Here's what we're seeing across our client portfolio right now. Platform algorithms are already rewriting how video gets distributed, discovered, and rewarded. The average global CPM climbed to USD$8.74 in 2025, up from USD$7.91 the year before. That's not inflation. That's competition for attention intensifying whilst most marketers are still treating video as a content format rather than a distribution system.
The 60% video budget projection for 2026 isn't about volume. It's about the shift in how search engines and social platforms prioritise video in their algorithms, which fundamentally changes the ROI calculation for every media channel you're running.
The Algorithm Shift You're Missing
When we first leaned into video as a core part of client strategy, the dominant assumption was that attention was the primary currency. Stop the scroll with strong creative, and everything else would follow. Success was framed around views, completion rates, and early engagement signals.
What we've since realised is that attention without intention is fragile.
The shift in perspective came as we started overlaying video metrics with downstream outcomes. Brand lift, search behaviour, site quality, and conversion pathways told a different story. Some of the most efficient video wasn't the most entertaining. It was the video that showed up at the right moment, in the right environment, with a message that aligned to where the audience already was in their decision-making.
Context began to outperform creativity alone.
This matters because platforms have become extraordinarily good at delivering exactly what they're asked to deliver. The issue is that most marketers are still giving them proxy instructions (views, reach, lowest CPM) and then expecting commercial outcomes to emerge downstream. That gap has widened as algorithms have matured.
What Platforms Actually Optimise For
Three shifts stand out in how algorithms prioritise video now.
First, algorithms now optimise for behavioural likelihood, not brand impact. If you ask for video views, systems will increasingly find people who watch a lot of video, regardless of whether they're in-market, brand-receptive, or even category-relevant. As inventory and signals have scaled, platforms have learned how to extract efficiency from behaviour patterns that have very little to do with growth. Left unchecked, this creates algorithmic efficiency, not marketing effectiveness.
Second, creative has become a signal to the algorithm, not just a message to the audience. Video is no longer just content. It's training data. Early engagement, watch time, and interaction all feed back into who the platform decides to show your message to next. Without deliberate constraints such as audience frameworks, frequency logic, or sequential intent, the algorithm can quickly optimise your media towards the wrong type of attention. At scale, that compounds fast.
Third, platform automation has outpaced measurement maturity. As platforms push broader targeting and black-box optimisation, marketers are left with fewer levers but higher accountability. The result is a false sense of security. Performance looks stable in-platform, whilst contribution to brand strength, consideration, or incrementality quietly erodes.
This is why "letting the algorithm do the work" is an incomplete idea.
Creative as Training Data
Once you accept that creative is teaching the system who to find, the brief changes in a material way.
Three years ago, most video briefs were still anchored in message clarity and storytelling craft. Those things still matter, but they're no longer sufficient. Today, we're deliberately designing video with an understanding that the first signals it generates will shape the audience it scales into.
The opening is now a targeting mechanism, not just a hook. We're far more intentional about what the first two to three seconds signal. Not just visually, but semantically. Category cues, usage context, price framing, or problem statements early on act as filters. They help the algorithm differentiate between people who find the content entertaining and people for whom it's relevant. Three years ago, we optimised openings for intrigue. Now, we optimise them for qualifying attention.
Brand presence is engineered for learning, not recall alone. Consistent brand cues (sonic, visual, or linguistic) help platforms build more reliable feedback loops around who responds meaningfully to the message. In practice, this reduces wasted learning and improves stability when we scale spend. Previously, branding was about memory structures. Now it's also about algorithmic clarity.
We design for variation, not perfection. Instead of chasing a single "best" cut, we build intentional creative families. Small, controlled variations in opening line, framing, or call-to-value allow us to see how the algorithm interprets different signals. This isn't A/B testing in the classic sense. It's a way of shaping the learning path.
The Budget Allocation Problem
For a marketer allocating a 2025 budget today, the risk isn't under-investing in video. It's over-investing in the wrong role for video.
By 2026, the most effective marketers won't be asking "How much should go to YouTube, Meta, CTV?" They'll be asking:
- Where do we want the algorithm to learn?
- Where do we want it to scale?
- Where do we want it to harvest demand?
Practically, that means reserving a meaningful portion of video budget for environments where learning quality is high, rather than just where reach is cheap. That's often mid-funnel digital video, selective CTV, or platform-native placements tied to intent signals, not pure awareness blasts.
We still see video budgets justified almost entirely on short-term KPIs. By 2026, the smarter approach is to explicitly allocate budget to audience qualification, message testing, and signal shaping. This isn't experimental spend. It's infrastructure. Marketers who don't do this will find their performance media becoming more expensive and less predictable, because the algorithm never learned who matters.
What B2B and Long-Cycle Brands Get Wrong
For automotive, universities and B2B brands, the mistake is assuming that because the outcome is long-term, the signals can be vague. In reality, algorithms still need clarity, even when the goal is consideration rather than conversion. The difference is what we're teaching them to recognise.
What we're reconciling is two timelines: the platform's need for immediate behavioural feedback, and the brand's need for slow, cumulative impact.
Instead of asking platforms to optimise for views or clicks, we anchor distribution around meaningful engagement proxies. Qualified video completion, time spent with higher-order messages, interaction with depth cues, or exposure frequency within defined cohorts. These don't create instant sales, but they do tell the system what considered attention looks like in a low-frequency category.
For a B2B or institutional brand, the most valuable learning isn't "who will convert today?" It's "who is willing to think about this category seriously?" Creative is designed to surface that distinction early. That might mean introducing academic outcomes, research credibility, or problem-framing rather than offers. The algorithm then learns to prioritise people who respond to substance, not spectacle.
Long-term brands benefit disproportionately from repeatable structures such as consistent narrative arcs, recognisable formats, stable brand cues. Over time, this gives platforms a clearer pattern of response and reduces volatility. Three years ago, consistency was framed as a branding principle. Today, it's also a machine-learning advantage.
The Five Video Levers You Control
Marketers don't control the black box, but they absolutely control the inputs that shape what the black box learns. As automation increases, those inputs matter more, not less.
The objective is the most powerful instruction you give the system. Algorithms don't understand strategy. They understand optimisation goals. Choosing views, reach, clicks, conversions, or value-based outcomes is not a tactical setting. It's a philosophical one. We're seeing brands stuck in low-quality learning loops because they default to easy-to-hit objectives that have weak correlation with growth.
Audience constraints define what the algorithm is allowed to learn. Broad targeting isn't the problem. Ungoverned broad targeting is. Marketers still control seed audiences, exclusion logic, geographic boundaries, and frequency caps. These act as guardrails. They don't strangle delivery. They prevent the system from optimising towards attention that's cheap but commercially irrelevant.
Creative structure shapes learning more than creative polish. As we've discussed, creative is training data. Marketers control what appears in the first seconds, how clearly the category is signalled, whether brand cues are consistent, and how variation is introduced. Two brands can spend the same amount with similar media plans and end up with very different learning outcomes purely because one gave the algorithm clearer signals to work with.
Measurement design closes or breaks the feedback loop. Platforms optimise to what they can "see." Marketers control whether that view is narrow or meaningful. That includes which conversion events are passed back, whether value signals exist, how incrementality is assessed, and whether brand and performance data are ever reconciled. If measurement is shallow, optimisation will be too.
Budget shape and time horizon influence algorithmic behaviour. Automation responds differently to steady investment versus bursty, reactive spend. Marketers still decide how much budget is stable versus opportunistic, how long learning periods are protected, and when to reset versus let systems compound. Through 2026, consistency will be an underrated lever because it allows learning to accrue rather than constantly restart.
What Happens If You Wait
The specific risk for a brand that decides to wait another year or two before restructuring how they approach video comes down to three compounding factors.
First, the cost of quality attention is rising faster than total media inflation. Competition for user attention intensifies each quarter. Brands that start building their algorithmic learning now will have two years of compounding knowledge by 2026. Brands that wait will be paying premium rates to learn what their competitors already know.
Second, platform algorithms are changing now, not in 2027; The way YouTube prioritises hyper-personalised content using deep learning models, how Meta's auction systems reward creative freshness, and how search engines integrate video into results pages. The brands that adapt their creative briefs, distribution strategies, and measurement frameworks today will be operating from a position of strength. The brands that wait will be reacting to a market that's already moved.
Third, the window for protected learning is closing. Right now, you can still allocate budget specifically for audience qualification and signal shaping without immediate pressure for returns. By 2026, when video represents 60% of social media ad budgets, that luxury disappears. Every dollar spent will be expected to perform immediately, which makes it nearly impossible to build the foundational learning that effective video strategies require.
The uncomfortable truth is that video spend will stop being about "how much" and start being about "where learning lives." The marketers who recognise that now will find that video doesn't just scale awareness. It compounds effectiveness across everything else they do.
Reach out to your ADMATICian to discuss how video can drive better outcomes for your brand.