Why algorithmic mastery is not about smarter algorithms alone, but about how intelligence is applied, tested and mastered in live addressable media.
Applied AI is everywhere in programmatic conversations. From bid optimisation to outcome modelling, it’s positioned as the engine behind smarter, faster, more effective media. But as adoption accelerates, one tension remains unresolved: how do we separate promised intelligence from performance that shows up in market?
In Addressable Media, that distinction matters. Applied AI tools increasingly sit inside live buying environments, shaping bids, targeting and optimisation logic in real time. When their impact isn’t clearly understood, teams risk defaulting to vendor narratives or surface level metrics rather than experience led evidence.
To move beyond assumption, our Total Addressable team set out to evaluate applied AI partners based on how they performed in live campaign conditions, rather than relying solely on stated capabilities or marketing materials.
Testing AI where it operates
Rather than relying on case studies or theoretical capability, each applied AI partner was assessed through live pilot testing within active programmatic campaigns, benchmarked against a standard control approach.
The ambition was simple but deliberate: understand whether these tools genuinely enhance media and behavioural outcomes, and whether they can do so consistently, at scale, in real buying conditions.
Each partner was evaluated across five consistent dimensions:
- Service – responsiveness, collaboration and quality of support
- Setup – ease and speed of onboarding and integration
- Activation requirements – complexity, resourcing and dependencies
- Capability – functional strengths and flexibility
- Campaign performance – results delivered against agreed KPIs
Applying a consistent framework across participating partners helped improve clarity and comparability while supporting a more balanced view of performance.
What AI performance really looks like in practice
One of the clearest learnings was that applied AI success is rarely driven by algorithmic sophistication alone. Operational realities (how quickly a tool can be activated, how intuitively it integrates, and how closely partners collaborate) play a defining role in outcomes.
Across the pilots, results varied by campaign context, implementation approach and operational fit. In some cases, tools aligned well with campaign goals and trading workflows. In others, additionalsetup, calibration or operational requirements affected how quickly value could be realised.
These findings should be understood as context-dependent observations from specific tests rather than universal conclusions about any provider category or solution. Key differences only surfaced through live, in market testing, reinforcing why AI must be evaluated within the full experience of buying, not isolated demonstrations.
Why operational experience matters as much as intelligence
Applied AI doesn’t operate in a vacuum. It exists within a wider ecosystem of traders, strategists, platforms and clients, meaning its value is shaped as much by how it fits as by what it can do.
In practice, factors such as implementation effort, data requirements, commercial setup and workflow compatibility can all influence how effectively a solution is tested and deployed. These operational considerations can shape the pace of learning and optimisation just as much as the underlying technology itself.
The takeaway is that effective Applied AI depends not only on optimisation capability, but also on how well a solution can be implemented, tested and managed within real operating environments.
Any discussion of specific partners or tools should be assessed carefully to ensure that claims are evidence-based, appropriately substantiated and aligned with relevant internal approval and governance processes.
In some test environments, partnerships combining sell-side decisioning with curated supply and the Dentsu Media Exchange have been assessed as part of a broader evaluation of applied AI in live buying conditions.
This includes partners such as Chalice, whose capabilities have been tested within these environments using consistent evaluation frameworks.
Layering solutions such as Chalice Curate AI across curated premium inventory can enable more streamlined setup, dynamic inventory selection and real-time optimisation aligned to campaign objectives.
These approaches can offer useful applications in specific contexts, depending on campaign requirements, implementation considerations and internal approval processes.
"dentsu's approach here reflects exactly what progressive agency partnerships should look like – a genuine commitment to testing cutting-edge technology in live environments, in service of real client outcomes."
Freddie Turner, Managing Director EMEA, Chalice AI
Shaping a more evidence‑led applied AI strategy
This work has helped establish a more informed point of view on how applied AI can be assessed and deployed within addressable media. Rather than treating AI as a universal upgrade, the focus shifts to understanding which solutions may be appropriate for specific use cases, subject to appropriate validation, governance and policy approval processes.
It also creates a stronger foundation for structured partner discussions around testing, feedback and opportunities for refinement over time, without presuming uniform results across different clients, campaigns or operating conditions.
As applied AI continues to evolve, this evidence led approach will be critical in ensuring innovation genuinely enhances media experiences, rather than adding complexity without clear return.
Final thoughts
Applied AI has the potential to meaningfully transform addressable media, but only when it’s evaluated with the same rigour, curiosity and creative intent as any other activation strategy.
By grounding decisions in controlled testing, consistent evaluation and real-world experience, teams build true algorithmic mastery and move beyond hype and towards AI solutions that deliver impact that’s felt, not just measured.
