Chalice AI

Chalice AI: Brand-Specific Models for Programmatic

This AdExchanger Talks podcast episode features Ali Manning, COO and co-founder of Chalice AI, discussing the company's approach to letting brands bring their own models into walled gardens, DSPs and SSPs. Chalice positions its software as continuously running, refreshable models (including generative AI) that optimize for brand-specific business outcomes rather than standard proxies like clicks or video completion. Manning describes challenges around the term "custom algorithm," explains Chalice's origin story, and gives a client example (Bayer’s One A Day) where Chalice used AI modelling and DSP bidding to drive offline "new to brand" sales. The conversation also touches on containerization and agentic bidding as possible ways to realize more precise programmatic outcomes.

Agencies Experiment with Vector-Based Targeting

Media agencies and adtech vendors are piloting vector-based targeting — a technique that encodes multiple audience signals into numeric "vector embeddings" to find and activate audiences across streaming, CTV and digital inventory. Major holding groups such as WPP and Dentsu are testing early-stage solutions and partnering with specialist vendors (Chalice, Equativ) and SSPs/publishers. Amazon offers a similar first-party targeting tool (Brand+) and publishers/SSPs and adtech vendors say vectors could improve cross-platform targeting where deterministic signals are scarce. The approach is seen as promising but nascent, with concerns about explainability, standardization and operational complexity.

Agencies Turn to AI Agents, Brands Move In‑House

Agencies are showcasing AI agent pilots, partnerships and products to speed media planning and buying — but face competition from advertisers building their own agentic capabilities. WPP is testing a media‑buying and planning agent for premium video, while Dentsu has partnered with Newton Research to trial media‑buying agents and faster analytics via Dentsu.connect. Independent agencies and consultancies such as Butler/Till and Dept are also deploying agent tooling (including Claude agents and a productised 'Agent Studio') and using Google models to accelerate delivery. Advertisers including Hyundai have built in‑house bidding agents with partners (Chalice and Canvas) using a containerized OpenXBuild solution, reporting a 67% reduction in online video CPM and a 20% decline in cost per defined 'high value action' during pilots. The piece cites WFA research on broad in‑house AI adoption and frames the trend as a strategic threat and opportunity for agencies.