Through MMA's AI Leadership Think Tank (ALTT), member brands run controlled experiments at five stages of the AI ad funnel: creative, evaluation, personalization, audience discovery, and contextual targeting. ALTT has run 75+ experiments. Its personalization lab, CAP, has completed 29 across 18 global brands, with an average lift over baseline of +160%.
Each lab is co-funded research with a named partner, run on live campaigns with paid media. I run the program with MMA.
// most pilots are free for MMA member brands. participation requires MMA membership.
AI message personalization in production
Production AI personalization model. 29 experiments, 18 brands, 4 countries. Recruiting for the next wave: video / CTV, brand lift, offline sales, cross-channel.
Agentic creative generation, end to end
Generate and iterate short-form video and other creative through Monks.flow, from research/brief through storyboard to final assets. Research-environment evaluation built in.
Pre-launch creative scoring
AI extracts the visual + linguistic features that drive performance for a given brand, then scores new creative pre-launch and recommends edits. Predicts conversion impact before spend.
AI-first audience discovery
Finds high-response cohorts continuously from CLV, CPA, conversion, and engagement signals. Of 300 campaigns analyzed, 90% showed incremental conversion improvement.
LLM-powered contextual targeting
Contextual targeting that uses precomputed LLM embeddings across a 4-billion-URL index. It reads what a page means from the embeddings, so it does not depend on keyword matches, and it runs at programmatic speed.
How Reddit communities shape AI answers
Tests and quantifies how activating Reddit communities influences Answer Engine Optimization (AEO) and branded responses surfaced inside AI assistants.
Measuring brand visibility in AI answers
Measures how often a brand surfaces in AI-generated answers (ChatGPT, Gemini, Perplexity), then tests which content changes move that visibility against a baseline and a control group, so a result shows cause. The pages models cite are often not the ones that rank in classic search.
Five of the labs cover consecutive stages of one AI ad pipeline: create, evaluate, optimize, discover, and target. Each stage has its own lab, and each lab feeds the next.
Brands and platforms join a pilot or propose a new experiment through MMA membership. Most labs are free for member brands.
If you want my read on which lab fits your stack and goals, tell me about your setup.