Zeta connects data…AGENTS Simulate TESTS (not well)…Publicis earnings.
This week, we saw developments that move AI in advertising agencies from another tool to a part of the overall operating model. Biggest news stories of the week:
ZETA INCREASES REVENUE FORECAST
Zeta has been pushing an “intelligence layer” called Athena that promises to pull together data points across a consumer journey into a single vision. They raised their forecast and delivered strong Q2 results. We will wait to see how implementation really happens, but it is clear they are moving from an add on to something central to the way marketing operates.
RESEARCHERS TEST AI AGENTS ABILITY TO SIMULATE TESTS
Lorenzo Masoero (senior applied scientist at Amazon) and Stefan Hut (senior economist at Amazon) published a paper titled “Can AI Agents Simulate A/B Test Outcomes? A Validation Framework for Agentic Experimentation”. They compared simulated tests, run by AI agents, against 67 historical A/B tests. So can the agents simulate outcomes? Sort of. The agents were able to predict the directional outcome 7 out of 10 times. Not bad results, but being wrong 30% of the time is concerning.
PUBLICIS MADE (AND SAVED) A LOT OF MONEY WITH AI
The big HoldCos are clearly going deep in AI. For Publicis, it seems to be working out with a reported second quarter 6.5% growth. The driver (they say) is using AI to help integrate the agency. We will watch for more signals like this: AI is moving from helping individual departments with individual tasks to being a layer of intelligence that both increases the ability to produce work, as well as accelerates learning.
Mover of the Week: AI-selected experiments
While not enough to justify a full investment, the fact that AI agents can predict results from 7/10 experiments is promising. Now that AI can help produce large numbers of options, the challenge is going to be deciding which ones are worth testing. Flooding the platforms with a million tests will just drag down the results of everything. Here, we see how AI can help humans make better decisions about what should be tested.
Thursdai Index: this week’s numbers
Experimentation Capability: 7.9/10Learning Quality: 7.0/10
Laboratory Reliability: 6.2/10
Market Traction: 7.4/10
OVERALL: 6.4/10
The Thusdai Index and report is the work of 61 Hawk founder Gary Stein with assistance from an OpenAI research agent.
