For years, paid social skill meant knowing the platform: the campaign structure, the levers, how fast you could move around Ads Manager. That still counts, but one theme ran through Meta Agency Summit from start to finish. As Meta’s systems automate more of the work, the job is shifting from managing the tools to shaping what the tools are told to chase.
So the question changes. “How do we optimize this campaign?” becomes “Is the system optimizing for what our business actually needs?” Four areas from the summit show what that looks like in practice.
Send Meta the signals that reflect your business
Meta’s automation optimizes against the signals it receives, and a platform conversion is not always a valuable business outcome. Two customers can trigger the same purchase event while delivering very different margins or lifetime value. A campaign can look healthy in Ads Manager while your internal dashboard tells a different story.
The summit’s answer was to move business reality closer to the optimization process, through custom attribution, order and value information passed back to Meta, and value rules that help the system tell outcomes apart. Revenue is a fine starting point, but if margin or lifetime value drives your decisions, they belong in the conversation with your media team now. Ask a blunt question: are you sending Meta the signals that reflect your goals, or the ones that are easiest to send? Conversions API (CAPI) is the foundation, because a reliable signal about the wrong outcome just helps Meta find the wrong customers faster.
Give automation room, with a reason behind every constraint
Meta talked about planning around your business calendar, including front-loading budget ahead of key moments, while letting the automated systems respond to demand as it shifts. Every manual constraint you add should have a purpose you can state, because overly restrictive inputs can stop a well-aimed system from finding opportunities.
Your part is knowing when demand is likely to change, setting guardrails that protect the business, and deciding where human judgment adds something the system can’t. Then you have to prove it worked. The summit kept returning to lift testing and incremental measurement, since a delivery improvement means little if those sales would have happened anyway. Ask your team how much of your reported performance is incremental, and how they know.
Build a creative loop you can learn from
Retail and ecommerce advertisers got the most attention here. Meta described connecting catalog data, creative automation and Pixel signals so the system can match relevant product and creative combinations to different people, instead of treating the catalog as a static feed and creative as a separate production line. It also covered Ads Creative Studio, a workspace meant to bring creative into one place, add generative AI to the workflow and show what’s working and why.
More assets only pay off if you learn from them. The loop to build runs from creative development to media performance to business outcomes and back again: which messages resonate, which product and creative combinations drive valuable actions, and what the next round of production should prioritize. One more distinction from the summit is worth keeping. Creators can be a media channel, because their audience is the distribution, and a creative asset, because their content can run in your paid program. The stronger plan uses both.
Treat AI agents as promising, not finished
The most interesting discussion centered on Model Context Protocol (MCP), which lets AI agents, including tools like Claude, connect directly to platforms like Meta. Early examples included pulling campaign information into a shared workflow, comparing channels to decide where budget should go, and building campaigns through MCP. Be clear-eyed about where this stands. It is early-stage, not a guarantee that every workflow is ready to run without oversight, and the integrations need proper setup before they deliver anything useful.
The same caution applies to the insights tools Meta previewed: explanations for why performance changed (competition, seasonality), customer experience insights, and forecasts of the trade-offs before you move a bid or budget. Knowing what changed is easy, and knowing why is hard, so these could be a real help. Meta also acknowledged the products are imperfect, which raises the right question: how accurate does a recommendation need to be before you act on it? Treat AI-generated explanations as input to your thinking, and check them against your own data.
