The layers of Meta's AI stack
Meta's AI advertising stack has three broad layers. Advantage+ is the automation layer for campaigns, most visibly Advantage+ Shopping campaigns, which collapse targeting, placement, and structure decisions into a largely automated setup aimed at a conversion goal. Automated placements let Meta decide across Facebook, Instagram, Reels, and the wider network where each impression runs. And a growing set of generative creative tools produce and vary images, text, and video to test more combinations than a human could.
Together they push Meta toward the same model as Google's PMax: you provide the goal, the creative raw material, and the signals, and the system makes most of the granular decisions.
Where the automation genuinely wins
Meta's machine learning is very good at two things: finding the people within a broad audience most likely to convert, and testing creative combinations at a scale and speed no manual process matches. For many advertisers, broad targeting with Advantage+ now outperforms the tightly hand-built audiences that used to be best practice, because the model has more signal than any manual segmentation.
The creative automation compounds this: the system can serve the right variant to the right person and retire fatigued assets faster than a human would notice. When the inputs are strong, the automation turns them into performance efficiently.
What still decides your results
Automation does not remove the levers that matter, it concentrates them. Three inputs still decide the outcome. Creative is the first and biggest: when targeting is automated and broad, the ad itself becomes the main variable, so the quality and volume of creative is the lever, not the audience. Conversion signal is the second: Meta optimises toward the events you send it, and with browser tracking decaying, a clean Conversions API feed is what keeps that signal accurate. Audience seeds are the third: the customer lists and value-based sources you provide still shape who the model looks for.
Give the stack strong creative, accurate signal, and good seeds, and it performs. Starve it of those, and no amount of automation rescues the account.
The guardrails iClick runs on Meta
The same discipline that fences Google's automation applies here. iClick verifies the Conversions API feed and deduplication before trusting Advantage+ optimisation, because the model scales whatever the events tell it. Creative is treated as the core job, not an afterthought, with a steady cadence of fresh assets tested against the funnel and retired on the data. Audience seeds are built from value-weighted customers so the model chases profitable buyers, not just cheap ones. And incrementality is watched, because broad automated campaigns can absorb existing demand and report it as new, the Meta version of brand cannibalisation.
The automation is a strong engine. The guardrails are what keep it pointed at incremental profit.
How to adopt it without losing the plot
The sensible path is to lean into the automation where it is strong and keep your effort on the inputs it cannot generate. Let Advantage+ handle targeting and placement, and pour your time into creative volume and quality, conversion-signal integrity, and value-based seeds. Test broad automated setups against your existing approach with a clear read on incremental results rather than platform-reported ones.
What does not work is treating Advantage+ as a set-and-forget button. It is set-and-feed: the automation runs continuously, but it runs on the creative and signals you supply, and those still need a human doing the work that actually moves the numbers.
