How your optimizer works

Your optimizer isn't a chatbot guessing at your numbers. Every recommendation comes from two layers working together — and you stay in control of what actually changes.

Not just an LLM

Most "AI" tools hand all your data to a language model and hope for the best. Zimmer splits the job in two:

• The analysis layer — the math. It works through your real performance data — far more than anyone could review by hand — to surface what's genuinely going on and filter out noise (the specifics are below).
• The AI layer — the judgment. A language model then reasons over those findings the way a senior strategist would: weighing context, choosing the changes worth making, and explaining each one in plain language.

Together, the two layers mean every recommendation is grounded in your real data and clearly spelled out.

What the analysis looks at

Across your connected accounts, it watches for things like:

• Winning and losing slices — which audiences, placements, devices, geographies, or times consistently beat or miss your target, so budget can shift toward what's working and away from what's draining it.
• Creative fatigue — is a creative wearing out? It watches for performance sliding as frequency climbs (the sign an audience has seen the same ad too often), so you know when to refresh.

How it runs

Optimizers run on a schedule (set by the kit), not only when you're watching. Every run follows the same path:

1. Schedule fires — the run starts on its own timer.
2. Fetch your latest data — it pulls current performance from your connected accounts, down to the dimensional level (audiences, placements, and so on).
3. Analysis finds what's real — the math layer surfaces the genuine signals and filters out noise.
4. AI reasons and ranks — it weighs context, decides which changes are worth making, and ranks them by expected impact.
5. Proposals — each recommendation is written up with the reasoning behind it.
6. You approve — review and approve in Approve First. Monitor stops here (nothing changes); Auto skips straight to the next step.
7. Execute — approved changes are applied on the platform, and logged in Runs.

So where a run stops depends on the agent's mode — Monitor at proposals, Approve First at your review, Auto all the way through.

Why you can trust it

• Built on your real data — every figure traces back to your account.
• You can check the work — see the full dimension-by-dimension detail behind every recommendation in Campaign Insights.
• Ranked by expected impact — the most worthwhile changes rise to the top.
• You're in control — nothing changes until your mode allows it.
• Everything is logged — every run and proposal is recorded.
• Works across your platforms — Meta, Google Ads, YouTube, TikTok, and DV360.