Create an optimizer agent

An optimizer is an autonomous analyst that watches your campaigns and proposes or makes changes toward a target you set. You build one from a kit — a ready-made setup for a platform and job.

Create it

Open Kits (opens in a new tab), pick a kit (Paid Media, E-Commerce, or Monetization), and select Use this kit.
Connect the platform (if it isn't already) and choose the ad account it runs on. An optimizer runs on one account, so create a separate agent for each account you want covered.
Set it up across its tabs (below).
Publish it. An optimizer only runs on its schedule once it's published.

The agent page

The agent page opens with a header showing its goal and publish state (with Duplicate and Activate / Deactivate). Below are four tabs.

Configure

The four tabs are the same for every kit — what changes is what's inside Configure. The steps below are the ad-optimizer setup (campaign scope and a target KPI). E-commerce and publisher (monetization) kits use this same tab for their own settings instead — their platform integration and sync or yield options (see the Kits pages).

Configure is a guided wizard with two steps — Platform, then Optimization. Alongside it, a Replay (last 7 days) panel estimates the impact of your current settings — proposals, pauses, bulk actions, and spend saved — and flags how many you've changed from the kit's defaults.

Step 1 · Platform

Connected resource. Choose the connected integration (Meta Ads, Google Ads, TikTok, DV360, and so on) and the ad account whose campaigns this optimizer can evaluate — one account per agent.
Allowed actions. Turn on exactly what the agent may do, with a live "X of Y enabled" count:
- Targeting dimensions it may adjust — which dimensions appear depends on the platform and kit. On Meta, for example: placement, device, operating system, geography, age, gender, audience, and schedule / dayparting; other platforms expose their own.
- Tiered actions, grouped by how broad they are — from surgical (within a single ad set or ad group, such as pausing or resuming an ad), through entity-level, to campaign-wide.

Step 2 · Optimization

Set your goals, scope, budget, pacing, and safety, plus how it analyzes and who gets notified.

Scope. Pick a scope mode — a single campaign, selected campaigns, or all eligible — and optionally auto-enroll new campaigns. Narrow with status and objective filters, then choose exactly which campaigns the agent manages. (On Meta, objectives are Sales · Awareness · Traffic · Engagement · Leads · App Promotion.) Then set the optimization levels it may act on — the campaign and the level beneath it (ad set, ad group, or line item, depending on the platform). Syncing is account-level; use Sync campaigns to refresh the list.
Success criteria. Set the primary KPI (for example View Rate, ROAS, CPA, CPC, CPM, CTR) and a target value, plus a direction — whether the KPI should land above, below, or within a band of the target. Zimmer sets the direction from the KPI (higher-is-better for ROAS or CTR, lower for CPA or CPC), and you can override it. You can also optimize for a second KPI with its own target, direction, and priority weight; when the two conflict, the primary wins. A readout shows the campaign's current platform setup (objective, delivery optimization) and reminds you that your target KPI is the optimizer's evaluation goal — it doesn't rewrite the platform's own delivery or bidding.
- Per-campaign override. From the campaign list in Scope, open a campaign to give it a different KPI, target, and direction (plus its own budget guardrails) just for that campaign. Campaigns you leave alone follow the agent's defaults.
Daily pacing. Off by default. When enabled, it pauses entities that hit a daily cap and resumes them the next day.
Budget & bid control. Choose reallocation (shift budget between a campaign's sub-units — ad sets or ad groups — keeping the total constant) or elastic (total can grow or shrink with performance). Set a minimum spread, and cap how far budgets and bids can move (max increase / decrease %, with optional hard caps). Bid caps apply to all bid actions (for example ad set / ad group bid, target CPA, target ROAS, max CPC).
Analysis mode. How the agent generates proposals — Judgment (weighs the curated evidence in two passes), Raw (the LLM works straight from the raw data, with no analyst curation), or Deterministic (statistically driven and most consistent — the same data yields the same proposals). Most teams keep the kit's default.
Execution mode. Monitor (advice only), Approve First (proposals wait for you — the default), or Auto (applies changes automatically — proposal-only actions still wait for you).
Notifications. Add email recipients and an optional digest, and connect Slack to post to a channel on a cadence you choose (for example, every run).

Brain

The agent's AI setup, in two parts:

AI Config — the model and how it responds (its instructions and response style). Most teams leave the kit's defaults.
Memory — facts, preferences, and corrections the agent carries across runs. Use Add Memory to write what it should remember, then set a type (fact, preference, correction, or instruction), a strength (from a Rule that always applies, down to a Note that stays available but isn't auto-applied), and a scope (this agent, or narrowed to a specific platform or campaigns). You can add or edit memories yourself.

Activity

This agent's own run history — when it ran, what it proposed, and what's waiting on you. You approve changes in Runs.

Settings

Basic agent details — its name and housekeeping (including deleting the agent). The optimization logic lives in Configure, not here.

Publishing & status

An agent is always in one of three statuses:

Draft — still being set up; it never runs. Every new agent starts here.
Published — live and running on its schedule.
Archived — retired and kept for reference; it doesn't run.

Publishing runs a quick readiness check — a popup that confirms the essentials are set (Configure, triggers) and flags anything missing before the agent goes live. After publishing, edits are saved as a draft and don't affect the running agent until you publish again.

Kits — the kinds of optimizers you can build
How your optimizer works · Runs

Overview How your optimizer works