Optimizations

Optimizations is where you review what your agents propose and decide what happens. Open it from the left nav. It gathers every recommendation waiting on you across all your optimizer agents, grouped by the campaign they affect, with the evidence behind each one a click away.

What an agent proposes depends on the kit — budget, bid, targeting, and creative changes, or a warning that a budget or spend limit needs attention.

The screen has two tabs: Awaiting review, where you decide, and Activity, the record of what has already happened. If your agents run in Auto mode, a third Auto tab appears alongside them.

Awaiting review

Everything waiting on a decision, newest analysis first. Recommendations are grouped by the campaign they affect, so you judge a campaign's changes together rather than one at a time.

Narrow the list:

• Search across campaigns, ad sets, and changes.
• Sort by — Impact (the default), Confidence, Risk, Share of spend, Share of impressions, Newest, or Oldest.
• Group by — Account structure (campaign, then ad set) or Change type (all the budget changes together, all the exclusions together, and so on).
• All agents — narrow to one optimizer.
• All status — Waiting · Needs you · Waiting for Script · Approved · Handled · Rejected · Failed. Handy for finding the ones that need you to act in the ad platform yourself.

What each campaign row shows

• the campaign name, with the platform it runs on;
• how many proposals it holds, and how many you've already decided;
• the primary KPI and its target, and the secondary KPI where one is set;
• the analysis window the agent looked at — for example Sep 22 – Sep 28, 2026 · 7 days;
• the agent that produced them, as a link to that agent.

Expand a row to see the individual recommendations inside it.

When there's nothing to decide, the tab reads Nothing awaiting review — "When the agents next run, anything they propose will appear here." If you haven't built an optimizer yet it points you at Start guided setup or Browse Kits instead.

Why this one

Select any recommendation and the panel on the right shows the evidence behind it. Until you pick one it reads "Pick a recommendation on the left to see the evidence behind it."

The numbers behind it — the value being changed, set against the things that justify the change: how this ad set or line item compares with the campaign, with your brand norm, and with your target. Underneath, the sample it rests on: impressions, clicks, and conversions, and over how many days.

Safety check — whether the change is allowed, blocked, or was adjusted to stay inside your limits (with the value it would otherwise have used). This is where you see your budget and bid caps doing their job.

Goals in force — the main goal the agent is working to, and the second goal where you've set one.

AI summary — the reasoning in plain language.

What else this changes — any knock-on effect the change carries, so nothing lands as a surprise.

Reviewing proposals

A campaign's Proposals are the specific changes the agent recommends. At the top you'll see counts — total, pending (awaiting you), blocked, expired, and manual (needs applying by hand) — and a summary line like "Reviewed 434 cells across one campaign — 137 changes recommended."

Search, sort, and group the list:

• Sort by — Impact (default), Confidence, Risk, Spend share, or Impression share.
• Group by — Ad set (default) or Lever, which sorts proposals by the kind of change: Budget, Bid, Bid Modifiers, Status, Targeting, Schedule, Scale opportunities, and Other.

Most proposals trim what isn't working. Scale opportunities are the opposite: the agent has found something worth adding — a keyword, an audience, a hashtag, or a channel that resembles what already performs for you — and proposes adding it rather than cutting something.

Whichever you choose, proposals are always grouped by campaign first, then by your Group-by choice within each campaign. A proposal's campaign name links to its full Campaign Insights detail — so you can dig into the data behind any recommendation before you act on it.

A blocked proposal can't be applied right now (a guardrail or platform limit is in the way); an expired one wasn't actioned within its approval window. Neither applies on its own, but both still carry the agent's recommendation and reasoning — worth a read, since you might make the change yourself or adjust what's blocking it.

What each proposal shows

Collapsed, a proposal already shows a lot: the action (for example, "Campaign.Set Budget → decrease"), an impact rating (stars, with its share of spend and impressions), its status (Awaiting approval, Skipped, Expired, and so on), a confidence chip (High / Medium / Advisory), how the references line up (for example, "target & peer agree"), and a plain-language AI summary of what the agent found and why.

Select Show more for the full Facts table — how the entity's value compares to your target, its peers, and your brand norm, with each reference casting a Worse / Better / Abstain vote plus a corroboration score. That's the detailed evidence behind the recommendation.

When proposals arrive as a group

If an agent wants to exclude two or more things of the same kind on the same target — say four placements on one ad group — they arrive as one card instead of four separate rows, so you can judge the set together rather than one value at a time.

Each value has its own checkbox, and the card shows what each one is costing you — spend, impressions, conversions, the secondary KPI where there is one, and the change proposed. A line underneath puts the set in proportion, for example "The remaining 12 are £340 between them, none above £45."

Use Approve all to take the whole set, or tick just the ones you agree with and use Approve selected. Anything you leave unticked stays pending and will be proposed again the next time the agent runs — rejecting is a separate, deliberate action. Long lists collapse behind Show all · pick individually, with a filter box for finding a specific value.

On an advisory group there's nothing to push to the platform, so the buttons are Mark handled (you did it yourself) and Dismiss instead of Approve.

Approving changes

When the agent is set to Approve First, you act on its proposals here:

• One recommendation — Approve it, Edit the proposed value first, or Reject it. If a change has no automatic lever (it can only be made in the platform's own UI), its button reads Mark handled — make the change yourself, then mark it handled to record who dealt with it.
• Several at once — tick the ones you want and the bar shows "N recommendations selected", with Approve, Reject, and Clear. This is the fastest way through a long list once you've read the evidence.

Approved changes are applied; rejected ones are not.

Changed your mind

Right after you approve a batch, a message reads "Approved N recommendations" with an Undo beside it. Undo pulls them back before they reach the platform.

The window is short and deliberate — once Zimmer starts sending a change, it can't be recalled, and you'll see "Too late to undo — already applying." After that, reverse it the way you would any other change: adjust it yourself in the platform. Anything already undone shows in Activity.

When you reject something, Zimmer asks why — pick a reason (Wrong KPI · Too aggressive · Not enough evidence · Wrong campaign or context · Timing is off · Prefer manual review · Other) and add a note if it helps.

The reason isn't just bookkeeping: it feeds back into the agent, which waits longer before suggesting anything similar. If you never want that suggestion again, tick Don't suggest this again — the button then reads Reject & don't suggest again, and the agent stops proposing that specific change for good.

Activity

The record of what has already happened, organised by analysis batch — one batch each time an agent runs. Each batch shows what the agent looked at and what came of it.

Narrow the list:

• Search campaigns or agents.
• All agents — narrow to one optimizer.
• All campaigns — narrow to one campaign.
• Last 14 days (the default) — or 24 hours, 7 days, 30 days, or 90 days.
• All status — Waiting · Needs you · Waiting for Script · Approved · Handled · Rejected · Failed.

If a batch is still running, or failed, this is where you'll see it.

What the statuses mean

Where your plan is leaking

Beyond individual proposals, Zimmer can show where a plan is losing ground — breaking the journey from impression to conversion into stages and pointing at the stage that's costing you, so you can tell "not enough people are seeing it" apart from "plenty see it but few act".

Each finding shows which stage it came from and what evidence sits behind it.

How agents act: the three modes

You choose how an agent acts when you set it up:

Approve from anywhere

When an agent is set to Approve First, you can review and approve (or reject) from:

• the Optimizations desk,
• the approval email Zimmer sends, or
• Slack.

Related

• Create an optimizer agent — set an agent's mode and what it's allowed to do
• Campaign Insights — see the results of approved changes