Ads and Pixels / AI Optimizer

AI advertising that shows its work

This is AI advertising built the honest way — math decides, the AI explains and proposes, and a separate piece of code enforces the limits. The AI never gets a direct write key to your ad accounts.

Start free See pricing

One place. Every screen.Explore below

01 /

How AI advertising decisions actually get made here

Every recommendation starts as a number, not a guess — the difference between an AI marketing platform you can trust and one that's just automatic advertising with extra steps. The scoring engine reads your results from Meta, Google, Microsoft, TikTok and Snap, adjusts for how long conversions take to arrive on each network, and calculates a probability — not an opinion — that an ad, keyword or audience is beating or missing its target. Only after that math runs does the AI planner get involved. Its job is to read the scores, write the plain-English explanation, and propose what to do next: scale this, pause that, test this audience. It cannot skip the math, and it cannot act on a hunch. If the data is too thin to be sure, the system says so and waits, instead of moving your budget on noise. That split — math decides, AI explains — is the whole design, and it's why every card you see comes with the number behind it, not just a sentence that sounds confident.

Every action gets a reason and an undo

This is what AI advertising with a real reason looks like. When the optimizer acts — or proposes an action for you to approve — it writes a card like this: "Paused 'Hook B' because it spent $142, 2.8x your $50 target, over 9 days with 0 purchases; 94% probability its true cost is above target." Every number in that sentence comes straight from the scoring engine, and a check confirms each one exists in the underlying data before the card is ever shown — the AI can't invent a statistic to sound more certain. Every action, whether you approved it or the optimizer took it on autopilot, has a one-click undo that replays the exact reverse operation. Nothing is ever deleted; ad sets and campaigns are only paused, so undo always has something to restore. An append-only audit log keeps who or what acted, the inputs, the reasoning and the network's own response for every change.

Explore Reporting

02 /

Guardrails that live in code, not in a prompt

Most AI advertising tools ask you to trust a black box. This AI advertising platform doesn't — the planner is never given a key that writes to your ad accounts — only the executor has one, and the executor only runs plans that pass a fixed set of rules first. Budget or bid changes are capped at 20% per step, with at least 72 hours between changes on the same ad set or campaign, because bigger or faster moves reset a network's learning phase and throw away its own progress. No more than 10–15% of a client's total budget moves in any single week. Nothing is touched while an ad set is still in a network's learning phase. The optimizer never edits a conversion event, an attribution setting, or the optimization goal — those stay in your hands. Brand keywords are protected from pausing by rule. And every guardrail is enforced the same way whether a human approved the change or autopilot did — the limits don't loosen just because no one is watching.

03 /

Autonomy you set, per client and per action

AI advertising autonomy isn't all-or-nothing here — you choose how much control to hand over, and you can change it at any time. Observe shows what the optimizer would do without touching anything. Suggest and approve — the default for every new client — puts every proposal in a queue for a person to accept or reject. Autopilot within limits lets the optimizer act on its own for the lowest-risk moves: pausing a clear loser or moving budget within the ±20% guardrail. Autopilot with a daily digest goes further but still reports every move in one email a day, in plain English. Whatever tier you're on, some things always wait for a human: launching a new campaign, building a new audience, publishing creative, and any change to bid strategy. If an anomaly opens on a client — a tracking outage, spend running away — automation for that client freezes until a person clears it.

Explore AI Media Planner

04 /

Proof, not a promised number

Good AI advertising has to prove itself, not just claim it. We don't publish a made-up "average return" because there isn't a real one — every account, industry and budget is different, and any tool that quotes you a single number is quoting a marketing claim, not a result. Instead, 10–20% of eligible campaigns are held out on rules-only management for 8–12 weeks, with no AI budget moves at all, and compared against the AI-managed group on real revenue and leads from your own backend — not the platform's own reported return, which each network is incentivized to report generously. That holdout comparison, not a case study, is what we'll publish as proof the optimizer is actually helping. Ad tests follow the same honesty rule: before a test runs, we calculate the sample size it needs, and if your budget can't reach that size in 14 days, the test simply doesn't run rather than reporting a "winner" that's really noise.

05 /

What the AI advertising optimizer doesn't do

It doesn't have a key to write directly to Meta, Google, Microsoft, TikTok or Snap — only the guardrail-checked executor does.
It doesn't touch your conversion events, attribution windows, or optimization goals, because those decide what "winning" even means, and that call stays with you.
It doesn't launch a new campaign, publish new creative, or build a new audience without a person approving it first, at any autonomy level.
It doesn't delete anything — ad sets and campaigns are paused, never removed, so every action can be undone.
And it doesn't act on a day's numbers before enough of that day's conversions have actually arrived; judging early data as if it were final is one of the most common automation mistakes, and the system is built to wait instead.

Explore MCP and API

A little more clarity

Frequently asked questions

Real answers before you make a decision.

Does the AI ever spend my budget without approval?

Only if you turn on an autopilot tier for that client, and only within the fixed limits: pausing a clear loser or moving budget by at most 20% per step, at least 72 hours apart. That's the boundary of what our AI advertising engine can do on its own — new campaigns, new audiences, published creative and bid-strategy changes always wait for a human, no matter which tier you're on.

What stops it from resetting my learning phase?

The guardrails are the same rules practitioners use to avoid a learning-phase reset: no more than a 20% budget or bid change per step, and at least 72 hours between changes on the same ad set. Nothing is touched at all while an entity is still inside a network's learning phase.

How is this different from Meta Advantage+, Google Performance Max, or other AI ads tools?

Those automate one network on their own terms, with no explanation and no undo. Our AI advertising optimizer works across networks, shows the number behind every action in plain English, lets you undo any single change, and proves its results against a real held-out control group instead of the platform's own reported return.

Is this AI marketing automation, or does a person still make the real decisions?

Both. Day-to-day AI ad optimization — pausing a clear loser, small budget moves within the guardrails — can run on autopilot once you turn it on. Anything bigger, like a new campaign or a bid-strategy change, still needs a human.

Can I see why it paused or scaled something after the fact?

Yes. Every action — human-approved or autopilot — is written to an audit log with the inputs, the reasoning, the exact numbers, and the network's response, and it's visible from the same card that showed the original recommendation.

What happens if something breaks, like a tracking outage?

The anomaly detector opens an alert and automation for that client freezes immediately, before it can act on bad data. A person has to clear the anomaly before automated changes resume — no AI advertising decision runs on data it can't trust.

Is there a minimum spend to use the AI optimizer?

No minimum is set by us; the AI advertising engine needs enough real conversions to judge a change confidently, so a very small account will see it wait longer between moves, not be blocked from using it.

Which networks does the optimizer manage right now?

Meta, Google, Microsoft, TikTok and Snap have full read/write control at launch, so AI advertising decisions apply across all five from day one. Other networks are reporting-only until their write access ships; the platform never claims control it doesn't have yet.

Your next move

See what it would suggest before you turn anything on

Connect an account and start on Observe — the optimizer reads your results and shows its reasoning with nothing turned on. Move to Suggest & approve whenever you're ready.

Start free See pricing