Every Pace run is a pipeline. The AI gets one shot to recommend a budget per campaign, and before that recommendation lands on your ad platform it has to clear a stack of safety checks that you and your account both control.
The same pipeline runs whether the campaign lives on Google Ads, Meta Ads, LinkedIn Ads, or Microsoft Ads. The data Pace fetches and the signals it reads change per platform. The safety model around the AI doesn't.
Here's what runs, in order.
1. Permission and safety preflight
Pace runs four checks before anything else:
- Pacing is enabled for your workspace.
- The account isn't currently in an overspend hold.
- You haven't manually excluded the account from pacing.
- The account is a standard ad account, not a manager or agency-level parent like a Google Ads MCC. Pace only writes to standard accounts.
If any of those fail, the run ends and no budgets move.
2. Fresh data fetch
Pace pulls three windows of performance data from the ad platform:
- A long-term trend window covering spend, conversions, CPA, ROAS, and CTR. This tells Pace which campaigns are stable performers.
- A recent diagnostic window. The signals here depend on the platform: Quality Score, impression share, and auction insights on Google Ads and Microsoft Ads; audience and placement performance on Meta Ads; engagement signals on LinkedIn Ads. This is the why behind the trend numbers.
- Yesterday's pulse, so Pace can see how the previous run's changes landed.
Pace also pulls month-to-date spend separately, because that one number drives most of the math that follows.
3. Memory
Pace carries context across runs. On every run it also loads:
- The strategic notes the AI wrote last cycle.
- The last action it took, with its reasoning.
- Observations the system has confirmed about your account over multiple runs.
- A verdict on its own recent predictions: confirmed, contradicted, or inconclusive.
That last piece is the calibration loop. Pace tracks whether what it predicted would happen actually happened, and the next run sees that track record.
4. Numerical sanity check
Pace validates its inputs before calling the AI. Month-to-date spend has to be a real number. Your monthly budget has to be positive. The calendar math has to be internally consistent.
If something upstream is broken (a timezone glitch, a missing field, a non-finite value), Pace aborts the run and changes nothing.
5. AI recommendation
The model gets the data and memory, plus your goals, constraints, schedule, and any context you've added on the account. It returns:
- A recommended daily budget for each campaign.
- The reasoning behind each recommendation.
- An updated strategic note to carry into next time.
- An optional prediction about what should happen as a result.
This is the only non-deterministic step. Everything else, before and after, is deterministic.
6. Safety clamps
Every recommendation from the AI passes through two layers before it can be written:
- A floor that prevents budgets from going below the ad platform's minimum daily spend. Each platform has its own floor, and Pace uses the right one for the campaign.
- A ceiling derived from your monthly budget, the days remaining in the month, and the campaign's previous budget. The strictest cap wins.
That means a single day can't burn a disproportionate share of your monthly budget, and a single campaign can't jump wildly day-to-day. Even if the AI was wrong, the worst case is bounded. If a clamp fires, the change record shows which one and why.
7. Per-campaign write
Pace updates one campaign at a time, with its own success or error status. A single failed campaign doesn't block the others, and a platform-side error on Meta Ads (for example) doesn't stop Google Ads writes in the same run.
Every change (successful, errored, or capped by a clamp) is logged with the old and new budget, the reasoning, and which guardrails fired. That's what shows up in your Changes tab.
8. Outcome measurement
A day or so after each change, Pace looks back and measures what actually happened to spend, conversions, and ROAS. It writes a verdict on whether the AI's prediction held up.
That verdict feeds into the next pacing run as part of the model's memory. Over time the AI sees its own track record alongside fresh data, which is what closes the calibration loop.
What the AI can and can't do
The AI can recommend a budget in any direction and shift spend from poor performers to strong ones aggressively. It can also exit a self-imposed hold when the data supports it.
The AI cannot:
- Push a budget to your ad platform without going through the floor and ceiling clamps.
- Spend a disproportionate share of your monthly budget in a single day.
- Run when your workspace's pacing kill switch is off.
- Run on a manager or agency-level account.
What you control
- The workspace pacing kill switch turns the whole system off in one click, across every connected platform.
- Per-account exclusion opts a specific account out while leaving the rest enabled, regardless of platform.
- Account budgets, schedules, and goals shape every recommendation the AI sees.
- The Changes tab gives you a full audit trail of every action, including the reasoning and any clamps that fired.
The Changes tab is the single source of truth for what Pace did on your account and why.