How does AdsGuru work?

Updated November 12, 2026

The daily loop at a glance

The loop has six stages, and the whole point is that only one of them needs you: the approval.

1. Scan: every morning AdsGuru reads both ad accounts over the official APIs - spend, conversions, CPA or ROAS, click-through rate, CPM, frequency, learning status, pacing, disapprovals.
2. Diagnose: movements beyond the account's normal noise are attributed to a cause: creative fatigue, auction pressure, a tracking problem, or a budget issue.
3. Propose: each fix becomes a proposal with the change, the reason, the evidence, and the expected effect.
4. Approve: you approve or reject in one click. Approved changes apply immediately through the platform API.
5. Autopilot (optional): routines you have marked as trusted apply automatically inside caps you set.
6. Log and report: every applied change is recorded with before and after values, and a short plain-language note tells you what happened in the account and what changed.

The rest of this page walks through each stage in enough detail that you could audit the logic yourself.

Stage 1: what the scan reads

The scan reads performance and configuration data from the Google Ads API and the Meta Marketing API, read-only at this stage, every day.

From Google Ads it pulls campaign and ad group spend, conversions and conversion value, cost per conversion, impression share data, ad disapprovals, budget pacing against your plan, and bid strategy status. Weekly, it pulls the search terms report, including Performance Max search themes, for the search terms cleanup.

From Meta it pulls spend and results per campaign, ad set and ad, frequency, click-through rate, CPM, learning phase status per ad set, and delivery issues. It also checks that pixel and Conversions API events are arriving and deduplicating, because every other decision depends on the conversion data being real. What the scan never touches: your billing settings, your audiences' personal data, or anything outside the connected ad accounts (details on the data page).

Stage 2: how diagnosis separates signal from noise

Diagnosis starts from a baseline: what does normal volatility look like in this specific account? Ad metrics bounce daily, and reacting to every bounce destroys accounts. A one-day CPA spike on 3 conversions is noise; a six-day trend across 80 conversions is signal.

When a movement clears the noise threshold, AdsGuru attributes it to a cause by checking the usual suspects in order. Conversions collapsing while clicks hold points at tracking, because real demand rarely vanishes overnight while traffic stays flat. Rising CPM with stable click-through rate points at auction pressure, meaning competitors are bidding more. Stable CPM with falling click-through rate and rising frequency points at creative fatigue. Spend pacing far above plan points at a budget or bid strategy change, often an auto-applied recommendation.

This cause gets written into the proposal, which is what makes the proposal checkable: you can open the platform and verify the claim before approving.

Stage 3: the anatomy of a proposal

Every proposal carries the same fields, so you can evaluate it in under a minute:

Proposals arrive batched rather than as a constant drip. Urgent items, like a tracking break or runaway spend, alert immediately instead of waiting (see alerts and monitoring); routine optimization waits for the batch so the account is not thrashed.

Stage 4: approve, reject, or tune

Approving applies the change immediately through the platform API, writes it to the change log, and confirms it in the next report. Rejecting dismisses the proposal and nothing changes in the account.

If a type of proposal keeps arriving that you keep rejecting, tune the routine behind it instead of rejecting forever. Every routine's thresholds are adjustable: the CPA level that triggers a pause proposal, the minimum spend before an ad is judged, the maximum daily budget movement. Tightening a threshold reduces proposal volume; loosening it catches more. Routines are explained here.

Stale proposals do not linger: if the underlying data changes, a proposal is withdrawn or refreshed rather than waiting to be applied against a situation that no longer exists.

Autopilot: optional, scoped and capped

Autopilot is per routine, never account-wide. You might put search terms cleanup on autopilot after three weeks of approving its proposals unchanged, while keeping budget moves in approval mode forever. Both are normal.

Every autopilot routine runs inside caps: maximum budget movement per day, minimum spend thresholds before action, floors and ceilings per campaign. Autopilot changes are logged exactly like approved ones and appear in the same daily note, so you always see what ran. You can demote any routine back to approval mode in one click.

Two things never run on autopilot: structural changes (consolidating campaigns, restructuring ad sets) and anything that would exceed a cap. Those always come back to you as proposals.

The change log

Every applied change, approved or autopilot, is logged with the timestamp, the routine that produced it, the reason, the before and after values, and who approved it. The log is the answer to 'what changed and why' at any point later.

The same changes are independently visible in Google Ads change history and Meta's activity history, because AdsGuru writes through the official APIs. That means you never have to take the log's word for it: the platforms' own records match, and your access can be revoked at any time without asking anyone.

The first two weeks are different

For the first 7 to 14 days after connecting, AdsGuru is building its baseline, and it says so rather than pretending otherwise. AdsGuru launched in 2026 and is honest about this: proposals carry weight only once there is enough of your account's own history behind them.

What you get immediately is the audit: tracking checks, quantified waste in search terms, settings traps, and structural red flags, because those do not need a behavioral baseline. What matures over the first two weeks is everything trend-based: fatigue detection, budget reallocation, pause thresholds calibrated to your account's real volatility rather than generic defaults.

The practical advice: keep everything in approval mode for at least the first two weeks, approve what checks out, and treat the period as AdsGuru interviewing for the job. Setup steps are on the getting started page.

A worked example of one morning

Here is a hypothetical morning, with illustrative numbers, for a store spending about $150/day across both platforms.

The scan finds three things. On Meta, the main prospecting ad's frequency has crossed 3.2 and its click-through rate has fallen six days straight; the diagnosis is fatigue, and the proposal is to rotate in the approved variant that has been waiting, with the old ad paused once the new one exits learning. On Google, the week's search terms include a cluster of 'free' and 'jobs' queries that spent $31 with zero conversions; the proposal is nine negatives added to the shared junk list. And the Shopping campaign has been pacing 15% above plan; the proposal is a budget trim with the excess shifted to the Search campaign currently beating its ROAS target.

You read three proposals, approve two, reject the budget shift because a sale starts tomorrow, and close the tab. That is the loop working as intended: minutes of review, with the reasoning visible.

Common questions

Related