Every AdOps release, newest first.

The AdOps AI rule builder turns a described goal into rule tasks. You type an outcome in plain language, AdOps assembles a prompt from the live metric catalogue held in its database, sends it to its Claude-backed rule service, and returns a set of tasks with a confidence score from 0 to 100.

Where does the output land?#

In the ordinary rule editor, not in a separate artefact. Generated tasks are pushed into the same task list a person edits by hand, each rendered with an AI Generated badge and a confidence pill. The modal shows example goals, runs a six-stage progress bar while it waits, and locks itself so the window cannot be closed mid-generation. Nothing reaches a campaign until the rule is saved and set live.

What is recorded?#

Each call writes a pending row before dispatch and updates it afterwards with the response body, the response time in milliseconds, the confidence score and a success or error status. Those records are browsable, scoped to the requesting user, and filterable by rule, so a team can review every generation attempt against the rule it was made for. The request times out after 30 seconds.

Two things to know#

The Generate with AI button sits behind a feature flag in the rule builder. And AdOps does not validate the model’s output against its metric grammar before showing it, so read every generated task, check its metric, operator and threshold, and edit before setting the rule live.

The AdOps Benchmark screen answers two operational questions: what is running right now, and what is slow. The Queue Progress tab lists in-flight batches with their start time, the rule, the ad account and an executed-of-total progress bar. The Slow Rule tab filters finished executions by a minimum execution time over a chosen date range.

How does the progress bar know?#

The worker recomputes progress after each campaign in a batch finishes and pushes the whole-number percentage back to the queue, so a batch of 400 campaigns moves in visible steps rather than sitting on a spinner. Each row shows its start timestamp with a relative age beside it, links through to the batch and to the rule, and batches that applied to no items at all are filtered out of the list.

What does the slow-rule profiler show?#

Set a minimum execution time in seconds and pick Today, Yesterday, Last 7 days or Last 30 days. Each matching execution prints the start date, the total time in seconds highlighted in red above 10 seconds, the rule name with the campaign and its ID, and then the detail: the task number and its action, the initial and new values, the per-step timing breakdown, and every condition’s value, operator, target and Pass or Fail badge.

The benchmark summary behind it covers the last 7 days and excludes runs that applied to nothing, so the list is the work that actually touched campaigns.

AdOps stores every ad account it discovers from Meta as inactive. Nothing is refreshed, measured or automated on an account until someone switches it on. The Ad Accounts screen raises a bulk bar the moment rows are selected, showing how many are selected next to an Activate button and a Deactivate button.

Why do accounts start switched off?#

Because the active flag is the platform’s blast radius. It is the single condition that scopes the month-to-date spend refresh, the Purchase ROAS and Aggregated ROAS cards, the spend breakdown, the campaign leaderboard and the account comparison. An account left inactive is discovered and listed but otherwise untouched, which makes connecting a Meta profile with dozens of accounts a safe first step rather than a commitment.

What does the bulk bar do?#

Selecting rows and pressing Activate or Deactivate opens a confirmation dialog. On confirm, AdOps applies the status change to every selected account and reloads the table. Around it sit the controls that make a large portfolio workable: a search over account name and account ID, filter pills for Active, All accounts and Inactive, and a sort menu covering account name, account ID and status.

Syncing from Meta keeps existing choices. Pulling the account list again adds what is new without resetting which accounts you had already activated.

The AdOps dashboard opens on two ROAS cards. Purchase ROAS and Aggregated ROAS are computed across every active ad account for the selected period and shown to three decimals with an x suffix, a trend arrow and a badge reading Profitable at 1.0 and above, Moderate at 0.5 and above, or Low below that.

Where do the ROAS numbers come from?#

Both account-level cards prefer Meta’s own attributed fields, purchase_roas and website_purchase_roas, and are requested with the unified attribution setting so the window matches the one configured on the ad set. Only when Meta returns no attributed value does AdOps fall back to summing action values, de-duplicating the umbrella purchase type against its sub-types so a conversion is not counted twice.

What else is on the screen?#

A Monthly Budget card with a spend bar, the days left in the month and the count of active accounts. A Spend Breakdown ranking the top 5 ad accounts by spend with each one’s share of the total, and a utilisation bar that turns amber at 80% and red at 100%. A campaign leaderboard split into Top Performing and Underperforming tabs, and an account comparison naming the best performer and the account that needs attention.

How fresh are the figures?#

Every widget is cached per user and per period for 1 hour, and the header states whether what you are reading is Live or Cached along with the time it was last updated. Refresh re-requests all five widgets with live numbers.

The AdOps rule log detail view is the evidence for a single run. For every campaign in a batch, AdOps renders one card per task carrying an Executed, Not Executed or Skipped badge, the action’s own parameters, the execution time in seconds and the date of the next scheduled execution.

How does AdOps explain why a campaign was not touched?#

The Condition Evaluations panel lists every condition the engine resolved, each with its metric label, its reporting period, a green Pass or red Fail badge, and the sentence Actual value, operator, Expected value. Nested condition groups render their own AND or OR badge above their children, so a group that failed can be opened to see which individual term was true and what number it saw. A task blocked by its action cooldown is not silently dropped: AdOps writes a skipped result carrying the skip reason and the time the action becomes eligible again.

What proves a value actually changed?#

Budget and name actions write a Value Changes block on the result: the previous value in red, the new value in green. The Action Details column beside it prints the action’s own inputs, such as the mode, the separator and the text for a naming action, or the increase amount, the maximum budget cap and the value type for a budget action, plus the level, the ad account and the timestamp.

Users with the ADMIN role also get an Execution Steps timeline: the numbered stages of the evaluation, each with its message, its elapsed seconds and the pass or fail of the metric it resolved.

AdOps can write live figures into a Meta campaign name. The Add to name action takes a multi-line template in which every brace-wrapped line is expanded at execution time: a date, a clock time, a metric over a chosen reporting period, a campaign field, or a custom metric.

Which tokens are available?#

Six: {date} for YYYY-MM-DD, {time} for HH:MM:SS, {metric|<metric_key>|<date_preset>} for a live Insights value over that period, {metric|<campaign_field>} for a field straight off the campaign object, and {metric|custom_metric|<id>} or {custom_metric|<id>} for a computed custom metric. The builder inserts them for you: the Insert Shortcode modal drops the token at the cursor position and shows a read-only preview of the result.

How does AdOps avoid clobbering a human-written name?#

The action carries a placement mode. TEXT appends the rendered text raw. APPEND writes after a bot separator, keeping the human-written part of the name intact. OVERWRITE replaces everything after the last bot separator, so a name refreshed on a schedule does not grow every run. Two sibling actions handle the rest of the housekeeping: Remove from name strips a substring, and Replace text in name swaps one for another.

Before any of the three writes, AdOps re-reads the campaign name from the Graph API rather than trusting the copy carried in the job, and records the before and after values on the rule result.