Case study: Autonomous ad ops across thousands of accounts
- Client: one of the big five marketing groups
- Scope: paid media across Meta, Google and TikTok
- Platform: Atlas Custom Agents and Runtime, deployed in the group’s environment
- Shape: pilot on 47 accounts, then rollout across the network
Results
Account-manager hours freed
64%
The preparation and review work before every decision now runs on its own.
Return on ad spend, pilot average
+22%
Budget and bid changes proposed per account against its targets, approved by the account manager.
Reporting per account
2 h → 12 min
Data pulled by API, changes marked, commentary drafted, on a schedule.
Pilot → rollout
47 accounts → network
A measured pilot against a baseline gave the group the evidence to roll out across thousands of accounts.
Figures as reported for this engagement: the pilot’s results against the baseline agreed at the start. Client name withheld under the terms of the engagement.
The situation
Every account needed a regular performance review. Account managers pulled exports from each ad platform, rebuilt the same spreadsheets, wrote the commentary and only then got to the decisions that move spend. About two hours of each cycle went on preparation, per account, before anyone could act.
Optimisation lagged the data by days. Across thousands of accounts, the hours lost to preparation were the biggest single cost in the operation, and the easiest to measure.
What we built
One decision layer over every platform. Atlas connected to the ad platforms directly, through their APIs, in place of manual exports. Every account got a unified cross-channel view, scheduled reporting with first-draft commentary, and agents that proposed budget and bid changes against the account’s goals.
Changes to spend kept the approvals the group already required. Access followed the client and account structure, and every action was recorded for review.
- A review is due
- the data is already together
- changes are proposed
- the account manager decides.
Connect.
Meta, Google and TikTok accounts linked by API. No CSV uploads, no stale data.
Report.
Scheduled workflows assemble each account’s performance, mark what changed and draft the commentary.
Recommend.
Agents propose budget, bid and creative changes against the account’s targets, with the reasoning shown.
Act, within approvals.
Spend changes go to the people who sign them off. Every action is logged.
How the value was established
Define the opportunity.
The goal was agreed as hours returned to account managers and better return on spend. The baseline was measured first: about two hours of preparation per account per reporting cycle, and each account’s ROAS before the pilot.
Put AI into operation.
47 accounts went live with connected reporting and recommendations. Account managers kept the final say on every spend change. Our engineers worked alongside the account teams through go-live.
Establish the value.
Measured against the baseline, reporting fell from two hours to twelve minutes per account, 64% of account-manager hours were freed, and ROAS rose 22% on average across the pilot. On that evidence the group took it to the network.
What this means for you
This is how Atlas engagements run. Custom Agents for the workflow, Runtime to run them inside your environment, Sylo to govern what leaves your network. The method is the one above: a baseline, a pilot and the evidence to expand.