Your agent can now perform deep analysis on campaign delivery and performanceExtends the MCP toolset into the campaign information behind our ‘Campaign
Optimization’ product. Your agent can see delivery data, targeted cohorts, and
day by day trends, combine these with the audience and cohort analysis tools,
and use them to diagnose a campaign and recommend optimizations.Status: Closed testing · Invitation only · Tool interfaces may change
New capabilities
Report on a finished campaign without assembling it by hand
“Give me a report on the campaign we ran for Organic Oats, showing total impressions and CTR achieved”Ask for a campaign by name and an agent pulls the campaign performance results. Performance can be broken down by:
- Cohort: How delivery was concentrated on the cohorts that actually performed
- Day: Show day-by-day trends of performance
- Line-item: How did one line item perform against another? Which was the most successful?
Optimise while the flight is still live
Campaign [campaign name] is live with [number] days remaining. Diagnose it and tell me what to change. Start with targeted cohorts indexed against the campaign average — where is delivery concentrated relative to performance? Then check line items grouped by creative format. Give me a ranked list of changes with expected effect and confidence.The most common finding is also the easiest to act on: budget weight sitting on a weak cohort while a strong cohort takes a fraction of the impressions. It requires no new targeting — just a reallocation. Creative format is usually the largest lever available mid-flight. Formats on the same campaign regularly separate by an order of magnitude on CTR, a far bigger gap than anything found between cohorts or between placements. Where budget can move between formats, that is the first recommendation to make. Tell the agent how many days the flight has left. The time remaining determines which recommendations are worth making at all.
Ground a pitch in what has already delivered
Check whether these cohorts have been targeted in past campaigns and pull the delivery figures. Group the line items by creative format so I can see which formats performed best for comparable advertisers. Reference other advertisers by category, not by name.Any cohort can be checked for its track record as a targeted buy. This replaces a projection with a proof point — “this audience has run for two comparable advertisers and delivered this CTR at scale” closes better than a reach number. Check every candidate, not only the one you plan to lead with. History is sparse and unevenly distributed: some cohorts have run many times, others twice in an irrelevant category, others never. The proof point often comes from the cohort whose name looks less on-brief, which is exactly why it is worth asking.
Audit your own setup
For campaign [campaign name], show me the share of delivery that matched our Permutive targeting, broken down by line item. Flag any line item delivering at scale with none of our targeting applied.Delivery carries the share of impressions that actually matched your targeting. Whole line items sometimes run at scale with none of it applied, usually concentrated in a single format or environment. This goes unnoticed unless something surfaces it, and it is much better found internally than in a client meeting.
Find expansion candidates from real delivery
Pull every cohort present in the delivered audience for [campaign name]. Show me who this campaign is actually reaching and how that compares to what we targeted, then shortlist the cohorts worth investigating as additions.A campaign’s delivery reaches far more cohorts than it targets. An agent can read that observed pool as a candidate list for expansion — a cohort has to appear there to be a sensible thing to add. See Worth knowing below before relying on this to find outperformers.
How it works
Two read-only tools over the GAM delivery plane:How are recommendations formed?The power of agents is that they combine capabilities (tools) in ways that best
match their tasks and needs. By combining the already-available
search_cohorts,
list_cohorts and
measure_audiences tools, agents are able to pick
new cohorts to recommend for campaigns. They can also analyse past performance
for cohorts to determine likely future performance.v1.2 of the Permutive MCP will release enhanced
analysis and recommendation tools, to give the agent ability to recommend
cohorts based on indexing and overlaps and time-line trends.