Supermetrics lets Claude build ad campaigns that stay paused until approved

A campaign drafted in a Claude chat can now land in Google Ads, Meta or LinkedIn and sit there, paused, until someone signs off on it.

The analysis side already has an audience. Supermetrics said in its PR Newswire announcement that more than 6,000 of its customers have used its Claude connector since the February launch, and that active users have risen by an average of 250% month over month. The company calls it one of its fastest-adopted launches. Paid-media reporting is currently the dominant use case.

So the open question is what happens when the chat that explains why performance changed can also propose the fix. For a performance marketing lead or an agency operations head, the answer comes down to two controls shipping alongside write access: an approval gate that keeps new campaigns paused by default, and a set of business rules the company calls Context Layer.

Six ad platforms, one approval gate

The write path covers Google Ads, Meta, Microsoft Advertising, TikTok, LinkedIn and Snapchat. On the analysis side, Claude can pull from a wider set of connected sources: ad platforms, CRM, SEO, ecommerce and data warehouses. A marketer can ask why a channel slipped and then review a proposed campaign change in the same conversation.

New campaigns created through the integration are paused by default. Marketers can review proposed edits to existing campaigns before they're applied. Customers can also set approval requirements for different accounts. This matters most where accounts carry different stakes, such as a team that wants tight sign-off on its biggest budget lines and lighter review on test accounts.

Thuneberg described the problem as a gap the last wave of AI tools left open. In the release, he said AI has made asking questions far easier, but marketers are still left to close the distance between receiving an answer and acting on it. The company describes the end state as analysis, recommendations and approved action, all without marketers giving up control of campaigns or budgets.

That budget line explains why the gate sits where it does. A paused campaign costs nothing, so the approval step is the point where a model's proposal turns into spend. Everything upstream of it, from the diagnosis to the draft, can move at chat speed without touching money.

Co-CEO and founder Mikael Thuneberg described the design goal plainly: the AI handles the manual steps between spotting an opportunity and acting on it, and the marketer keeps the final say.

The rules marketers write down

When Claude asks for campaign information, Supermetrics returns normalized data, so nobody has to export and reconcile figures from each platform by hand. On top of that data, teams can set their targets, naming conventions, reporting rules, events and client-to-account mapping in plain language, at the team, platform or account level. Those settings carry across conversations, so nobody has to re-explain how the marketing operation works with every prompt.

While the AI only read data, those settings mostly shaped reports. Once the AI can write campaigns, a team's naming convention is an instruction the system follows. The same logic applies to client-to-account mapping, which could be what separates a correctly placed proposal from one aimed at the wrong client's account. Thuneberg made the related point in the release: a general-purpose model doesn't know a company's business, how its data fits together, or what good performance looks like for it.

Supermetrics is betting its positioning on that gap. The company argues that the next battleground for marketing AI won't simply be access to ever more powerful models. Instead, it will be whether those models can reach the trusted marketing context they require to make useful decisions. In practice, that pitch gives marketers a conversational front end across their stack while the data infrastructure and controls stay underneath it.

The bet has a practical edge for buyers. If the model itself is interchangeable, the durable asset is the written-down business context, which suggests the hours a team spends codifying targets and mappings in Context Layer are the real implementation cost, not the connector setup.

The one customer result Supermetrics cites comes from the analysis side. Layer, a Norwegian digital marketing agency, cut a bespoke client-reporting task from up to 10 hours of spreadsheet work to roughly 20 minutes using Supermetrics and Claude. It's a useful benchmark for reporting time. It isn't evidence about campaign creation, which is the new capability and doesn't yet have a published customer figure.

The handoff Supermetrics' own survey flagged

This product direction lines up with a report Supermetrics released in July. For its AI Readiness Gap report, the company surveyed 435 marketing leaders in the U.S., U.K., Germany, Australia and Singapore and asked what keeps teams from acting on an insight. The biggest blocker, named by 40% of SMB marketing teams and 34% of enterprise teams, was poor integration between analytics tools and activation platforms. Slow manual data handoffs came up for another 20% of SMB teams and 27% of enterprise teams.

The Claude write path goes straight at the first of those blockers: the diagnosis and the campaign change now happen in one conversation rather than across two disconnected tools. The manual-handoff group gets help more indirectly, through the normalized data Supermetrics returns when Claude asks for numbers. For operators facing either problem, a write path inside the same conversation could shorten the step from insight to action.

Still, the report's broader point concerns ownership. Supermetrics uses the term "AI readiness gap" to describe how far an organization's real ability to deploy AI accountably falls short of the pressure it feels to adopt the technology. In its view, this is more an ownership and execution problem than a technology problem. The report also found a mismatch in who controls what. Teams held responsible for AI-driven performance frequently have no say over the data strategy, the system integrations or the activation workflows, and those are the things that determine whether AI delivers reliable results.

That finding lands squarely on the approval gate. Per-account approval rules only work if someone clearly owns sign-off for each account, so the first setup decision for many teams could be organizational rather than technical: who approves what, and on whose behalf.

Two more figures from the same report sharpen the picture. Among respondents, 42% said their wait time for ad hoc data doesn't meet their needs, and just 15% reported a defined AI roadmap with measurable success metrics.

Those numbers also cut against the speed argument. An approval gate stops a bad campaign from spending money. It does nothing about a sound-looking proposal built on weak data, and the report itself warned that without stronger data ownership, AI risks speeding up broken workflows instead of improving decisions. CEO Anssi Rusi made a similar point in July, saying AI needs high-quality data, transparency and trust, and that marketers need to understand what it does and keep control of it. Keep in mind too that the survey comes from the vendor, and its 85% figure is an early read from a poll that's still running.

Which teams benefit most depends on what they've already built. Those with clean cross-platform data and written-down naming and mapping rules have most of the Context Layer inputs ready. Teams still waiting days for answers to ad hoc data questions will likely spend their first weeks on that groundwork before any AI-drafted campaign is worth reviewing.

From query tool to campaign builder in a year

The roadmap has been built in steps. In November 2025, Supermetrics launched Supermetrics AI, a suite that included built-in agents and an MCP Server letting outside AI tools such as Claude and ChatGPT query a company's marketing data through Supermetrics' Enterprise APIs. One of those agents, the Insights Agent, was pitched as closing the "now what?" gap by analyzing data and recommending next best actions. Recommendations, at that stage, still ended with a human doing the work in the ad platform.

The launch video described the core problem as decision velocity: how fast a team can get from insight to decision to action. The company said in that video that data access was a problem it had solved years ago, and it described a proprietary knowledge base and guardrails meant to keep AI output accurate and consistent. Context Layer reads as the customer-editable version of that idea, and the write path is the next step.

The demand case Supermetrics made at the time came from its 2025 Marketing Data Report, which said marketing leaders were dealing with 230% more data than in 2020, and that more than half couldn't make sense of it. A year on, the company is moving from helping marketers read that data to helping them act on it.

The scale behind that move is large. Supermetrics says it's used by more than 200,000 businesses, and its company description puts it in 120 countries, processing 15% of global advertising spend across more than 100 data sources. Each product release has moved closer to the ad account itself.

Thuneberg also signaled that the company is testing the approach on itself. He said that if Supermetrics asks marketers to trust AI with increasingly important decisions, it also has to be willing to apply AI to its own hardest problems. The stated direction goes further than a chat window. It describes a model in which data and business context feed analysis on a continuous basis, flag meaningful changes, recommend next steps and allow approved action inside predefined guardrails.

There's a reason to expect a slower ramp on the write side. A separate Supermetrics release, headlined on its PR Newswire page, reported that campaign optimization ranks last in AI adoption among retail, ecommerce and CPG brands. That makes reporting's early lead unsurprising. The next adoption figures Supermetrics publishes will show whether campaign creation and editing join reporting as a main use, or whether marketers stick with asking questions and keep making campaign changes themselves.