
Pixeon's Growth team runs paid media for two very different B2B healthcare companies: Pixeon, with systems for hospitals, clinics, laboratories and radiology centers and a long sales cycle, and BoaConsulta, management software for clinics and doctors' offices, with more volume and a shorter cycle. The goal was an AI agent that does more than read the accounts: it reads the data every day, flags what is off, proposes an action with a reason and executes what gets approved. But with a long cycle, the ads manager shows the lead while the sale happens weeks or months later, in the CRM. For the agent to judge a campaign, it had to see the whole funnel, from click to sale. So before thinking about agents, the team built the data foundation on Nekt.
Seven sources feed the base, so every answer starts from the full funnel:
| Systems | What they power |
|---|---|
| Meta Ads, Google Ads, LinkedIn Ads | Spend, campaigns and execution through MCP |
| GA4 | Site behavior between the click and the lead |
| Salesforce (sales CRM) | Qualified leads, opportunities and sales |
| RD Station | Marketing automation, with events sent back by Nekt |
| Customer conversations | Service context for every lead |
With the data in place, the agent did not start by proposing changes. It read the accounts and interviewed the team over nine rounds of questions in one week, and every decision became a written rule with its reason. That context stopped false alarms: a conversion that looked broken after a landing page change, and campaigns that looked forgotten but are paused on weekends on purpose, because weekend leads convert much less. Autonomy has levels: the agent reads, analyzes and alerts freely, everything it creates is born paused, and every change to an account still goes through the Growth team's approval.
A promotional campaign with an exclusive offer for one ICP looked great in the ads manager: low cost per conversion and conversions coming in every day. Crossing those conversions with the CRM, the agent saw that only 14% of the contacts were from that ICP. It found the ad that was attracting the wrong audience and proposed pausing it. In the days after the pause, 93% of the contacts were from the right ICP, with the campaign still running.
The agents work because the base was ready before them: ads, site, CRM and marketing automation in one place on Nekt, plus a log of every account change kept since 2024, gave the agent the full funnel and the context of each account from day one. It is the same idea behind context engineering for AI agents and a governed MCP server for data. Learn more about Pixeon.
