Improvado Improvado Improvado home
Get your demo

AI agents

AI agents on your marketing data

An AI agent for marketing data answers questions about your campaigns in plain English instead of making you build a report. Improvado's is trained to understand your business, reads your governed marketing layer, and returns the answer with the sources it came from.

Where does each agent connect?

Through one governed connection, documented per agent below. The platform page for the connection itself is Improvado MCP, which lists every source you can put behind it.

  • Claude

    Ask in plain English, get source-cited answers.

    Claude reads your governed marketing layer directly, so the answer arrives with the sources it came from instead of a number nobody can trace.

    See the Claude setup

  • OpenAI Codex

    Query and schedule reports against clean data.

    Codex works against modelled, deduplicated marketing data, so a scheduled report keeps meaning the same thing next week.

    See the Codex setup

  • Your own agent

    Any framework, one governed connection.

    Bring the agent you already run. It reaches the same governed layer over one connection, with the same access rules your team already has.

    See the custom setup

  • What is an AI agent for marketing data?

    It is an agent you ask a question in plain English, which answers it from your own marketing data instead of sending you to build a report. Improvado's AI Agent is trained to understand your business and deliver tailored insights, answering most day-to-day data questions and ad-hoc stakeholder inquiries on marketing performance. That gives every marketer a way to uncover insights and make data-driven decisions without waiting on an analyst.

  • What does Improvado's AI agent do?

    It takes you from data to insights in a single prompt, across the work a marketing team actually asks for. It compares and analyzes results across every platform, uncovers the revenue impact of every touchpoint once your multi-touch attribution dataset is ready, gives a holistic view of marketing effectiveness from a marketing mix modeling dataset, surfaces winning keywords from combined SEO and paid search data, and identifies your best-performing creative assets, formats, and placements. It also schedules automated reports and insights, so the analytics team can focus on complex, mission-critical work.

  • Can the agent reach live data without building a pipeline first?

    Yes, through Discovery API, the AI Agent's live line to your connected data sources. It talks to a platform's API directly, using the credentials of a connection you already have in Improvado, and returns the answer in chat right now, without loading anything into a warehouse first: it does not require an extraction pipeline, a report type, or any loaded data. The same channel works in both directions, so you can pull live metrics and inspect account structure, and also pause a campaign, adjust a budget or fix a name, with an explicit confirmation before each change. What can be created, changed or only read is a property of the vendor's own API, so the write surface differs per platform.

  • Can the agent build the data work, not just read it?

    Yes. The AI Agent interprets your natural language requests to build, configure, and manage complex data extraction and loading workflows, so it can create an extract using the optimal template, schedule it, and create a load that runs immediately after each extraction completes. It builds recipes the same way, a recipe being a data transformation process that combines aggregated data from multiple digital marketing channels into a single basic performance model: it discovers and analyzes available data sources, proposes a complete data model design, builds the recipe step by step, and validates the final output. Nothing is applied until you are comfortable with the plan, and at every stage you can review, refine, or adjust the work.

  • What is MCP, and what does Improvado expose through it?

    MCP, the Model Context Protocol, is the open standard AI agents use to reach external tools and data. Improvado exposes your entire connected marketing stack through one MCP endpoint: 1,400+ data sources, a unified schema, and knowledge-graph memory. Connect any source to Claude, ChatGPT, Cursor, or any MCP client, and your agent reads, writes, and monitors with enterprise governance.

  • Is my data safe when an agent connects?

    Agents connect through Improvado's MCP layer with scoped, role-based access and a full audit trail. No raw database credentials are ever shared with an agent, and every query is logged: in production, every call and its payload are logged for audit. The platform is SOC 2 Type II audited and GDPR-ready.

  • Can I control exactly what my agent sees?

    Yes. Access is scoped per agent and per data source, least privilege by default, expandable per use case, fully audited. Underneath, access is controlled by two independent layers: role permissions decide which features a user can use, and share settings decide which specific objects a user can see, so giving someone a role that contains every permission in the product does not make existing recipes or dashboards appear for them. You assign roles and permissions, monitor usage, and ensure only the right people can access or modify your data, and the AI Agent has its own permissions inside that model.

  • How long does setup take?

    Minutes. If your sources are already connected to Improvado, you add one MCP config to your agent and authenticate with OAuth: no pipelines to build, no schemas to map.

  • Can Claude, Codex, and a custom agent share the same data?

    Yes, the MCP layer is agent-agnostic. Point Claude, OpenAI Codex, or a custom agent at the same connection and every one of them sees the same schema and shares the same knowledge-graph memory. Both read and write the same Improvado knowledge graph, so a target set in one session is recalled in the other, and agents and humans stay consistent.

  • Which agents and frameworks work with this?

    Anything that speaks MCP: LangChain, custom Python or TypeScript orchestrators, internal platforms. Claude supports MCP servers natively, and Codex works with MCP servers too, so adding the Improvado endpoint to either config makes every connected source queryable. The MCP layer is model-agnostic and works identically across models.

  • Can an agent schedule recurring reports?

    Yes. A script written against the MCP layer can be scheduled to run on a cron, so exports, alerts, and reports keep running without a manual pull. Failures surface back in the same workspace.

Put an agent on your marketing data

Claude, Codex, or your own - connected to your stack.