Kevin Burke, Associate Director of Analytics at Noble People, built an in-house AI agent he calls Rocky. Improvado is the backbone underneath, pulling every paid channel the agency runs into one clean BigQuery warehouse. Through the Improvado MCP (beta), the agent queries that governed, reconciled spine directly, never raw client data, and a person reviews every output before anything reaches a client. Reporting that took days now takes hours, and a new-client dashboard that was a multi-week build lands in about 48 hours.
The setup
Noble People is an independent media agency in New York that pairs inventive media thinking with real analytical rigor. Kevin Burke joined to keep the agency's analytics pipeline running through a team transition and ended up rebuilding both the ETL pipeline and the client reporting from scratch. Today the analytics team covers 20+ clients and a dozen platforms, and along the way Kevin made two decisions that define how the agency works now. He standardized the data layer on Improvado, and he built an AI agent on top of it. Ask Kevin where everything begins, and the answer is immediate.
"It's the base of everything. Everything starts with Improvado in our ETL pipeline. That's the starting point for everything that we do, whether it's a prospect, whether it's a client, whether we have some one-off analysis that we need to do. It all starts with our own data that we own for the client work that we do."Kevin Burke, Associate Director of Analytics, Noble People
The bottleneck was the handoff
The team saw the promise of AI early, but the technology had a clear gap. Claude could reason brilliantly about the data, yet it had no way to go and fetch it, so the connecting happened by hand. Export from the warehouse, drop a CSV into Claude, or hand-run a query and paste the results back. Every analysis started with a data pull, and iterating meant re-exporting. The day-to-day turned reactive: A steady stream of messages about a broken report or a client who could not see their numbers. Each one was troubleshot by hand, and each one ate the day. The reasoning was strong and the data was rich, but the manual handoff in the middle capped what either could do.
"Claude could reason about the data, but it couldn't go get it, so we were the bottleneck."Kevin Burke, Associate Director of Analytics, Noble People
Clean spine first, then the agent
The build happened in the right order. First the spine: Kevin made Improvado the ETL layer, pulling every paid channel the agency runs (Google, Meta, LinkedIn, TikTok, Reddit, X, Bing, YouTube and more) into BigQuery. The result is one unified, consistently named warehouse feeding reconciled views and client dashboards. The merged social and display connectors now "all live happy ever after in one table," as Kevin puts it. On that foundation sits one normalized view across the whole client book, 20+ clients and a dozen platforms, all spend-mapped. Then the agent: Kevin built Rocky, named after the alien in Project Hail Mary. Kevin now directs the agent's work. Connecting Rocky through the Improvado MCP, still in beta, closed the gap that had made the team the middleman. Rocky works through the Improvado MCP and the BigQuery MCP together, writing the SQL, building the views, and drawing on every dashboard the team has already shipped. Repeatable cross-client work turned into playbooks, durable agency process rather than one-off scripts, and Improvado plays a part in every single one because it generates the data everything else depends on.
"One of the biggest pains of all time is joining data that doesn't match. Improvado solves that immediately."Kevin Burke, Associate Director of Analytics, Noble People
Governed by design
For an agency handling client media data, the guardrails are not an afterthought; they are the reason the whole thing works. Rocky queries Noble People's own governed, reconciled warehouse, never raw client data. That warehouse is the set of BigQuery views Improvado feeds, so everything the agent touches is already normalized and reconciled. A person reviews every output before anything reaches a client: analysts walk through each dashboard ahead of weekly client calls, so the numbers arrive with human judgment attached. The pipeline watches itself too. Every night an automated audit checks each view the team relies on, and every morning Kevin gets a data-quality email flagging anything that lagged or broke, so issues get fixed before a client ever sees them. That discipline is what turns speed into trust.
"Our data is a lot more trustworthy. It's clean, we know how to use it, and it really helps me sleep at night, because I trust the data that we have now."Kevin Burke, Associate Director of Analytics, Noble People
What changed
With the clean spine and the agent working together, the shape of the work changed. A data discrepancy that used to eat days of back-and-forth now resolves in minutes. Cross-platform reporting that took days lands in hours. A new-client dashboard that was a multi-week build goes from kickoff to live in about 48 hours, and research that ran weeks, sometimes months, comes back in days. Knowledge transfer changed too: because the whole pipeline is documented and the agent knows it, onboarding into the analytics function takes days rather than months. The payoff is reclaimed time that goes into better analysis and more coverage, not fewer people. Analysts spend that time reviewing dashboards before client calls and turning numbers into recommendations, the strategic work clients actually pay for.
Ask Kevin what the reclaimed time is actually for, and he doesn't talk about efficiency. The point, in his words, is to put "the human touch back into analytics." The dashboards were never the finish line; the recommendations built on top of them are.
Directional by design. The pattern is the story: The manual data-engineering work moved to the data layer, and the analysis moved back to people.
Rolling out AI without the fear
Kevin's approach to bringing the team along reframed the fear most people have about AI. Instead of a tool that threatens jobs, he ran sessions with a simple premise: "Let's automate what you hate." Pick the part of the work nobody enjoys, and hand that to the agent. The team came along quickly, and the effect spread. Colleagues with completely different specialties now pick up Claude and the Improvado MCP and find their own use cases, each one different from Kevin's, all of them reading from the same trusted spine.
What Kevin tells other agencies
His advice to a peer thinking about connecting marketing data to an AI agent is the distilled version of everything above. "Get your data foundation right first, because the AI is only as good as the spine it is querying. Once your data is unified and consistently named in one warehouse, connecting an AI agent through something like the MCP is the step that turns it from a reporting tool into a teammate that can actually fetch, build, and automate." The spine does the data engineering, the agent does the legwork, and the people do the analysis. Asked to sum up what the combination means for Noble People, Kevin lands on the part that matters most.
"The main benefit of the Improvado MCP is knowing that our data is clean, accessible, all in one place. Turning client data into actionable insights went from weeks to days, leading to better recommendations for our clients and better wins for our internal team. It taught us that analytics doesn't stop at the dashboard. We create dashboards so quickly now that we have more time to spend on the human side of things, finding the insights that actually matter to our clients' campaigns."Kevin Burke, Associate Director of Analytics, Noble People
.png){kind=logo}
