Industry: Legal advertising · Engagement: ~2 months

The problem

Pareto PPC runs ads for lawyers. Twenty active clients, significant spend, real results for the firms it works with. The data infrastructure told a different story. Their internal data engineer built a pipeline for each client, slightly differently every time: one client had client_name, another had client-name, a third had something else entirely.

That inconsistency cuts two ways. When something breaks, there's no pattern to trace, so debugging means twenty different systems instead of one. And every new client makes it worse: onboarding meant repeating the same process with slight variations, adding more fragile code and more naming inconsistencies each time. Every client they signed became a liability to the infrastructure holding the business together.

Pareto had a standard report template, in theory. In practice, every new client meant copying it and hand-editing the result to match that client's data structure. They tried pointing Claude at the data to speed up analysis; it got confused, hallucinated numbers, and gave answers nobody could trust.

The rebuild

Orbit started with an audit: full access to the existing data lake, client samples, and reports, looking for every gap and structural problem they didn't know they had. There were plenty. The fix wasn't patching the old system. It was replacing it, following the same patterns proven at Glamorise and other long-term clients.

  • One unified pipeline. Every client flows through the same system, same column names, same structure, same transformations.
  • Onboarding in minutes. Connect the Google account, connect the Facebook account, name the client. Everything downstream handles itself.
  • One filter, not one rebuild. Clone the reporting template, add a single filter for the new client, done.
  • Time tracking, unified. ClickUp data now flows into the same warehouse, so leadership can see time spent per client without anyone filling out a spreadsheet.
  • AI-ready. A single warehouse with clean documentation Claude can actually navigate, instead of guessing.

Data quality assertions now watch the pipeline. If a table stops syncing, an alert fires to Slack and email immediately, instead of a client noticing something looks off days later and asking why.

The results

  • 20 clients, all normalized to one structure
  • 1 pipeline replacing twenty different implementations
  • Minutes, not days, to onboard a new client

Before: every client wired differently, onboarding creating tech debt, reports rebuilt by hand, and AI tools unable to make sense of the mess. After: one pipeline, one structure, one template, with improvements that compound across the entire client base.

We basically modernized their stack and made it ready to be used with AI analysts and AI tools. Before, they'd point Claude at it and it would just get confused and hallucinate numbers.

This is where the engagement is heading next: using the cleaned-up infrastructure as the foundation for AI-assisted analysis the data can finally support.