SurgeRank

Agencies managing local search for dozens of clients were assembling reports by hand from four different tools. The reports took a day a week to produce and told the client what had happened, never what to do next.

We scoped the product around one measurable outcome: an account manager should open the platform on Monday morning and leave with a prioritised list of actions per client, not a chart.

That reframed the AI work. Instead of a chat box bolted to the side, the language model runs a constrained audit — it reads the tracked data, and it may only return findings from a fixed taxonomy with a citation to the metric that triggered them.

Four months. Grid-based rank collection with scheduled jobs, competitor snapshots, an audit engine, and white-label PDF reporting that agencies send under their own brand.

The heaviest engineering was not the AI. It was making the scheduled collection cheap and idempotent enough to run thousands of grid points a day without the bill scaling faster than the customer base.

Next.js and TypeScript, a Node.js job runner, PostgreSQL for time-series rank data, OpenAI for the audit pass, deployed on Vercel with the workers on a separate schedule.