Unblocking every team to query and build with warehouse data, with Rig's context layer and governed Rig MCP
A Fintech is building an AI-powered finance tool for consumers. As they scaled, the warehouse outgrew the people who could query it, and fraud, marketing, engineering, and analytics were all bottlenecked by tribal knowledge and manual SQL.
A 4,000-table warehouse, four blocked teams.
The client's business outgrew its data access. The warehouse was big and sophisticated, but using it required tribal knowledge, SQL fluency, and a queue.
Stitching transaction data by hand, every case.
30–60 minutes assembling context before the analyst could even make a decision.
Every data question had to wait in a queue.
Answers queued behind engineering. Decisions bottlenecked on a manual pull.
Context-switching between IDE, BI tool, console.
Wanted to query the warehouse without leaving Cursor or Claude Code.
The whole org routed through one queue.
A bottleneck on every decision, not a force-multiplier.
What Rig built
Rig deployed four things:
- A self-healing data context layer over the organisation's entire warehouse
- A hosted data chat
- A Rig MCP to talk to data and build with data in Claude / Cursor
- Specialised automations and data apps for fraud/disputes, marketing, and CS
Usage exploded in non-tech teams — Marketing, CS, Executive team (chatting to data in Claude), and Fraud/Disputes (pulling the custom data they need).
Source: client ROI dashboard. Monthly Rig MCP calls, Data App runs, and chat messages, from rollout onward. Rig MCP calls overtake every other surface by Month 2 and dominate by Month 3; the current month is month-to-date.
Usage over the first months crystallised something important about what Rig is actually for. "Rig MCP only" users grew beyond engineers active in Cursor or Claude Code, with business teams using the Rig MCP in their tools to work with internal data. It became the active feature of user choice within the first few months and has stayed dominant since.
The conclusion: the load-bearing piece is safe, accurate, governed data context delivered wherever the user already works, not natural-language-to-SQL. Every LLM can write SQL now. The problem worth solving is giving those LLMs the right context, with the right permissions, against the right tables. The client's adoption curve is the proof.
Five functions, one data layer
Sixteen weeks after first deployment, Rig had moved from a fraud-team pilot to a data access layer across different teams in the company, used by customer-facing functions and internal ops alike.
- Finance & DisputesLoan limit alerts · Repayment risk warnings · Credit dispute reviews · Dispute backlog triage
- MarketingTop-performing ad detection · Channel performance reviews · Card-ads campaign tracking
- OperationsData pipeline health checks · OKR progress reports · AI cost monitoring · Weekly active-user digests
- GTMConversion-rate analysis · Seasonal demand tracking · Pipeline performance pulse
- Customer SupportWeekly support summaries · Conversation-quality reviews · Escalation pattern tracking
Different functions shipping their own automations against the same governed context layer.