All work
LLM Development Analytics SaaS 2025 14 weeks

Grapho AI

47% increase in new customers

A natural-language analytics layer that lets non-technical teams query their warehouse in plain English and get charts back in seconds.

More new customers
Query accuracy on eval set
Faster time to insight
01

The challenge

Grapho's customers were sitting on rich product data but only three people in the company could write the SQL to reach it. Every question routed through a backlog that ran two weeks deep, so most questions simply never got asked.

02

What we built

We built a retrieval layer over their schema and metric definitions, then fine-tuned a query planner that turns a plain-English question into validated SQL. A verification pass runs every generated query against a dry-run endpoint before results ever reach the user, so a malformed query fails silently instead of returning a confident wrong number.

03

The outcome

Self-serve analytics went from a three-person bottleneck to something the whole company uses daily. The data team now spends its time on modelling instead of answering ad-hoc requests.

“We stopped being the bottleneck. That alone changed how the company makes decisions.”

Priya Raghavan Head of Data, Grapho