CASE STUDYMARROW & CO *AI ASSISTANT

A support assistant that resolves 62% of tickets on its own.

How a growing support backlog turned into an AI assistant that reads the whole knowledge base — with a human always one tap away.

62%

Tickets auto-resolved *

40k

Documents in the RAG index

6 wks

Concept to production

<2s

Median response time

PROBLEM

Support volume was growing faster than the team could hire. Answers lived across 40k scattered documents, and response times were slipping past SLA.

[ image zone — retrieval + handoff flow ]

APPROACH

We built a RAG assistant with tuned retrieval over the full knowledge base, grounded citations, and confident human handoff when the model is unsure.

Every answer ships through an evaluation harness that measures accuracy continuously, with fallbacks and observability baked in from day one.

OUTCOME

The assistant now resolves 62% of tickets before a human sees them, median response time dropped below two seconds, and agents focus on the hard cases.

"Our AI assistant went live in six weeks and now resolves most support tickets before a human sees them."

J. Marrow * — COO, Marrow & Co