Client name withheld (NDA)

AI knowledge assistant for a SaaS support team

An assistant inside the helpdesk that drafts answers from product docs, past tickets and release notes, with sources, so support agents reply faster and more consistently.

Industry:
SaaS
Region:
UK
Duration:
5 weeks
  • ~25%Faster first response
  • 100%Of drafted answers cite a source
  • 5 weeksFrom discovery call to live pilot

// The challenge

  • Support agents searched five different sources to answer the same questions over and over.
  • New agents took weeks to get up to speed, and answers differed from person to person.
  • Management wanted to use AI but was worried about wrong or made-up answers reaching customers.

// Our approach

  1. 01

    Discovery and use-case selection

    We went through the ticket data together and picked the repeat questions where an assistant would save the most time.

  2. 02

    Prototype on real tickets

    A working prototype answered the previous month's tickets so the team could judge the quality before anything went live.

  3. 03

    RAG assistant in the helpdesk

    It drafts answers that cite the source article or ticket. Agents edit and send them, and the assistant never replies to customers directly.

  4. 04

    Evaluation and tuning

    A test set of real questions, weekly quality reviews and content fixes wherever the assistant got things wrong.

Related Posts