Skip to content
infoloop
Work

Case study 2 min read

A fintech scale-up

An AI support assistant that cut manual ticket work by 72%

A fintech scale-up was buried under customer messages. Infoloop built an AI support assistant that does the routine part of the job while a person decides, and we still run it today.

Industry
Financial services
Services
AI support assistant, Ongoing run and monthly tuning
Timeline
5 weeks to first version live

Ask how this would work for you

Introduction

The vision: routine questions answered in minutes, with a person still deciding

A fintech scale-up was signing up customers faster than it could hire people to look after them, and the support inbox was buried.

The brief was an AI assistant that does the routine part of the job while a person decides, live quickly, and run by Infoloop after launch.

The challenge

Most of the messages were the same handful of questions asked in different words.

  • Where has my refund got to, why is my account on hold, what was this payment for: the same three questions, all day
  • Answering any one of them meant opening three different screens and copying numbers between them
  • Agents were writing sentences they had already written many times that week
  • The wait for a first reply had crept from minutes into hours, and customers had started to say so out loud
  • Hiring another row of desks would only buy a few months

The business needed the routine questions answered quickly without taking judgement away from the people on the desk.

Our approach

We read a week of real messages instead of asking managers. Three kinds of request followed rules the team knew by heart, using information the company already held: exactly the work software can take over. Everything else stayed with a person, agreed on day one.

  1. An assistant that sees what the agent sees

    Joined up to the support inbox, the payments platform and the customer records, so it can see what an agent would otherwise have to go and look up.

  2. Ready-written replies

    Ready-written replies and one-click actions for the three commonest kinds of message.

  3. A person approves anything involving money

    A person has to press approve before anything involving money actually happens.

  4. One screen for the queue

    How much is coming in, how quickly it is answered, and how much the assistant handled on its own.

The results

A working version shipped in five weeks. Routine questions are now answered in minutes, agents work the cases that need a person, and we still run and tune the assistant every month.

Technology used

A quick look at what runs behind this build.

Technology
  • AI support assistant
  • Support inbox connection
  • Payments platform connection
  • Customer records connection
  • Approval step for money actions
  • Live reporting screen
What we did
  • AI support assistant
  • Ongoing run and monthly tuning
Runs today
  • Live, run by Infoloop

Keep reading

All case studies

Take the routine out of your support desk

Infoloop builds AI assistants that draft the routine replies while your people decide, and keeps them tuned every month.