All case studies & publications

Retail Customer Service

Email Triage and Agent-Assist Automation for a Car Rental Company

Anonymised engagement
Client
Car rental · Europe · multi-billion-euro rental group, hundreds of thousands of vehicles
Industry
Retail Customer Service
Department
Customer Service & Contact Centre
Region
Europe

Every rental extension, cancellation or vehicle swap meant an agent clicking through screen after screen of Salesforce while the customer waited, and every email meant someone reading it just to work out what the customer wanted. We put UiPath robots behind the agents: emails for five request types sorted and actioned automatically, and the same requests on a call handled with one click.

A working model of the approach on sample data. It plays on its own and pauses when scrolled away; nothing here touches a live system.

The results

5
request types automated
Extend, cancel, roadside assistance, change vehicle, change drop-off. Scope, not a saving.
2
channels on one automation layer
Email unattended, phone attended. Scope, not a saving.
1 click
per agent action on a call
Instead of several Salesforce screens and status changes. No call-time figure was recorded.

The challenge

Our client is one of Europe's largest car rental groups, a multi-billion-euro business running a contact centre across several offices on Salesforce.

Customers phoned and emailed with the same handful of requests: extend a rental, cancel it, send roadside assistance, change the vehicle, change the drop-off location. Each one meant an agent working manually through multiple Salesforce screens and status changes. Every email first had to be read by a person just to work out which of the five requests it was, before any work could start.

What it was costing them

  • Call time. Every extra screen an agent clicks through is added to the call, and to every call after it.
  • Queue length. Agents busy navigating Salesforce are agents not taking the next customer.
  • Email triage by hand. Someone had to read each email just to sort it before anything was done.
  • Half-updated records. Multi-screen status changes, done under call pressure, are where records end up half-updated.
  • Lost transactions. A slow or abandoned interaction is a rental not extended or a booking not changed, and that is revenue.
One rental change, two ways
  • Several screens and status changes by hand
  • One click during the call

The approach

We built UiPath automation integrated with Salesforce through both the API and UI automation, serving two channels: unattended robots that read and action customer emails, and an attended agent-assist tool that turns a multi-screen Salesforce task into one click during a call.

Email channel: unattended

  1. Email understanding. Communications mining reads each incoming customer email and labels its intent across the five request types, with a confidence score. Nobody reads emails just to sort them.
  2. Confidence gate. High-confidence emails become transaction items in a queue for an unattended robot. Lower-confidence emails are not auto-actioned, so the robot never guesses.
  3. Processing in Salesforce. The robot carries out the request in Salesforce: the extension, cancellation, roadside request, vehicle change or drop-off change.

Phone channel: one-click agent assist

  1. Pop-up menu. When a call comes in, the agent sees a pop-up of the relevant actions.
  2. One click. Choosing an action triggers the automation behind the scenes, instead of the agent working through multiple screens and status updates.
  3. The agent stays on the conversation while the robot updates Salesforce.
Two channels, one Salesforce automation layer
  • Email: unattended robot → Salesforce automation layer
  • Phone: one-click assist → Salesforce automation layer
On the email path, a confidence gate keeps uncertain emails away from the robot.

API where possible, UI where needed

Both channels write into Salesforce through the API and through UI automation, so every action could be automated, not only the ones Salesforce exposes through an API. One integration layer serves the email robots and the phone agents, so a request is handled the same way whichever channel it arrives on.

Technology

LayerWhat we used
AutomationUiPath Studio, unattended and attended robots
Email understandingCommunications mining: intent classification with confidence scores
CRMSalesforce, through the API and UI automation
Agent toolingAgent-assist pop-up for contact-centre agents
Work distributionQueue-based transaction processing

The outcome

Agents click once instead of working through screen after screen, and the most common email requests are sorted and actioned without anyone reading them first.

BeforeAfter
Sorting customer emailsRead by a personLabelled automatically, with a confidence score
Actioning a high-confidence emailBy hand in SalesforceUnattended robot
A rental change during a callMultiple screens and status changesOne click
Low-confidence emailsRead by a personNot auto-actioned, never guessed
The same request by email or phoneTwo manual routesOne Salesforce automation layer
  • Shorter calls. Agents complete in one click what used to take several screens, and call durations came down.
  • More transactions per agent, and higher productivity across all of the client's offices.
  • Emails for the five request types sorted and actioned without manual reading whenever confidence was high.
  • A contribution to revenue. The client saw revenue rise; this is recalled by the delivery team, not a documented figure.
  • Directional, not measured. No volumes or timings were recorded for this engagement, so the figures above describe scope, not savings.

The risk most teams miss

Salesforce ships three major releases a year, and each one can move the screens a UI robot depends on. An agent-assist button that fails mid-call is worse than no button. Using the API wherever Salesforce exposes the action keeps most of the work independent of screen changes; the UI-automated steps are the ones to retest around each release.

The simpler sibling of this build: in-flight wifi refunds for a UK airline group, where a single-purpose mailbox made intent classification unnecessary.

Built with

The tools this ran on, by the layer each one carried.

  1. Apps & integration
  2. Automation

Next step

Start the same conversation

Tell us the process that keeps breaking. We will come back with what it would take to fix it, what it would cost, and what we would measure.

Or write to us directly: •••••@•••••••.•••

  • A reply from a partner within one business day
  • Fixed scope and fixed fee, agreed before we start
  • The maps and the backlog are yours either way