claudia · makes ➧ TOBi Chat case study

My role

Designing for TOBi meant working at scale, with millions of customer interactions and measurable business impact.

Led a team of 2 designers, 1 QA analyst, and 1 UX writer
Owned the process: research, analysis, and final design decisions
Designed 20+ journeys for key drivers, like billing, sales, and troubleshooting
Defined reusable design patterns for faster, more consistent journey design
Improved target metrics: Containment > 80% + NPS > 60

Key insights

Position chatbot as a self‑service tool

Not as a gatekeeper to speaking with a human agent.
Generic responses feel unintelligent

Leverage account data to give specific, relevant answers.
Edge cases are inevitable

Keep journeys on track with context‑aware guidance and fallbacks.
Fluency and technical know‑how varies

Use plain language and illustrate concepts with examples.

Design process

The goal was improving efficiency and experience metrics. Journeys were monitored after launch, and measured against containment and NPS targets.

- Step 1: Discover and Define
- New journey
  - Stakeholder role:
    Business requirements
  - UX team role:
    Define use cases
- Journey redesign
  - UX team role:
    Audit current flow
  - Analyze customer logs
  - Map issues and opportunities
➭
- Step 2: Design and Test
- - Stakeholder role:
    External input
  - UX team role:
    Design and prototype
  - Test usability
  - Iterate
➭
- Step 3: Implement and Validate
- - UX team role:
  - Document
  - Hand-off
  - Stakeholder role:
  - Implementation QA
  - UX team role:
  - AB test
    For journey redesigns
  - Measure KPIs
UX team

Stakeholders

Working with the QA analyst, I analyzed anonymous customer logs to find where journeys broke down and why (surfacing gaps, edge cases, misalignment between the chatbot and human agents), often pointing to new feature opportunities.

Stakeholders were involved throughout: Customer Service QA checked compliance, Brand and Marketing covered sales and promotional journeys, and Development flagged technical constraints.

Featured journey

Troubleshooting

Many customers reached this journey via channel switch from phone support.

Customers should always stay on their preferred channel, but some tasks, like technical troubleshooting, benefit from visual support. Channel switching makes sense in those cases, however, it’s important to work towards channel parity, so customers can get the help they need on any given channel.

This journey authenticates the customer, surfaces known service issues, runs automated connection and equipment checks, and offers further troubleshooting or agent hand-off. A short NPS survey is shown at the end of the journey.

What wasn’t working:

Service selection wasn’t obvious to all customers.
False containment created by customers marking the issue as solved to end the troubleshooting.
High abandonment, particularly at the advanced troubleshooting step.
When automated tests fixed the issue, NPS was very high. When customers gave up, it was very low.

A customer tells TOBi their home internet isn't working. TOBi acknowledges the issue, explains that he needs to check their service status, and asks for the customer's account number to do so. Near the input field, a suggested reply reads "I don't know my account number", offering a fallback for customers who don't have it on hand.

What changed

Service selection

Identifying the affected service wasn’t always obvious to customers. The redesign made service selection clearer and easier, with a simple way to switch if they'd chosen the wrong one.

Automated tests

Automated checks ran in the background with little feedback. The redesign made each test stage clearer, so customers could follow the progress and understand what was happening.

Resume journey link

Customers who couldn't troubleshoot at the moment were told to return to the chat on their own. The redesign introduced a link, sent via SMS and valid for 48 hours, that picks up exactly where they left off.

If the customer does return on their own, TOBi will ask if they want to continue the troubleshooting either way. But the link reinforces the idea that they won’t have to repeat information.

Optional troubleshooting

Complex steps were mandatory before agent hand-off was offered. The redesign makes them optional, with hand-off available immediately after the automated tests.

If the customer calls after completing the basic troubleshooting, they're connected directly to a human agent, avoiding the frustration of having to explain the issue again.

NPS survey

Typed input led to frequent errors, corrupting the data used to measure journey success. Buttons with a deliberate submit step make the choice more intentional and the data more reliable.

We give dissatisfied customers the option to speak with a human. It’s a good opportunity to solve pending issues, and turn detractors into promoters.

User feedback

User feedback confirmed the changes were positive, but also pointed to possible improvements. This feedback informed future design iterations. It’s tempting to implement all feedback, but we were careful to always balance the information and features offered, so choices didn’t become overwhelming.