VIETTEL MONEY
[PROJECT DETAILS] :
I. Background
Viettel Group asked VDS to move customer care to a Digital First, self-service model, with two aims: raise service quality and cut the infrastructure cost of running support.
Analyzing the situation
What the call center cost
Staffing 11,497,000 VND per seat each month
Hotline infrastructure (call charges) 555,337,652 VND a month
Total running cost roughly 1.07 billion VND a month
How it was performing (Q1 2022)
Inbound calls ~143,317 a month
Answered by an agent ~117,764 a month
Never connected ~13,785 a month
Load per agent 94 calls a month on average
Average wait time ~10.3s per call
Average handling time ~260s per request
→ Demand was high and each request took a long time to resolve. The hotline was regularly overloaded, and customers felt it.
Key Insight
The call center was expensive to run and still couldn't keep up. Every unanswered call cost money and chipped away at trust in the Viettel Money brand.
II. Design Challenge
Design an in-app Help Center, backed by a chatbot, where customers can find answers, raise requests and track them without ever picking up the phone.
What success looked like
Business Goals:
Retire the traditional call center and the staffing and infrastructure costs that come with it.
Bring customer care in line with the company's wider digital transformation.
User Needs:
Solve problems on their own, without leaving the app.
Get help faster when they do need a person.
III. Research
Research Questions
Internal documents gave me the business side. To fill in the rest, I framed my research around three questions:
1. How do other apps approach in-app support today?
2. What do people expect when something goes wrong mid-task?
3. How should we measure whether a help center is working?
General Findings
A literature review and a round of user interviews surfaced the following:
How other apps do it
Most treat the help center as a knowledge base with a route to a human agent. The better ones go further:
• They surface help at the moments people are most likely to need it, not just in a settings menu.
• They organise content by topic, so browsing, searching and raising a request all feel like one flow.
What users expect
• A choice of channels, depending on how urgent or complex the problem is.
• Clear answers to common problems so they can fix things themselves.
How to measure it
• Request volume per day and month, and requests or complaints per customer.
• Average handle time how long it takes to resolve a request.
• First contact resolution the share of issues solved in one go.
• CSAT how satisfied customers are with the help they got.
IV. Ideate & Design
The research pointed to three design decisions:
1. Put entry points where problems actually happen
2. Organise support topics so customers can find an answer or raise the right request quickly
3. Keep support channels and FAQs within easy reach
V. Validation
I turned the designs into a high-fidelity prototype and ran usability tests with four participants. Each task was scored on completion and first-attempt success, and the feedback shaped the recommendations for launch.
Analysis
Task
Success Criteria
Success
First attempt
Find service info (answer is in the FAQ)
Home → FAQ answer
2/4
0/4
Find service info (answer is not in the FAQ)
Home → Chatbot, directly or via the FAQ
4/4
0/4
Request help from a failed transaction
Transaction result → Chatbot
4/4
4/4
Request help from Home
Transaction history or Help Center → Chatbot
4/4
2/4
Finding
The core goal held up: not one participant thought about calling the hotline at any point during testing.
Participants responded well to getting help in the moment, right where the problem appeared.
Most found the Help Center and chatbot easy to understand on first use.
Reading the results
Finding information
When the answer was in the FAQ only 2 of 4 got there, and nobody on the first try.
When it wasn't all 4 succeeded, but it took them a while.
Requesting help
From a failed transaction all 4 got there first time.
From Home everyone finished, but only 2 of 4 on the first try.
What this told us
Information lives where the feature lives. People didn't think to look for a central knowledge base under Profile. They either went into the feature they had a question about, or left the app to Google it or ask someone.
Requesting help was easy. There was some hesitation on the first try, but everyone completed it. The standout moment was being able to ask for help straight from a failed transaction.
In their words
“It took me straight to an explanation, so I was a lot less anxious when the payment failed.”
“Seeing information right away made me feel safe enough to just wait it out.”
“Quick and sensible. It reassures me first, and if that's not enough I can send a request straight away.”
“I like that I find out why a transaction failed the moment it happens. I wouldn't know what to do on my own, but the app gave me a fix right there.”
Recommendations
Next steps for the Customer Care team (phase 2)
• Give the Paydi chatbot a warmer, more human voice so it feels reassuring rather than robotic.
• Tell people when to expect a reply, so waiting feels predictable.
VI. Achievement
15k
sessions a day
0.3%
bounce rate (users leaving within 3 seconds)
20s
average session length
89%
of surveyed customers rated the Help Center helpful
Only 0.3% of sessions ended in a support request
That low number is the point: most customers found what they needed without having to contact anyone.
Calculated as support requests ÷ Help Center sessions
