MTN Mega Khava MTN Mega Khava MTN Mega Khava MTN Mega Khava MTN Mega Khava MTN Mega Khava MTN Mega Khava MTN Mega Khava

Self-service WhatsApp bot - 2023

Key skills: Information design and copywriting

5 - 10 minute read

Key Players

Client:

MTN, the African telecom giant; one of the most recognisable and trustworthy brands on the continent.

Service provider:

aYo Holdings, a microinsurance provider serving seven African countries.

Elevator pitch

Collaborate with the team at aYo to design a self-service WhatsApp chatbot to be the automated interface for their family funeral cover in South Africa.

The management team called the UIX team in to help with copywriting and visuals, however we soon uncovered flaws in the user journeys – leading to a more comprehensive evaluation.

The cast

–

Summary of my role

My role was to coordinate the UX effort, take responsibility for copywriting, and organise the user journey. That said, teamwork and collaborative feedback played a key role.

The timeline

Research

Research tells us…

MTN’s competitive prices and value-adds has them in the running – but the truth is, good customer service would be a deal breaker.

Connect this with WhatsApp, the most popular messaging app in South Africa and home to numerous successful chatbots. Suddenly, the future of MTN Khava looks promising.

User research:

With no budget for new user research, I did the following…

Leveraged aYo’s 2022 research in South Africa (conducted by TRi Facts).
Learned from existing usability studies on chatbots (Nielsen Norman Group, IBM).
Studied aYo’s user research on the South African lower to middle class microinsurance target group (2019).

Other research:

Funeral cover research: Finmark Trust 2020.
Tech research: workshops with Clickatell, our development agency.
Competitor research on microinsurance (Vodacom, Cell C, Telkom) and funeral cover providers (Sanlam, Old Mutual, Avbob).

Average cost of funeral: R15,000 – R100,000
Average monthly salary of target audience: R5,500
Average monthly contribution to funeral cover: R300

Empathise

Empathy map: The research indicates that the user base has limited financial resources yet highly values funeral insurance.

Aggregated user persona: Meet Innocent – a family man shouldering many debts.

The User Experience of Chatbots (2018). Nielsen Norman Group.

Journey Mapping Insights

Main Problems

WhatsApp is a platform for conversation, yet our bot lacks language processing. This leads to issues.
Decision trees lack flexibility. This is a pain point for task-oriented users.
Messaging is expected to be conversational and easy-to-read. Understanding insurance requires more complexity, and more reading.

Main Opportunities

Communicating the limitations of the bot (being a tool), while offering access to outside help (from a human) prevents drop-off.
In the absence of search functionality, a hierarchical and simple decision tree is the best aid for task-oriented users.
Utilising media uploads allows messages to be brief (as expected) and documents to be expansive (also as expected).

Decision tree mapping

Decision tree: broad overview

My identified IA guidelines

Simple tree, simple flows, simple tasks.
Use logic and hierarchy.
Limit choices.
Quick access to key functions.
Adhere to WhatsApp limitations.
Limit user error by replacing typing with buttons.

Major limitation

Navigation allowances on WhatsApp are limited: there is no intuitive way to provide flexible navigation.

To solve for this, I conducted competitive studies of WhatsApp bots (DisChem, Discovery, SpecSavers, etc.) and team brainstorming. The outcome? A not-so-perfect escape hatch for users who must type ‘help’ when stuck.

An intuitive solution was not possible, so ease of interaction (i.e. moving forwards) was prioritised over ease of navigation.

Copywriting

Bible #1: Terminology

South African funeral cover goes by many names, with no consistent rhetoric for benefits and value adds. MTN had the same problem.

The UX team made a logical choice and enforced it. The product is named “Khava” which translates to “cover” in Xhosa. I simply decided to use this name consistently.

Here is how I mapped out naming conventions for all other benefits.

Terminology map

Bible #2: “That feeling”

Creating a VOICE for our bot meant honing in on the “feeling” I wanted users to experience. Through a team brainstorm I identified “sunny and informative”, “reassuring”, and “compassionate” as the most important characteristics for our bot.

My job was to design one voice to carry each moment, without invoking the knee-jerk reaction to a complete tonal shift. I used Torrey Podmajersky’s voice chart technique as a blueprint.

How to have a nice chat

Just like human conversation, which unfolds in small bursts of back-and-forth, I wrote messages to convey one concept at a time.

Before users entered a long process, I provided a roadmap to set expectations.

Instructions were clear, emojis and fluff kept only for conveying sentiment.

Limitations

I limited buttons to 18 characters, menus to 10 items, and messages to 160 characters for at-a-glance readability.

The chat only serves as a rough outline, other media must convey detail. Remembering users aren’t here to read was critical.

Prototype + test + iterate

* A note on testing

All testing was conducted internally with aYo staff standing in for actual users. Frequently, these tests resulted in a propensity for lengthy, information-dense messages, which we know to be less effective on WhatsApp. Instead, I focused on interpreting feedback to enhance usability.

16 journeys later

Outcome pending…

After several iterations and continuous collaboration with design, development, and client, we finally landed on a chatbot with a little style and a lot of usability.

The product is due for launch early 2024.

Key Performance Indicators

The following KPIs will be evaluated once the product reaches real-world users.

User uptake – are we meeting expected market standards?
Conversion rates – are new users converting into paying customers?
Error rates – how many errors are users experiencing?
Drop off rates – where and why are users dropping off?
The claims process – how fast is a claim on average and would it be improved with upload functionality?