Building a Chatbot for Civic Engagement

How we designed, tested, and built a prototype chatbot to help Cambodians engage with their local government.

TLDR

We wanted to work out what a good platform for citizen engagement with local government services in Cambodia might look like, and how it might work. We ended up creating a pilot of a chatbot that worked in the Khmer language, using Khmer script, to respond naturally to people’s requests for services via Facebook Messenger.

The project was exciting - we believe this is the first time a chatbot for civic engagement has been created in Cambodia that responds naturally in Khmer, and we were able to see a lot of potential for extending this approach in future. Our work was designed and implemented through a partnership between Glean and Pact, with the support of the USAID-funded Development Innovations project.

The Challenge

Tech platforms can help citizens to build a good relationship with local government. In the past though, encouraging citizens to engage their local government through technology platforms has been a challenge in Cambodia. Information dissemination can be unpredictable, and technology has yet to help streamline the process.

That meant that our challenge was to find a more efficient way for Cambodians to engage with their local commune officials, find out processes for services, and register their issues with the right people.

An Assumption

In our eight years working in Cambodia we have seen Facebook take off as an unrivalled platform for digital engagement in Cambodia. Facebook’s reach in Cambodia is so great that a lot of Cambodians don’t know the difference between Facebook and the internet, and we’ve seen time and again that Facebook Messenger is one of the most popular communications tools here nationwide.

The assumption we wanted to test was whether we could build, within Facebook Messenger, a way for citizens to get the information they need, based on a simple chat interaction. Our idea was to use a chatbot that would be able to take citizens through a simple flow, and provide contextual information based on prompts from the user.

Researching

Before we got designing and developing a chatbot prototype, we needed to better understand how Cambodians use their phones, especially Facebook Messenger.

We designed a simple field test with a range of user demographics, with the aim of understanding how social media was being used in the communes we would be testing in. Our testing groups would include commune officials, youth, men, and women, all representing a variety of age groups.

After our first round of research, we identified two groups who would gain the most benefit from a chatbot intervention: high school students and young citizens (age 18-35 – a demographic which in Cambodia contains more than 70% of the population). These two groups had greater access to a smartphone with data, and had the most familiarity with the Messenger platform. If there was any potential in the idea, we’d see it when working with these groups.

Chatbots for Governance: Trust Matters

Our research work with sub-national administration officials – local government people – in Cambodia provided some unique insights into how they use technology, and some of their perceptions around the apps they use on their phones. Based on the sessions we ran and the officials we talked to, 9 out of 10 sub-national government officials had a smartphone with access to data.

The top apps the sub-national authorities were using were: Facebook, Messenger, and WhatsApp.

What was interesting in this group was that they preferred to use WhatsApp to communicate, as it was perceived as safer – due to its encryption – than Messenger.

Designing

After our initial field testing – which was focused on understanding how citizens and stakeholders use their technology – we went to work on designing the chatbot flow. There were a few considerations when we were designing the technology: firstly, how we could build something that was engaging to use; secondly, whether we could provide transparent pricing on government services, and thirdly, whether we could build something that would easily be maintained by our local partner.

We worked with our partner, Pact Cambodia, to map out the relevant government services that we should provide pricing and information for through the chatbot. After researching the latest available data on priority services in communes, the top services we decided to test with our prototype were: National ID Cards, Birth Certificates/Records, Marriage Certificates.

The Right Chatbot Platform for Governance

We also looked at the various chatbot platforms available to best determine which we could use to build our prototype. Our initial research focused on open source bot platforms, which provide a lower initial cost to scale; but ultimately, we landed on Chatfuel, a commercial platform from a startup based in the US. The main reason we decided on Chatfuel as the primary chatbot platform was its easy user design interface. With no coding knowledge a team member at Pact would be able to update the tool as demands or services changed, and sustainability like that is hugely important to ensure tools can be scaled effectively and enable organizations to stay focused on fulfilling their aims and delivering change.

Looking for a way to organically engage users within the target group, we decided to try to bring in some type of gamification to the chatbot. Our initial assumption was that designing a simple story-based game would be a good approach, to take users through an educational or informative journey, earning points along the way. But as we mapped out the decision tree for the game, we realized how complex the game would need to be in order to be compelling and engaging for users, especially given the high level of awareness and discernment amongst Cambodian youth around mobile games. Because of this, and due to the time constraints of our activity, we decided to first focus on transparent service information and local government engagement, and decided that if we could get those working in this prototype, then we could loop back and test the gamification of the program later.

In addition to providing information, we wanted to make sure the chatbot felt friendly, like chatting with a friend or family member. We worked with local artists to help us draft a simple character that would feel like an older brother, to make users feel safe sharing potentially sensitive information.

Testing

To test our prototype, we conducted a series of user experience tests at the local level, with a varying demographic of users. In the tests, we gave users a phone with the chatbot loaded, and asked them to try and start a conversation with the bot. Interactions like this are always fascinating, and we try to ask as few questions as possible to start with – instead, we want to see what users do naturally. Good design should feel natural, and that’s particularly true when creating places and tools for interactions which users haven’t experienced in their own language before.

For youth and young adult demographics, we observed a natural understanding of how to use the tool. Since the tool was based on a communication method they already use, they were able to easily type and request information as if they were messaging a friend or family member.

Conclusions

The chatbot worked well within its constraints. Running the bot in an environment that almost all users already understood (Facebook Messenger) meant that confusion over UI was almost non-existent, and allowed us to see how the design of the service itself worked.

The thing we’re very excited about here is demonstrating that a chatbot can work in the Khmer language, using Khmer text, in a way that users could understand and respond to naturally. That means we have a usable format for delivery of service information, which functions in a way that feels natural, and as far as we know, that’s a first for Cambodia.

There’s still more to do. As a pilot program, the bot functioned with basic content-design and very little iteration, and we were delighted about how well it worked with users; but more user testing will help us refine the content design and work out how to expand it to other services.

The scope of the testing, too, needs to extend. Our approach was very lightweight, and allowed us to get a sense of how relatively tech-savvy users would respond to services delivered this way: we need to go further and research tech access and usage amongst other demographics in Cambodia. It’s also worth investigating other platforms (such as Telegram), which are already embedded in governance in some places and might therefore be more appealing to government stakeholders.

Finally, bringing together services like this isn’t just about citizen users but government users as well; more research, to help us find out how services like this are currently thought about and delivered through government, will help us to make the chatbot more effective in all directions, covering more services, and reaching more people.

Send us an email

Congratulations, you made it to the end! If you're interested in exploring how technology can impact your work, we'd love to chat.