How to run a successful chatbot development project
In this article we’ll share a few things we’ve learned from building chatbots. We’ll cover how to select the right messaging platforms, the importance of conversation design and why we’ve built our own platform for rapid development of bots.
What does success look like?
Before starting any project, it’s important to define goals and ways to measure success.
One of the unique things about bot products is that you can often fake it until you make it. Why not pretend to be your bot for a while and learn about your users’ expectations? Hey, if it’s good enough for Facebook M…
If this is a new startup you might want to check out our thoughts on how to evaluate a new digital product business before spending anything on development.
We use an Agile development process that’s simple and flexible. In order to hit the ground running there are a few things we need to figure out first.
Choosing the right messaging platform
Ultimately the choice should be driven by your audience. One of the key reasons people are building bots is “app fatigue”. The stat everyone throws around is “the average American installs zero new apps per month” — it’s not quite true, but the point remains: We need to meet customers where they already are. What is your audience’s favourite messaging app?
US smartphone users’ number of app downloads per month
We’ve produced an extensive review of messaging app platforms that are suitable for chatbots, which is good background reading. Make this decision based on hard evidence if you can, rather than trying to match up the demographics of your audience and the apps. In the same way that you would review Google Analytics before rebuilding a website in order to know which browsers and devices you need to support, you should aim to collect data from your audience on which messaging apps they use.
Facebook Messenger is by far the most popular option, with over 1 billion monthly active users and strong support from Facebook to grow the platform. We’d recommend looking at Messenger first unless you have particular anonymity requirements or you need to support basic devices that only use SMS.
Interactions can be delivered differently for SMS and Facebook Messenger bots
We like to capture requirements as user stories, since they force us all to think from the end user’s perspective and help us to prioritise the requirement. A user story always has the same format:
As a <user role>, I want <requirement>, so that <reason>
For example, if we were designing the classic pizza-bot we might identify the following requirement:
As a previous customer, I want pizzabot to remember which toppings I dislike, so that I don’t receive inappropriate recommendations.
User Journeys and Conversation Design
The next step is to prioritise our user stories, take the most important ones and figure them out in more detail. This is where we start getting into designing the conversation at a high level.
Atchai works with a conversation designer to provide full-service strategy, design and technology for AI bot products.
Before we consider the bot’s personality, we must design conversation flows that optimise the process for our customers. This could be seen as the equivalent of user journey mapping and wireframes for visual UI projects.
Tone of Voice
Before we start writing final copy we need to develop a shared understanding of the bot’s personality and tone of voice.
There are several techniques to achieving this but the simplest is to develop the character of the bot as if it was a complete person who is just doing their day job by answering the user’s queries. This results in a bot character story that explains who they are, their motivations and quirks.
Developing a fully fledged character is the key to writing a script that doesn’t grate after multiple uses. It also helps you see when humour is an effective tool and when it’s an annoying distraction.
We aim to deliver a working product, deployed and ready to use by real users, right from the very first sprint. The idea is to then iterate on this product, adding more features until we start to meet the goals we set out right at the beginning.
It’s important that we test the product with real users throughout the process. There is no better way to understand whether we’re building the right thing to meet their needs.
Bot scripts are written incrementally, starting with the core functionality and then expanding into personality driven intents and multiple responses as the project develops.
Conversation design also involves training the natural language processor to understand your users, again focussing on the core terms for the key functionality at the start and then expanding as user data becomes available.
The process of matching user intents to existing features or highlighting the need for new responses is a continuous cycle that moves the bot from a decision tree structure to a natural conversation over several sprints.
We’re constantly improving our technology offering, and we’ve invested in developing a core platform that benefits all our customers. Out of the box we can offer:
- Support for major messenger apps (e.g. Facebook Messenger) and SMS.
- A powerful engine that allows for complex conversation workflows.
- Human in the loop — override the bot and talk directly to your users.
- Integration with advanced NLP products like wit.ai.
- RESTful API for quick integration with any web service.
- Fully managed solution that we can host and support for you.
Typically you will find that your users are spread over multiple platforms, so we allow you to handle this transparently, providing you with a single interface to all your users regardless of the messaging app they’re using.
Real-time cross-platform messaging UI, allowing you to interact with your users.
Once your bot is out there in the wild, you will want to keep learning from your users’ behaviour. We will set up analytics tools so that we can all monitor the performance and make informed decisions to continually improve the product.
How much does it cost?
Good question! And we’ll give you a simple answer…
It can cost as little as £15k for a simple MVP — that’s a 1 week discovery phase and a single sprint to get a working product out there. We can deliver this cheaply and reliably because of the technology we’ve already built and the valuable experience we have to share with you.
Of course the full answer is that it depends on how complex the product is that you want to create. If you want to integrate with other technology, or use sophisticated NLP / AI techniques then it can take many sprints to train learning algorithms to be effective and meet product goals.
We’d love to help you realise your vision. If you think we can help you, please get in touch.