How we use AI to help users get answers instantly and increase customer satisfaction.
Case Study Of e-Residency, An Estonian Governmental “Startup”
A couple of months ago, we launched an Artificial Intelligence (AI) customer service assistant for the Republic of Estonia’s e-Residency organisation. Together, we decided to tackle the challenge of increasing customer support response time without hiring additional people. The solution was simple — an artificial intelligence chatbot with a live-chat interface. We are really hyped that the AI has been able to handle 45% of customers’ questions from live chat on it’s own. This is a case study on how we did it. Hope you enjoy it 😉
About The Project
The Republic of Estonia’s e-Residency organisation is running a tight ship. With a small team they’re connecting digital citizens all across the world with Estonian state services. It’s an extremely lean and efficient organisation. One could say it’s almost like a governmental start-up. An experiment, that has turned into a unique phenomenon in the whole world.
Naturally, they’re getting a lot of questions from people who are interested in becoming an e-resident or have already submitted their application and are now wondering how they can receive their digital identity or what they can do with it.
For each individual, these questions are unique, however, from the organisation’s standpoint, they often repeat. This poses a challenge since team members have to spend a lot of time re-writing the same answers into different conversation fields. As humans, there are better ways we can use our time.
Estonia is one of the most innovative countries in the world, and so are its organisations. To tackle the issue, e-Residency turned to us and artificial intelligence that is already smart enough that it can help handle a lot of the customer service issues without the need for human intervention.
Software can help solve people’s problems faster, as well as save human resources for the organisation and, in this case, save money from the taxpayer pockets.
This is a real-life case study of how we worked together with the e-Residency team of Estonia to help bring business automation and conversational AI to a public sector company through implementing live chat with a chatbot assistant.
The Results
Right now the E-Residency AI has a 45% success rate in solving customers’ problems. This means that the AI was able to solve customers’ issues/questions completely on its own for almost half the questions asked.
On average, one could expect that an AI will handle 10–20% of the customers’ problems (it also varies case-by-case). Suffice to say, we were very positively surprised by the results delivered with this AI assistant.
These are preliminary results after only a couple of months of operation. We expect to improve the success rate to at least 50–60% with further training of the AI.
What does it mean for your business?
- You will need twice as less customer support team members to handle the same amount of inquiries
- You will get an automated “customer support service” that is online 24/7
Is It Worth The Time, Money & Effort?
In general, a customer service AI would make sense if you have either no customer service team at all or you have a large and overloaded one and you’d like to have the first level of support be instantaneous for common issues.
In e-Residency’s case, it made perfect sense since they are not offering live support. This way users can have answers to pressing issues instantly, leaving more time for the team to deal with complex issues via e-mail.
1. If done right, the level of customer service goes up
If you don’t use AI to distance your company from your customers but rather use it to help people with pressing issues faster, then AI can be a great tool at your disposal. It means that people with simple questions get an instant answer, rather then having to wait.
More complex issues get handed off to human agents automatically. This means that humans can spend more time and focus on solving problems that the AI cannot handle. It’s a win-win.
You can also get closer to your customers, being able to deploy an AI assistant on various platforms — on your website, Facebook Messenger, Google Assistant (voice), Telegram, etc.
2. AI gets smarter over time and doesn’t need vacation days
An AI assistant is not a one-off solution. It’s performance should be continuously monitored to make it perform better and become smarter over time. Once you’ve taught something to the AI, it doesn’t forget how to handle that problem (or need to take a coffee break, or have vacation days). It’s available to your customers 24/7, 365 days a year and it only becomes better over time.
3. Cost-saving
Being able to handle 20–60% of incoming tickets with an AI assistant can mean great cost-savings for your company. It doesn’t matter how many people are talking with the AI, it can handle them all at the same time.
In fact, we’ve created a simple tool that can help you estimate how much you could save if you used an AI assistant. Check it out here:https://chatcreate.com/
Common Misconceptions Around Customer Service Automation With AI
1. Isn’t it just another answering machine?
No. A true AI bot is very different from “press 3 to talk about your delivery issues”. Provided that it has been trained well enough, it can take questions from users as they come and respond to them almost like another human being would. If it can’t, then it hands over the conversation to a real human.
2. AI will replace all customer service agents.
Probably not. There’s still a long way to go until AI will be smart enough to fully replace human jobs. Right now it’s more like a helpful assistant that can help handle mundane tasks while letting humans focus on more important things.
3. It takes a lot of time and money to set it up
Usually it takes approximately 3–4 weeks to launch an intelligent assistant. However it doesn’t stop there. After the first launch, the bot needs to be monitored and continuously made smarter. This is something that we can do, it doesn’t require much time from the client.
We map the most common questions together with the client and then take it from there. To see how much it would cost and how much you could save using AI, check out the tool we have built for that here: https://chatcreate.com/
Main Question: Is It Useful And Do People Actually Like It?
Absolutely! With e-Residency, in 45% of the cases users are able to resolve their questions instantly, instead of having to wait on hold, or wait for an email response for a few business days. This provides for an outstanding user experience.
Building The AI Chatbot Assistant
Setup
First step before setting up the bot, is to analyse the current situation. That includes understanding how many conversations people usually have, identifying most common questions and concluding whether the bot can be used in this case.
After we have defined that a bot is a worthwhile investment with good ROI, we set out to map the questions people ask. This is something that is done together with the client. There are different methods for this — you can automatically map the questions based on existing chat logs, work together with the client’s team or analyze conversations manually.
With e-Residency, we mapped the questions together with Aet, a member of their team. E-Residency had a really good understanding of what people want to know, what they actually mean when they ask their questions, in which different ways they ask those questions and also what kind of follow up questions should they themselves ask from people in order to understand customers’ problems better.
It’s also important to gather a lot of variations for questions, otherwise the bot will fail to understand people. People often ask about the same thing but word their questions differently. This means that “What is e-Residency?” and “Tell me about e-Residency” both mean the same thing — a customer wants to know about e-Residency.
In addition to variations, a good bot also has to look out for typos. This was especially important for e-Residency as people often say “digital citizenship”, write “eresidensi” or make other kinds of weird typos.
These kinds of customer support bots require natural language processing. This means that instead of only looking for certain keywords, the bot is actually using machine learning and complex language models to understand what users want.
People also like to just chat with bots. Good bots are capable of small talk. This means questions like “I like you” and “What’s up?” are something that your bot has to know how to answer. Without this, people can feel disappointed and have a bad brand experience.
It’s very hard to teach an AI to understand everything that a user might ask it, so safeguards have to be set in place. In e-Residency’s case, the bot asks people to rephrase their questions — if that still doesn’t work, the conversation is handed off to humans via e-mail.
Maintenance & Training
The first month after launching a bot is very important. This is when you get your first results. Based on these pilot conversations, you can make changes, add new questions and train the AI to be even better.
This is an ongoing process that either we or the client can do.
What if the user says something that the bot doesn’t understand?
As it turns out, in about 50% of the cases, the bot is not smart enough to give a satisfactory response to the user’s question on it’s own and human help is needed. Actually that’s not a huge problem, in this case we ask the user to rephrase the question one more time. If the AI still doesn’t understand what is asked of it, then it gives a contact email, where a person can turn in their questions. Alternatively, you could also have a live support agent jump in and take over seamlessly.
It can happen that the bot responds with something completely irrelevant. This happens when the bot either has too many intents to track, is improperly trained or hasn’t had enough time to optimize after launch.
Conclusion
We are very happy with the first results of the AI and the value it has been able to deliver both for e-Residency and their customers. There’s still a lot of room to grow. We’ll expect to teach the AI to handle 50–60% of the questions customers have. It’s important to get more people to use the live chat feature instead of other more expensive channels too.
If you’d like to learn more about what chatbots can do for your business, feel free to chat with our bot here as well as check out our website chatcreate.com where a quick calculator can help you decide whether AI support automation for you is worth it or not.
Play with our bot & check out other case studies: → m.me/chatcreatebot
Check out the e-Residency chatbot: → e-resident.gov.ee
More about AI customer service automation: → chatcreate.com
Case study created by Jorma & Juhan, co-founders of ChatCreate. We build chatbot solutions ranging from simple lead generation bots to complicated Natural Language Processing bots with custom integrations. Companies we work together include small businesses, government organisations as well as enterprises.
