Every company provides customer support on many different channels. Not every channel is easy to automate and not every channel should be automated. This is a list of insights into how to provide good service and still automate some interactions with your customers.

Knowledge Base & FAQ

Many companies complain that although they have an FAQ section or a knowledge base on their website, customers rarely look at them. The problem in most of these cases is that customers don’t expect a knowledge base to answer their question. And so they try to contact someone first.

Place the search box to the knowledge base on the contact page or integrate it into the contact form. This enables customers to find the relevant section in your docs themselves. They are more satisfied because they get help right away and your company doesn’t even have to handle a support request.

Of course this experience hinges on the quality of the search. If customers can’t find an answer, contacting a real human still has to be extremely easy. Never hide your support team.

Chat

Chatbots seem to be everywhere at the moment. But a conversational interface is still hard to do and so most chatbots are a friendlier and chattier version of the knowledge base search bar. Expect a glorified self-service knowledge base UI and be pleasantly surprised. Having said that, it might be possible to construct conversations with specific goals - e.g. booking a flight.

Email

For most customers email is the default option to get in touch with you. Fully automating email replies is nearly as difficult as fully automating chats but has a few advantages over chats: Email is an asynchronous channel and so customers don’t expect a perfect reply right away but may be delighted to get an instant response. In some cases it might be possible to automatically retrieve a matching section from your documentation and attach it to an automated reply right away.

This has to be done with caution, though. The obvious hard requirement is that suggested articles match the question. But also how the email itself is worded plays a role. Customers will be more lenient if the email emphasizes it's just the first email intended to help immediately. An actual human will read the email and respond. The email should only be marked as resolved if the customer explicitly says so.

Semi-Automation

Automation doesn’t have to be a binary choice. It's a continuous path with several milestones:

• Collect and structure documentation so your support team can reply faster.
• Make it easier for your support team to pick a pre-written response.
• Suggest pre-written responses and improve suggestions over time.
• Send some replies automatically - if the software is confident enough.

The first two milestones are organizational tasks. They are essential when preparing for automatic replies. Especially collecting and structuring support documents has the positive side-effect of building a knowledge base that customers can also browse before they contact your company.

The third step is a semi-automation strategy: A person checks replies before the system sends them. This lets you evaluate the system while already benefitting from increased automation. Once you are confident enough the last step is to let the system reply autonomously in predefined circumstances.

The semi-automation strategy is easiest for emails but can also be applied to chats.

The Future

Over the last decades we got a glimpse into automation with mixed results. OCR (optical character recognition), speech recognition, machine translation, text generation are all examples people use to show that AI "just isn’t there yet". But over the last couple of years deep neural networks have swept into every aspect of machine intelligence and have raised performance - sometimes even above human levels. This means that everything you remember about a system’s performance from before around 2012 to 2015 is outdated and really does not predict future performance.

Aside from text-based communication, an area to watch closely is speech-recognition and text-to-speech. For many specific use-cases this might already be a viable option and improvements are happening fast. Phoning call centers and explaining a problem to a machine that can answer some questions is probably around the corner.

Conclusion

Automation is not a binary choice. It can be rolled out slowly and carefully. Often semi-automation (making suggestions to a human team) makes support or sales teams more efficient from day one and can be combined with existing processes.

Once everybody is confident about the system’s quality, full automation can be used in more and more situations. Letting your team focus on difficult cases that need their full attention.