Designing Simple Chatbots for Customer Service
Social customer service is not a new phenomena. Many brands and companies encourage customers to use social channels to send enquiries. Social media managers are increasingly asked to fill customer service as well as marketing roles. A well designed chatbot is an excellent way to supplement a customer service team and provide one-to-one service at scale.
Most consumer facing companies will have an FAQ posted on their website. Website FAQs’ quality can range from being somewhat useful to completely unhelpful. FAQs quickly expand to unwieldily lengths, so as to answer the majority of potential questions. Search within FAQs can help FAQ overload but often feels like a step too far for many customers. Instead of typing a query into a search bar, a customer could simply send a message to a customer service operative and get a precise answer on their specific issue. Generic FAQs therefore suffer from two main downsides, when compared to customer service.
- If the FAQ is long enough to cover most questions, it’s probably too long to be easy to use
- FAQ answers need to be interpreted by the user to fit their precise issue, as FAQs are often written as generic responses
The issues with simply relying completely on a customer service team however are obvious: it’s expensive and difficult to scale up quickly if needed. Supplementing a customer service team with an FAQ chatbot is an easy way to answer a large number of routine customer enquiries while freeing up the customer service team’s time to answer more difficult ones.
Designing such a chatbot is fairly straightforward, involving research, planning and development. I’ve outlined my general process below drawing from my experience working with simple chatbots on WeChat, though I think my suggestions are applicable to all platforms.
Data, intuition and knowing your customers’ problems
In a previous role I consulted with a client whose customer service team was swamped by enquiries through their official WeChat account. Once we had gone through a large enough sample of customer enquiries, the solution to their problem seemed clear: an automated system should handle a majority of customers’ questions.
In this case I was lucky to have a good amount of raw data to work with however internally many at the company knew where the potential customer pain-points were and (despite doing their best to minimise these) had predicted exactly the kinds of questions they would receive. A brief interview with the in-house UX designers and
a few customer service staff could have produced a fairly similar list of enquiries as my data-led approach, however would have made phase 2 more difficult.
Once we had a clear picture of the scope of enquiries, we planned a number of answers that covered each enquiry and briefed our copywriting team.
Keywords and trigger phrases
Planning keywords to trigger FAQ messages needs to be done with care. The chances to confuse customers with the wrong answer are high. Additional confusion to enquirers with customer service issues can easily result in added work for customer service representatives, making the chatbot a hinderance rather than a help.
If you have the data of previous enquiries to work from, this is the logical first place to look for keywords. Word frequency in this data set should show you the common terms used in each question. If you don’t have enquiry data to mine, then this is time to go talk to your closest SEO expert about keyword finder tools. Google, Baidu and other keyword tools can help you determine relevancy and frequency of search terms. SEO keyword tools aren’t designed to just show customer service type enquiries to Google, so UX designers should rely too heavily on such keyword lists for chatbot responses and be careful to sense-check any keywords chosen. However I have found them useful places to start.
Finally, scheduling regular testing and refining for any chatbot is vital. Testing user journeys and analysing queries to refine FAQ and keywords will ensure that your chatbot will continue to support the customer service team.
Advanced development
The ultimate goal for a customer service chatbot is for it to be integrated with customers’ account information. For instance, a customer might be able to enquire about the status of an order or receive specific after-sales service. This involves a developer to connect the API of the messenger to the API of the company’s customer management tools. As messenger platforms are opening up their APIs then this kind of integration will become more common, and expected by customers.
