“I can’t believe Vanessa, my bride, my one true love, the woman who taught me the beauty of monogamy, was a chatbot … all along.”
At this point, just about every online or offline retail customer has in some way interacted with a chatbot — an artificial-intelligence-based algorithm that impersonates a human during an online chat conversation. At a baseline level, chatbots can provide intelligent service at scale, reducing organizational support costs. However, chatbots based on sophisticated machine learning algorithms can often times be better than a real human — giving customers a natural and intuitive self-service experience. Also, chatbots never sleep and are available on any digital channel, enabling customers to engage on the services or channels they prefer — the same services they use today for social interactions (such as Facebook, Instagram, etc.).
Customer Service Chatbots
Nowhere else is AI more prevalent than in the customer service department. Most of the large retailers have an online chat function – and a surprising number of the chat facilitators, aren’t actually human. (I know, creepy right?). But instead they are AI algorithms that learn appropriate responses to questions over time.
These chatbots have varying levels of sophistication. The most basic ones are simply a programed answer to basic FAQs. If you click on an FAQ link, then a pre-programmed response will be sent back. While this is better than nothing at all, it is hardly that much better than just reading the website. Unless your question is an exact match, then there will be no response.
The next level is a “decision-tree” based chatbot system. This is usually a complex mapping of if-then scenarios: If the customer asks this, then the chatbot replies this way, and then if they ask that, then the chatbot uses a different reply.
But if the customer either asks something that is not anticipated, or if they request more information that is not pre-programmed into the canned responses, then the customer is simply out of luck.
The next level of chatbot is one that uses machine learning. Based upon the customer’s questions and responses to the chatbot’s dialogue, the chatbot algorithm continually improves the relevancy and accuracy of the response over time. ML chatbots — for obvious reasons — are quickly becoming the norm.
Finally, the best chatbots actually link into a number of other databases so that it can actually offer the customer products and solutions to their questions — so customers can ask what is the best chef’s knife for boning fish, or where can I go and pick up that cut of beef, etc., the chatbot can actually make a product recommendation with a link to buy.
The graphics below show an example of an advanced chatbot session for a make-believe car insurance company. You can see that not only does the bot answer the customer’s questions, but also can present multiple choice questions to help narrow down the customer’s car type and actually provide the customer with a quote based upon their supplied data.
Another type of chatbot that is becoming more frequently used is one that helps customers by orchestrating multiple tasks, including finding products, making recommendations or pushing promotions. These chats can then be sent to the customer on their cellphone or PC, either through an APP or just a plain text.
Below is an example of excess inventory being a trigger for an automatic promotion. These kinds of automated promotions can be triggered from a number of events besides just overstock — such as weather, holidays or sports events.
This kind of chatbot specifically recommends the right product for the customer’s needs. Below is an example of a recommendation bot for apparel, and they can also be used for wine, cheese and perhaps even A/V accessories.
Examples of Chatbots
Below are some recommendations for chatbots available to any retailer:
Live Tiles (for employee productivity bots)
Tallan (for consumer-facing bots)
Bizzy (building bots without code)
Microsoft’s QNA Maker (a free service by Microsoft that lets you build, train and publish a simple Q&A bot based on FAQ URLs, structured documents or editorial content in minutes)
Just so you can quickly get in the know, here are some essential chatbot terms:
Affordances: Affordances are all the possible actions for an object, environment, or app. Conversational interfaces have fewer visible affordances (controls, menus) than traditional apps. This can make it harder for users to discover possible actions. It is critical that chatbots provide clear prompts, menus, responses, etc., to make it obvious what tasks the chatbot can/can’t perform.
Chatbot prompt: The chatbot prompt is designed to initiate the user experience. Prompts can be open ended or offer explicit, contained lists of options.
Chat window: The chat window is the container for the chat interaction. On a phone it will use the full screen. On the web it may exist as a separate screen within a larger web experience. When the chat window is a separate screen, resizing and responsiveness requirements are important design considerations
Conversation date time stamp: This indicates the date/time of the interaction, which may be useful for time sensitive tasks.
User input prompt: This prompt indicates where the user types in their chat content.
Utterance: In language analysis, an utterance is a smallest unit of speech, a continuous piece of speech beginning and ending with a clear pause. In the context of chatbots, utterances are the user’s words or phrases typed by the user. It is critical that chatbots understand the wide range of possible utterances users will attempt to use to communicate with the chatbot.
Actor identity: This identifies who is communicating (user or chatbot). We can anticipate group chats in the future with multiple people and/or multiple chatbots involved in a single conversation.
Feedback: Well-designed UX keeps the user in control at all times by communicating what is happening (feedback). Types of feedback include implications of choices, task status, system status and error conditions.
Scrolling interface: The conversation thread is a scrolling interface. Scroll bars are important to enable users to easily move back and forth to refer to info earlier in the conversation.
Conversation pace: Humans do not communicate with each other in sub-second responses. In human interactions there are short natural pauses. To make chatbot interactions more human, responses should feature a short, natural delay in responding.
Auto suggest: Auto suggest is a UI feature to speed data entry. Auto Suggest queries predefined tables and displays records that match the text users type into a field. As a user types text into a text field, one or more possible matches are presented for selection.
Constant info: Record info, such as account balances, may need to be visible alongside the conversation thread to provide data context for the users. This snip shows “floating” account info in a banking chatbot.
Human agent: Users will sometimes need to switch from conversing with a bot to interacting with a human to complete a task. There should be an obvious way for users to request to initiate interaction with a human to help them complete their task
Action controls: These are controls that enable users to take specific action(s). Control can also be accomplished via prompted key words (e.g. type approve to approve this expense report).
Noah Herschman is a Microsoft retail industry senior architect with over 30 years’ experience in CE retail, including stints at Tweeter, Amazon, Staples China, eBay, DHgate and Groupon Goods Asia. His partner ShiSh Shridhar has worked at Microsoft for more than 20 years, currently as retail industry lead for data and analytics. Together they are creating technical solutions that are sophisticated in design and specifically targeted to improve businesses by engaging customers, empowering employees, optimizing operations and transforming products.