A Chat Bot by Any Other Name…

OpenMarket – December 13, 2016

chat bot, machine learning, keywords, SMS, rose

Chat Bots, Machine Learning, Human-Assisted Virtual Assistants, Expression Language, and Keywords: what an interesting bouquet of roses we have! Each smells sweet, but has its own particular thorns.

Keep in mind these technologies represent different communication approaches to address this key problem: What options do businesses have to provide an SMS engagement experience that they can afford, and that provides the mobile experience that will satisfy their customers?

My thesis is that there is a balance between the cost and benefit for applying each of these technologies to different mobile engagement use cases rather than businesses having to use a single technology for all use cases in order to provide an acceptable mobile customer journey that the business can afford. Let’s examine each option more closely.


Keywords have been offered in two-way SMS ever since businesses starting texting their customers. As an example, here is a simple SMS message: “Text JOIN to 10991 to opt-in to this SMS program.”  JOIN is the keyword and you have to spell it correctly and not include any characters before it in order for it to be recognized correctly.  Another issue with using keywords is that some assume that they can respond with synonyms. For example, a customer might receive a text message from a business that says “Were you happy with the service you received?  Reply Yes or No.”  If the customer responds with YEAH or Y, then those keyword aliases might not be recognized if the business didn’t anticipate all the likely synonyms and add them as keywords to their messaging program. Another challenge arises if you are using a short code that is shared between two SMS programs.  In this case, both programs must have distinct keywords for each SMS program so you can’t have both programs prompt for a Yes/No response. The other solution to this dilemma is that and you have to use two sequential keywords (e.g. Yes 1 or No 2) to disambiguate the response from the customer.  Let me throw a final consideration with the limitations of keywords.  Imagine that a customer receives a message that says “Text HELP for assistance,” then they text back “I need HELP now and I’m willing to pay for it.”  Since HELP isn’t the first word in that sentence, the SMS program won’t recognize “I” as a valid keyword and thus, will respond with a message to the effect of “unrecognized command, please try again.”

Expression Language

To solve this last problem, you need software that can read what the customer texts in and interpret its meaning. OpenMarket’s Mobile Engagement Platform uses Expression Language to identify HELP in the text message: “I need HELP now and I’m willing to pay for it”, and respond correctly by sending the HELP message to the customer. So you can see how Expression Language can start to tackle some of the limitations with keywords. Using keywords and expression language is particularly good at addressing problems that have a limited and well defined set of expected responses from customers. For example, one hotel is using this approach to address the most frequent requests from their hotel guests instead of requiring their guests to call and speak with someone for assistance. For instance, they can request a wake-up call, request room service, or request a taxi.  This approach was very successful in reducing the amount of live phone calls that took up hotel staff time, as well as allowing guests to self-serve on a communications channel that they found more convenient: texting.   But even expression language has its limitations.  Expression language is essentially a fixed set of rules that are applied repeatedly to the same text messages it receives….thus it doesn’t learn to avoid making the same rules-based mistakes repeatedly.

Machine Learning

Machine Learning (ML) software is required instead of rules-based scripting language like expression language to have a solution that learns from its mistakes. So when a hotel customer texts “Please make a hotel restaurant reservation for my spouse and me”, the ML software is trained to understand that “my spouse and me” means 2 people. Machine learning software requires training of the Artificial Intelligence software rather than explicitly programing it.  Machine learning software will work best for use cases where the context of the person’s text message is well understood and narrowly defined.  This means that ML software can be trained to handle reservations where it is expecting to get information on the number of people in the party and the time of the reservation.  However, people can be unpredictable. A customer texting for a reservation request might say “I want a reservation for tonight at 8 for a party of 4, but can you confirm that you have vegan dishes on the menu as well?”  The ML software can be trained to make the reservation and fail to understand the request about whether there are vegan dishes on the menu.

Human-Assisted Virtual Assistants

Machine learning supplemented by human assistance is analogous to Interactive Voice Response (IVR) systems today.   When your customer calls your business’ IVR and is given a set of self-service options, they can only self-serve when the IVR recognizes the caller’s request.  When that fails (and it inevitable does for some percentage of the calls), the caller is transferred to a live agent: a human being who can recognize their issue and assist in getting it resolved.  This model of using trained machine learning software to handle the initial messages from consumers with humans handling the unrecognized requests enables the highest level of correctness.  This approach provides greater understanding of the customer’s intent and context of their message while still attempting to have the majority of the customer’s messages be processed and responded to automatically via software, but it has the highest cost due to the inclusion of human assistance.

Chat Bots

My broad definition of chat bots is that chat bots are software designed to respond to the conversational abilities of humans. I believe that keywords, expression language, machine learning, human-assistance virtual assistants are all valid forms of chat bots because they are all part of the answer to this question: What options do businesses have to provide an SMS engagement experience that they can afford and that provides the mobile experience that will satisfy their customers? The four options I listed above were presented in order from least expensive with the lowest quality, to the most expensive with the highest quality.  You get what you pay for, but you should really only pay for what you need for a particular use case.  Why use human-assisted virtual assistants when keywords will work fine for a customer satisfaction survey?  Conversely, why wouldn’t you use human-assisted virtual assistants for use cases that are revenue generating and result in customer purchases or acquiring new customers?!

Why OpenMarket

I’ve just laid out 4 different types of roses (aka ‘communication technologies’) and explained how each smells nice and what the particular thorns are. The good news is that OpenMarket offers all of these technology options, as well as the mobile messaging expertise to guide you in selecting the right type of chat bot that will produce the business results for your particular use case. We support two-way SMS to over 80 countries and can work with you to select the right message originator choice for your particular use case. I invite you chat with us to learn more from our mobile engagement specialists about the right type of chat bot for you.

See all blogs

Related Content