No EQ? No Problem, for AI with Machine Learning Capabilities!

Imagine the following scenario: you’re a very conscientious content writer, and you want to make sure that you’re using phenomena or phenomenon in the correct singular or plural context. You go to Google to search for “is phenonema plural”? As you’re typing in the search field, the first suggested search after typing “is phenomena” is “is phenomena contagious.” Curious about what could possibly make the word phenomena related to infectious disease or contagion, you select that suggested search. And you get the following top result: “is pneumonia contagious?” on the NHS’s website.

To be clear, this happened to me today. Google didn’t even assume for a second that I meant to search for the word “phenomena” in relation to the word “contagious” and promptly redirected me to results about whether pneumonia is contagious. The results didn’t even come accompanied by its usual gentle prompt of “did you mean…” prompt at the top, instead assuming the best for my search needs. The fact that multiple people must have misspelled phenomena so drastically to actually mean to go to pneumonia is equal parts hilarious and frustrating. And now this has become a top suggested query for the word “phenomena” to make sure that I get the information Google thinks I need. All this to say that technology is steadily becoming smarter in order to meet needs we didn’t even know we had, in the efforts to improve our experience with a given platform.

Building on AI Algorithms with No EQ

We once hosted a lunch and learn about how machine learning could be used to humanize artificial intelligence algorithms. We described how AI with ML capabilities could bridge the gap of algorithms with no EQ, or emotional intelligence quotient. This way, the AI is able to continue learning using the ML technology and continue adding to its knowledge map of how to parse and understand requests of its programming. At OSG, our technology uses both AI and ML to summarize and understand data and to build algorithms and models for continuing to learn and understand customer data and behaviors.

We’re still seeing versions of this humanization of AI/ML technology in Google’s newest announcement around launching new AI algorithms, called Multitask Unified Model or MUM, for their search functionality, specific to natural language processing (NLP). Google is working on getting their search engine to process more and more complicated requests and questions. With NLP capabilities, Google search results will be able to utilize its existing knowledge map functionality to present all possible related information to your search, by better understanding all possible language nuances of your query. The search engine will be able to correct misspellings, understand complex questions, and even fill in the gaps on questions you haven’t even fully fleshed out when asking! Look forward to the advent of MUM and superpowered search results in the future!

A Case Study of AI with No EQ in Chatbots

In December 2016, a customer of Lemonade Insurance Company hit “submit” on an insurance claim for a $979 Canada Goose coat. Approximately 3 seconds later, Lemonade’s claims bot, AI Jim, had reviewed the claim, cross-referenced it with the policy, ran 18 anti-fraud algorithms on it, approved the claim, sent wiring instructions to the bank, and informed the customer that the claim was closed.

However, not all chatbot experiences are like the Lemonade experience. Despite the sudden surge in chatbot usage, most of us have had clunky conversations with chatbots on retail or financial services websites.

These include, but may not be limited to:

• A poor user experience with bot not understanding the dialect
• Confidence in the product offering decreasing
• Providing us a good reason to go somewhere else that offers a similar product
• Redirecting ultimately to a call center, which is not what we wanted to do in the first place

Why are AI chatbots today not so smart?

Many retailers are embracing chatbots for the sake of chatbots – which in most cases is unlikely to yield any meaningful results. Retailers need to integrate chatbots in ways that interest consumers and make their shopping experience more personalized and entertaining to their needs. It’s all about working the goals of a consumer once that interaction happens, which is often not planned effectively with any chatbot integration.

The key issues chatbots battle with today are:

1. They use historical data gathered from your CRM, social feed etc. to attempt personalization
2. They are unable to learn in real time. This means chatbots cannot follow live requests being made by a customer in a chatFor most organizations, chatbots are insufficiently trained. They cannot understand things like human expressions of frustration or delight, acronyms, etc., and thus end up giving inappropriate reactions to situations
3. Contrary to the perception of an automated service, chatbots need continuous tweaking and monitoring to get the best from business goals, which in most cases change at a rapid pace

How Does AI Learn and Grow?

Often, there is a trade-off between the best technology, speed of integration, & actual consumer POS requirements. Without understanding the core needs of their consumers and what drives their purchasing decisions, many retail and financial services organizations are being shaped ineffectively by the “potential” of messaging apps, nudge technology, and chatbots as the future for improved customer retention and churn rate.

A frantic race to full automation and the AI “blockchain dream” is also to blame. Many organizations want to use chatbots to connect with customers in as many ways as possible, without thinking of the financial implications, technological dependencies, and consequences.

For example – designing sophisticated chatbots that:

• Give out retail advice and product usage information
• Award frequent users with discounts and loyalty benefits
• Assist with shipping and logistics
• Provide bi-directional marketing, i.e. give out information about the product to the consumer, whilst also simultaneously collecting information for market research through surveys and feedback

Emotions run deep in every human interaction and are unfortunately ignored. Deciphering these human emotions by way of studying voice modulation, facial expressions, text tonality etc. can reveal a wealth of information that can add context to a customer interaction with a chatbot and make the interaction more fulfilling. Only if AI has the capability to empathize with users’ feelings and learn in real-time about what the customer wants can it design the perfect responses, leading to outstanding user experiences.

This means that organizations must:

1. Ensure that chatbots quantify what matters to customers when making a purchase decision, and by how much
2. Ensure that chatbots can determine a customer’s emotional state

How Can OSG Help?

OSG uses its proprietary trade-off methodology ASEMAP™ to let customers prioritize benefits that are best suited to nudge their behavior toward purchase. By using a combination of cognitive analytics (historical data that looks at the “who” and “what” behind customer decision-making) and future-looking behavioral analytics (that go beyond traditional analytics to identify the “how” and “why” behind customer decision-making), OSG can help you and your technology learn in real time what matters to your customers. Our big data analytics platform can parse text, tone, and voice analysis to understand customer sentiment, helping you to humanize your chatbot during an interaction. We know how best to combine AI and ML capabilities to give you the actionable insights you need to make your brand and customers more connected. Contact our experts today to learn more about how our technology can work for you!