By Marc Mandel.
I’ve been fascinated with the potential brought about by AI since I first saw the seminal film, 2001: A Space Odyssey in which a computer, HAL, was able to carry on intelligent discussions with crewmembers of their spaceship. “Imagine”, I thought, ”a computer could be taught to think well enough to understand and participate in a personal conversation”. The future was right in-front of me, and I was hooked.
I was eager to have similar experiences with my first personal computer which in the late 1970’s was a Radio Shack TRS-80 (an early home PC) that offered less computer power than my current garage door opener, and I remember when I was able to try the first and most rudimentary chatbot, “Eliza”, a computer program that simulated a conversation with a virtual person.
I remember how excited I was to type “Are you smart?” and it responded, “Yes I am.” but that was it. There really wasn’t more “AI” in Eliza other than matching some words to a list and responding with a prefabricated response. The answers were little more than my toy “Magic 8-Ball” could make by shaking it and seeing an answer to any question appear in the little plastic viewer window. My interest in Eliza quickly waned, yet I remained fascinated by the idea and potential.
Using AI to improve user experience
Fast forward to the late 1990’s, when I joined the leadership team of a software company selling intelligent website avatars designed to use burgeoning conversational AI to engage a web visitor and help guide them.
Think of that annoying “Clippy” in the old Microsoft Word on the bottom of the screen that looked like an animated paper clip. I was pumped to be able to get my fingerprints on tools like this to enable better customer experiences, specifically in burgeoning ecommerce use cases.
The tech was simplistic. As it turned out, the brains of the software were little more than “if/then “statements that could identify keywords in something someone typed, much like Eliza, and respond either with direct written feedback selected from a list of canned responses, or website navigation assistance to what the person indicated they were looking for.
The software was priced well into the six-figures and targeted all the VC-backed “dot coms” who didn’t care how much money they were burning because it was all about the sizzle.
Ask what’s inside a Chicken McNugget
We had a big sales lead, McDonald’s, the fast-food giant in Chicago. They envisioned an animated “Ronald McDonald ” on their website with whom you could type requests for information about their products such as nutritional or availability questions and “Ronald” would help. It could have been magical, and we made it to the finalist round in the pursuit of that business.
We were invited into their corporate HQ to give a demonstration of the tool, all programmed and polished for the occasion and we followed a tightly choreographed, scripted presentation that went well. As the end of the meeting approached and we were feeling pretty good about ourselves, we opened the demo to the folks in the room and took requests for the virtual helper. A voice in the rear of the room shouted, “Ask what’s inside a Chicken McNugget ” which, mind you, wasn’t part of our script and nor had we prepared with keywords for it.
We cautiously typed it into their website and got a very strange answer:
“Ray Kroc is inside a Chicken McNugget ”
Kroc was their storied founder and a hero to many in that room, and the answer that came back horrified them. It could not have been more wrong and come at a worse time. They threw us from the room and that was it. We were eliminated from the deal and were laughing stocks.
Terrible, for sure, but again, like with Eliza years before, it made both a positive and negative impression on me. Positive, for the potential of what could be, and negative for what was. From a pure customer experience, had McDonald’s adopted our approach, they would have antagonized thousands of their customers who were looking for important information and got truly bizarre outcomes. The risks outweigh the rewards and they ultimately concluded that the tech wasn’t prime-time ready.
Leveraging AI to understand customer sentiment
It took about ten more years but again I was given an opportunity to work in the customer experience AI solutions field when I joined a well-known industry startup. The nascent company barely had what you might consider a minimally viable product (MVP) when I joined them in 2007 but I was excited to be part of it. The notion that AI could help a business to understand customer sentiment about experiences was the killer app in my mind and that the company was going to be a rocket ship to the stars.
That MVP was barely functioning and largely unpolished. We sold it to companies looking to understand survey data better and unlike before, began to apply natural language processing (NLP) techniques that went well beyond keyword pattern matching.
The progression of the AI was astounding to me. Tools could begin to identify not only what someone said, specifically in what words they used, but could start to recognize the semantics or said otherwise, what they meant, regardless of the words themselves. This was liberating and boosted the value of tools like these orders of magnitude higher than just the earlier keyword spotters.
Multiple platform suppliers continued to strengthen their offerings with greater abilities to integrate natural language understanding (NLU) approaches, expanding their ability and pushing the envelope more deeply as people became increasingly aware of these tools and their benefits. Imagine, for example, being able to “tag” a customer comment with a topic identifier of, let’s say, “Legal” regardless of whether the verbatim was “I’m going to sue you!”, or “I’m calling my lawyer on you!”, or “I’m going to drag you into court!”. The software was able to see these as equivalents and all connected thematically to “Legal” and label them as such, even in situations where the comments were in complex, non-English languages.
We recently published a blog that talks about AI marketing tools; why don’t you check it out for more ideas?
The fear of employing AI for responding to customer complaints
Significant commercial use cases sprang up, none more important than complaint handling in financial services. In the US, regulators have put stiff rules and huge penalties in place for banks, insurers and investment companies who don’t respond to written customer complaints timely.
At first, as more companies started to deploy survey programs to measure and assess customer experiences, there was pervasive fear about these regulations and opening surveys to written comments.
A survey response even suggesting mishandling of a customer’s funds was a huge thing and would trigger compliance workflows and reporting that had to happen very soon after the complaint was issued and failure to comply would bring large fines and other, even more serious consequences.
At first, these companies were so fearful, they refused to ask anything open-ended on their surveys as they did not want to trigger this process should someone respond and complain.
The software’s ability to “read” open-ended feedback and identify patterns of complaint language proved over time to make this workflow much easier and faster and even more accurate because it was less subjective than a human evaluator. A complaint could almost always be found, labeled, and routed, often irrespective of the volume of feedback with nearly immediate access. Even just this one example made AI a homerun, both for the company and customer.
Again, a few more years later, AI enabled other types of customer experiences. One such example grew in popularity in the insurance industry who found AI could not only be used to identify signs of potential claims fraud, but could also be used to process claims holistically. Purely from a customer experience standpoint, a strong predictor of a good customer experience in insurance is speed and accuracy of service. AI delivered on both.
A claim could be filed, processed, and closed nearly instantly with AI acting as the “middleman” to ensure both fast and accurate outcomes for the customer and for the insurer. The claims handler took on a more supervisory role, overseeing the software and ensuring the process. The very term, “AI” in this case took on an interesting and different definition from its original definition of Artificial Intelligence.
AI was more about “Augmented Intelligence” and not so much replacing the human with bits and bytes but rather augmenting the human ability to serve to higher and higher levels of output, quality, and speed.
Customers were the big winner here, often creating a surprise and delight opportunity out of a stressful one.
Unlocking original responses
Again, fast-forwarding a few years, AI has continued to evolve and begin to test the waters of not only understanding the language but truly generating original information instead of just regurgitating a pre-canned response.
Some of the very same building blocks that played a role in teaching a computer to first identify basic keywords and later tie them together into linguistic phrases and determine meaning and intent were now able to be used to unlock original responses generated by the AI.
Many industry folks refer to this as “Natural Language Generation”, “Generative AI”, or just NLG and it’s all the range, although still in its infancy as I write this in early 2023. Much like the computer users of the 1970s who saw Eliza appear to be a breakthrough and then prove not to be, or the 1990’s early “bots” enter and then leave the scene, often in despair, early NLG is both groundbreakingly exciting and terrifyingly dangerous. In some ways, and in the wrong hands, it’s incredibly full of risk.
Artificial Inteligence in 2023
The promise of what’s to come is nothing short of amazing and game-changing.
We’ve seen the news about ChatGPT everywhere, with stories about how it can be used to create original content and theoretically threaten content creators everywhere.
AI it’s only in its infancy, and as such, makes for an exciting demonstration or proof of concept, but I would not put automatically generated content of any kind into business use cases that would expect and rely on original, thoughtful, accurate information.
While night-and-day better than Eliza or even the website bots and early NLU technologies, NLG is here, on the scene, and exciting as hell, yet only just now beginning.
In many ways, we are still in the “Eliza stages” of NLG.
Over time and with millions of people coding experimental applications to leverage various versions of NLG tools, these technologies will improve exponentially but in their overall capability and accuracy.
There is a ton of math to back this up and in time, we’ll see something in these tools that will take your breath away.
As for the future of AI, I fully expect we’ll start to see our Alexa smart speakers truly understand and accurately respond, much like the fictional HAL computer did back in the 1960’s movie.
No, I don’t mean go on a murderous rampage the way it did, but automate a conversation to not just reply but to truly understand. Link that together with some simulated emotion (assuming such a thing exists) and you can get machine-empathy.
Wow. When this day comes, we as CX practitioners will see an almost endless list of potential benefits, from call center headcount reductions to higher sales conversions to overall more delighted customers, the payoffs will be huge and the horizons for these impacts are near.
Again, I am excited for what’s to come and can’t wait to see it play out and touch all our lives in ways we’ve never imagined.
Would you like to check what the trends for this year are? Then come along and download our first-ever collaboration with CX and VoC industry specialists sharing their perspectives on the key trends for 2023.
What the guide includes:
- The top CX trends (Speed to market/response, doing more with what you have, Proving ROI, AI specialized services, automation)
- Bonus trends+
- The recession effect in CX
- Wrapping up with the key trend takeaways
- About the authors
Get the trend guide now and dive into the key CX trends delivered under the lens of six remarkable CX industry experts for this year.