Will IBM Watson help customers make better choices?
IBM Watson, the natural language processing platform that once won Jeopardy, has taken a job in customer service. The North Face has implemented solution provider Fluid’s Expert Personal Shopper (XPS) software, which uses Watson as its basis for decision making.
When customers visit the XPS page on The North Face website, they are greeted with the question, "Where and when will you be using this jacket?" The customer’s answer leads to further questions about the product and its intended use. Customers can answer the questions in plaintext (ordinary language), rather than having their answers restricted to a menu of choices. The responses help to refine the product selection and guide the customer towards the jacket that best suits their needs.
XPS currently only supports searching for jackets. If a customer responds to the initial question looking for shoes, the program returns a message indicating that Watson hasn’t been trained to deal with shoes and directs the customer to The North Face’s website.
Source: thenorthface.com/XPS
The use of natural language processing to help streamline customer service inquiries is a growing trend. Earlier in 2015, TechCrunch reported a solution called DigitalGenius, which acts similarly to XPS only through text message rather than a web interface. DigitalGenius is meant to answer repetitive customer service questions. Last year, Motherboard reported on artificial intelligence platform Amelia from IPSoft, intended for the same purpose.
Advocates see the technology as having the potential to free customer service to focus on more complicated questions. Detractors worry it could lead to job cuts. In fact, the Motherboard article reported that overseas call centers were concerned that lower-tier support now outsourced abroad would be more easily managed through a natural language processing solution.
But in the case of XPS, it’s not a resolution the customer is looking, just a product suggestion. One wonders how helpful a program could truly be in this regard. Processing plaintext input is one thing, but Watson doesn’t have a sense of what looks good. At least, not yet.
- The North Face, IBM and Fluid Launch New Interactive Shopping Experience Using Artificial Intelligence (AI) – The North Face
- Your Personal Intelligent Shopper Beta – The North Face
- DigitalGenius Bring Artificial Intelligence To Customer Service Via SMS – TechCrunch
- The Artificially Intelligent Call Center Operator – Motherboard
Discussion Questions
Could a recommendation engine like Watson/XPS become a valuable part of customer service, or is it just a novelty? Are other uses for natural language processing, such as answering common call center questions, a good or bad thing for business?
Poll
BrainTrust
Ken Lonyai
Consultant, Strategist, Tech Innovator, UX Evangelist
These are still very early days for AI, especially in retail. Compared to Watson’s capability, this is a very simple implementation, but a good start. Watson “learns” from a corpus of data, so the more data given to it and the more analytics acquired and related back to the data, the stronger its insights get.
There are a bunch of low level natural language service “agents” I’ve encountered on websites and they’ve all been miserable, being easily stumped and throwing out unhelpful canned responses. Watson has a lot of power when NLU (natural language understanding) and TTS (text to speech) APIs are layered over its core.
Any forward-thinking retailer needs to give this sort of AI serious consideration ASAP or be prepared to be left in the dust sooner than later.
Even if it were just a novelty, it would help bring a bit of fun back into the shopping experience. I think as a product recommendation engine it’s a great idea.
As a call center tool, not so much. The call center experience is annoying enough already.
The ability to provide consumers with great recommendations for products or services is a critical gap in today’s online shopping experience. The standard online shopping approach of “browsing” through poorly-indexed hierarchical categories needs to be dramatically updated to the more fast-paced and immediate gratification needs of today’s shoppers. My hope is that with processing power and heuristic capabilities a service like Watson can provide more relevant recommendations to consumers.
Assisting in customer service might be the first application of machine intelligence in retailing, but the real play might be to gain a better understanding of customer preferences, leading to increased sales.
One of the benefits of customer loyalty programs has always been to provide a contract where striking up a dialogue with customers is more natural and the information exchanged more valuable — can we say, more “real”?
Think if the same types of insights could be gained via Watson or something like it, and in real-time while a customer is in the shopping environment. That is powerful.
For recommendations a screen image would help. For helplines — I would still just keep hitting zero till a human answers. Watson or other NLP tools for helpline are a failure. Ask Comcast. I am dropping their service mainly due to their terrible customer service. No matter the technology, if you cannot serve your shopper directly, human to human, then you are in trouble. I am tired of the machine saying “I didn’t quite get that — please re-enter your phone number.”
In thinking about it this morning I don’t think I’ve ever actually bit on a “recommendation” from Amazon, eBay or any other company trying to help me out. Will a semi-solicited recommendation from Watson make it more palatable and on-target because they may know a little more about me? I don’t know. In general, at this stage of its technological development I am not in favor of breaking the human bond for AI intervention especially when it comes to higher-end merchants. While it may be fun for some shoppers with time on their hands to play with a recommendation engine I think this will just be a time-wasting novelty for quite some time.
The buzzword(s) of the 2016 marketing world may be “cognitive analytics.” IBM Watson is starting to think, not just report. When asked a number of questions, Watson makes recommendations. Even for something as simple as, “What should I buy my 12-year-old nephew for Christmas?” Watson can help me. It will probe for more information and make excellent recommendations.
Is this good for business? We’ve reached a point where the quality of language processing is good enough to be accepted. And the software driving the consumer’s experience is working. Customers are enjoying the experience. That said, some still want to talk to a live person. That option has to be available and the employees must be well-trained to answer the customer’s questions or resolve their problems.
You have to start somewhere and the North Face beta sounds like a nice start. Kudos to the company for getting the ball rolling rather than waiting for a large-scale, bug-free solutions (which would require someone else grabbing first-mover advantage). The open-ended text-intensive interface could be polarizing — annoying to slow finger/thumb typers, awesome for those whose fingers float across a keyboard. If the solution is truly intuitive and leads to a more engaging and satisfying customer experience, wider adoption will naturally follow.
Having attended IBM’s recent Insight conference, my head was spinning (in a good way) thinking about the many retail applications of cognitive, Watson-based solutions. The possibilities are endless and, regardless of the short-term results from various tests, the train has left the station.
Here is my problem with the example and the first question “Where and when will you be using this jacket?” Like all e-commerce sites today I think many shoppers being asked this question will believe that North Face assumes they are ready to buy in this session, when there is perhaps a 4 percent at best conversion on product page views. Yes, the website might suggest adding the item to a wishlist if the viewer leaves without buying, but shouldn’t this technology be more customer-centric and mirror the in-store experience? I doubt sales associates are trained to go right up to someone at the jacket rack and say “where and when will you be using one of these jackets?” That puts the shopper on the spot, and also assumes they are ready to buy and not just killing time by browsing.
And to take this to the next extreme, let’s say the shopper pauses, looks confused and walks away when the the sales associate asks the question. Can you imagine the associate following the shopper out into the mall and saying: “Since you can’t tell me, let me start guessing.” In effect that’s what AI predictive analytics is doing, guessing. And yes, it’s getting better, but what’s missing still is to simply ask the shopper for consent to speak to them about marketing the products in front of them on criteria the shopper wants to set for their permission.
If the recommendation engine was solely a part of the customer selection process, its use over time might increase beyond anticipated levels. The ease of use and reliability increases to error-free calculations are, as always, the deciding factors as to expanding use.
The next level of consumer demand is speed and execution especially at check out. Most companies that experience consumer evacuation after an initial astounding success fail to keep up with the growth of system access and utilization. The root problem is the cost of off-peak system idleness devouring profit dollars at an alarming rate, diminishing the funds needed for the system fixes. When you add to this dilemma the nightmare of software controls for scheduling enterprise distributed processing for corporate accounting processes and reporting, you may start to see things like a shutdown here or there and maybe everywhere.
Information Technology ( IT ) vendors are all united in the recommendation(s) which will always call for more software tools and hardware which recovers none of the sustained loses.
As for after sale customer service, the 4th and 5th generation software now available does not have true intuitive capabilities. The systems must use inquiry and selection methods to determine consumer needs and/or wants better known as grandiose “if then” programing code. This process is doubly expensive in terms of system size and utilization adding to the client(s) dissatisfaction with the sale results, therefore yielding a larger loss on investment.
For many retailers of any size, it is often not a bad idea to get to know the third party IT vendors supporting the latest proven technologies that are of interest. This will provide very valuable company live information when the decision to bring targeted process capabilities in house is on the table.
I think recommendation engines are a nice idea in retail in order to help shoppers when they aren’t able to talk face to face with an employee. Live chat has already been implemented on a number of eCommerce sites and filtering out the basic questions will free up employees to talk about more complex questions with consumers.
Since this idea is in such an early stage, there’s no telling what impact it will have on the bottom line and call center employment for retailers. At the end of the day, I agree with Tom, talking with a real person is much preferred. There are so many nuances that a recommendation engine can’t grasp just yet. This will be an interesting story to follow.
About 30 years ago I had a consulting company that was a member of the Help Desk Institute. Does anyone remember that? We started using AI techniques to assist in customer service and call center service. AI tools were everywhere. My company built custom solutions, much like primitive mini-Watsons. Most everything knowledge-based back then was based on heuristic engines. Watson goes so much further and I am happy to see where the technology is going. BUT back then there was a little thing called the Second AI Winter that hit around the early 1990s. (The first AI winter hit R&D mostly, a decade or so earlier.) It resulted from too much hype and too little delivery of useful results. It slowed things down in the AI realm for a good while.
I have been doing this stuff since I was an undergrad in the early 1970s. With each cycle of rebirth things jump ahead and new terms are coined for modified technologies. Thus it all seems new full of promise. I am always amused by people asking me if I have heard about the new AI tools that will change the world! Advantage now is that CPUs and storage are so much faster and cheaper. If today’s vendors over-promise and people expect too much too quickly, it is going to get cold again.
Now that the new hardware, the web and the cloud make everything doable quickly and cheaply, I am taking my 20-year-old AI projects off the shelf and finding investor interest. If all goes well you will be using such in a few years — with sexy new techie names, of course.