The potential of AI for customer facing applications

There is considerable interest regarding the potential of AI for various customer facing tasks, from market prospecting, to sales and after-sales. But there are also considerable technical limitations to keep in mind, when planning the deployment of AI-powered solutions to carry out tasks that have a direct impact on the customer experience. If deployed in the wrong setting, AI solutions can actually backfire, and destroy value for the firm, instead of adding to it.

 

I have been reflecting on how the characteristics of AI can support vs hinder customer service (and other customer facing tasks), and I have come up with the following formulation:

Here is my train of thought.

 

In terms of context, it is important to remember that AI works best for specific tasks, in pre-defined contexts. AI is ideal for applications such as recommendation systems, where the algorithm can learn from past purchases (this customer’s, as well as other customers with similar purchase histories) to suggest additional items that the customer may be interested in. By reducing the choice pool to relevant products, AI-powered recommendation systems may actually increase customer satisfaction, in addition to generating additional sales.

 

In addition to being specific vs. generic, tasks can also vary in terms of the type of skill required. Huang and Rust (2018), for instance, distinguish between mechanical, analytical, intuitive and empathetic skills, as thus:

• Mechanical intelligence concerns the ability to automatically perform routine, repeated tasks
• Analytical intelligence is the ability to process information for problem-solving and learn from it
• Intuitive intelligence is the ability to think creatively and adjust effectively to novel situations
• Empathetic intelligence is the ability to recognize and under- stand other peoples’ emotions, respond appropriately emotion- ally, and influence others’ emotions

 

As Huang and Rust note, skills are learned and honed very differently:

 

 

Tasks heavy on mechanical skills, mostly require the application of rules, and this is something that computers do much better than humans, because they don’t get bored or tired. Tasks that are heavy on analytical skills require large databases from which the algorithm can learn from. Human brains have great powers, but computers are much faster (and, possibly, more cost-effective) than we are at processing large volumes of data. Intuitive skills-based tasks require an understanding of the context. For instance, if I say “That’s Gucci”, it is likely to be a statement about the brand of an item; however, if this sentence is said by a teenager, it is likely to be a statement about the state of something. Some of the contexts (e.g., location) are easy to define (and, hence, are easy to learn from), but are others are subtler and more difficult to define, meaning that humans are likely to be better at this (or, at least, more cost effective).

Finally, tasks heavy on empathetic skills require the ability to understand the feelings of the other party, which is usually developed through experience – either our own experience of feeling X, or our experience of someone else’s reaction to Y. Computers may be able to read certain facial expressions, or be trained to detect changes in tone of voice. But, overall, humans are much better than machines at this,

 

Putting these perspectives together, I would say that, as a general rule of thumb:

• There is very high potential for deploying AI-powered solutions in customer facing tasks that are very specific, and which are high on mechanic and analytical skills. For instance, queries about store opening hours.
• Tasks that require mechanic and analytical skills, but which are quite generic such as interpreting medical symptoms will require AI to be augmented by a staff member.
• In contrast, in those tasks that require intuitive and empathetic skills, but which are very specific, we are likely to see AI augmenting the human. For instance, sentiment analysis of tweets for a specific brand.
• Finally, in very generic settings, requiring intuitive and empathetic skills, we are likely to continue to lead with the human touch. For instance, entertaining a crowd.

 

Hence, this formulation:

 

I would love to know what you think: Do this helps you understand the potential of AI in customer-facing applications? How does this need to be improved?