AI and Customer Support

Two years ago we embarked on a journey to install solar panels for our house. With Austin’s snow storms growing increasingly disruptive it felt like we needed the right design, number of solar panels, batteries and a provider who would understand everything from the slope of our roofs to the size of the house. 

After talking to a very well-known provider in the space (a human too) we decided to go with a different company. Our reason? Everything beyond the initial discovery and sell phase would be automated with the well-known provider, leaving us with limited to no chances for highly personalized answers and traditional “white-glove” support. The project proved to us why we had made the right choice as well, just given the amount of human interaction needed to debug and reconfigure through various phases of the install.

More recently, every professional conversation I seem to be having veers into the use of Generative AI in customer support. The use of AI for customer support is neither new nor untested but the notion of using more powerful generative AI for it, is top of mind for companies.

Companies with a mission to be known as customer-centric or customer-first are actively deploying automated agents or chat bots that are much more complex in nature (relying on LLMs rather than simple rules). The key here is to measure the success of this automation via “successful” resolution rates, the ability to prevent a second contact about the same issue within a set period of time, etc. 

Business Process Outsourcing (“BPO”) providers are anticipating the usage of smart chat bots (based on Gen AI rather than simple rules) to drive up efficiency by almost 30% while delivering on higher quality responses. A word of caution; yes automation increases efficiency but for customer-centric companies efficiency should and rarely is the only measure or customer support success. 

It doesn’t stop at front-line support & chat bots, though. There are a host of improving solutions today thanks to AI. Think multi-lingual support, better knowledge-management, training, research & insights, reporting etc. There are platforms that purportedly reduce data extraction & formatting time. Augmented messaging will allow for more seamless hand-offs between AI and humans.  Tone detection tools help with sentiment analysis and better responses.  I also anticipate entire business processes to be re-designed internally at companies thanks to fast-developing AI solutions.

But, many models & platforms continue to have their pitfalls such as inaccurate responses and inadequate governance over information security and privacy. This is where companies will need to think about the entire industrial operations behind any AI (data labeling & annotation, quality, experimentation, tuning, risk management & information security, potential regulatory impacts, etc.). 

Let’s also talk about the real-world impact. A recent “future of jobs” report listed analytical and creative skills as the most critical ones that will continue to be a requirement in workers. On the other hand LLMs (Large Language Models) are currently missing common sense and real-world context. 

A better use of LLMs is to deploy them towards Tier 1 activities while moving skilled workforces up the customer support value chain. Re-skilling workers to focus on Tier 2 and 3, and specialized escalations, driving insights, providing quality labels to improve the underlying AI, up-sell, etc is the way to go in the future. 

The World Economic Forum’s Future of Job’s Report states: “Six in 10 workers will require training before 2027, but only half of workers are seen to have access to adequate training opportunities today.”  In practicality, GenAI has the potential to impact regions with lower access to education. 

This necessarily shifts the burden to local institutions, companies, and governments to amp up their educational efforts. Imagine a scenario where a Tier 1 customer support agent is the bread-winner for the family and is the only type of job they have ever known. I’d expect companies to utilize their skills sets for data annotation and labeling if not Tier 1 as it stands today and invest in further re-skilling them to ensure they continue to thrive in the coming decades.