Cutting Through The GPT Hype - Chatbots and Customer Service in 2024

A Brief History

The first ‘chatterbot’ was created in 1994 by Joseph Weizenbaum at MIT. He named it ELIZA. Google searches for ‘chatbot’ didn’t really show up in earnest until 2016, when chatbots found their way into customer service interactions. Searches for chatbot remained at that level until it exploded in early 2023 when ChatGPT hit the scene, with the hype causing everyone to take another look at what chatbots can do. Right now ‘chatbot’ as a search term performs roughly half as well as it did in February 2023, suggesting the AI craze popped, hit a bottom, and is now finding a sustainable position in the overall market as it’s slowly been rising since September 2023.

This chart is an easy way to get a better understanding of the market as a whole for ‘chat bots’ - They became a thing in the dot-com era, but it wasn’t clear what practical application they served. In the 2012-2016 timeframe there were many companies that created answer machine products for live chat customer service interactions, and they called them chatbots. Features and functions ranged from ‘after hours simple response robot’ to highly complex logic trees built and maintained by an army of humans. All of them were quite clunky, none of them felt very personalized, and all have generally been described as “It’s better than nothing” by most leaders in customer service.

OpenAI launched ChatGPT in November of 2022 and by early 2023 everyone (and their grandmother) had heard of it. Immediately there was a rush to market for new companies using GPT integrations, as well as legacy chatbot companies looking to enhance their existing products. Customer service as a whole was abuzz with excitement (and fear) about the possibility of AI taking over every support interaction. This unrealistic expectation has waned throughout the year as more and more people have come to understand the AI revolution for what it is (an everything-enhancement tool) and what it is not (a magic wand). 

Ticket Deflection, Ticket Resolution, and Customer Service

To better understand how AI can be used beyond the hype, it’s important to establish some ground rule terminology. If you’ve worked in customer support you know these well -

Ticket Deflection is customer self-service. It’s the cheapest tool that you can use to stop a support team from getting swamped with tickets. Generally it means building and maintaining an accurate and robust self-service help center and adding chat bots and search capability to guide customers to the correct help center article to find the answer they are looking for. For each correct answer found, you just saved yourself a ticket to solve. At the core it’s really saying “hey - help yourself and leave us alone, please”. Technically speaking, removing the ability to create a ticket in the first place is ticket deflection…you just won’t survive very long as a business that way. Deflection tools and tactics absolutely have a place in support, of course, but it can be incredibly difficult to find the fine line between “let me help you, help you” and “bleep bloop go away” or even just a never ending stream of “was this answer helpful?”. No. It was not. 

Ticket Resolution is a totally different animal. It’s generally expensive, it requires people to manage queues, manage customer expectations, and balance the customer experience with cost to Solve a ticket. Any ticket can be solved quickly, solved well, and solved cheaply - But you only get to pick two of these per ticket. This is the typical experience you have as a customer when you are emailing back and forth with a support person.

Most support orgs have some sort of tiered system - from Triaging tickets to Tier 1, Tier 2, 3, etc, which is generally following some sort of escalation path where the ticket complexity moves it to the next level. Most Tier 1 tickets are not substantially more complex or difficult to solve than a question that might have been easy enough to get deflected by our help center, but the cost for those two interactions for most orgs are dramatically different. A deflected ticket might cost us under $1, but a ticket that sneaks through into Triage might be 10-20X that. It’s no wonder why most organizations will put so much effort into deflection or even outsourcing the lower tiers to try to cut down on costs, but that typically comes at a high cost to the customer experience.

Customer Service is not ticket deflection, nor resolution. Customer service is not just solving customer problems. Those are obviously components of what many customer service teams do, but the core mission of every customer service team is to add value to the customer experience in whatever way they can. When done well, this additive effect can be more impactful and productive than Sales, Success, and even Product combined. Adding value is easy when you have low ticket volumes, customer service teams that love their jobs, a perfect product that never fails, an unlimited budget…etc.

The reality is that we find ourselves in a position of constant compromise, attempting to keep ticket backlogs low (they aren’t), employees happy (support has the highest turnover of any dept), and budgets met (“do more with less” said 2023). That compromise brings cost-saving tools like chatbots, which were invented around the same time as the Sony Playstation. The first one. And have only been iteratively updated since then, now with a shiny new GPT badge on them to clearly, concisely, and efficiently say “Please leave us alone” to anyone that shows up to interact with us. It’s not really their fault, it’s just that we’ve already trained society at large to LOATHE a chatbot. No matter how great the experience is, it’s always going to feel like a compromise. 

What about ticket resolution? Why can’t we use AI to attack this side to bring costs down, keep the customer experience high, and stun customers with a support interaction that is fast, accurate, and inexpensive? Because it’s really, really hard. Ask me how I know.

If you can inject a virtual AI assistant into every ticket interaction, you’re creating a support ecosystem where (real) people can use the robots to do the busy work for them. Let the robots run around and find answers, create a message for you, and present it to you to check and revise. Doing this with tickets instead of pre-ticket interactions means that it’s the most impactful, trusted, and expensive part of the support experience - compared to the work the chatbots do.

This should feel quite natural to any support org that does onboarding for new agents by having them do research and post internal note answers on tickets for more experienced team members to check (pro tip - have sales people onboard this way and you’ll thank me for it later).

By applying AI on this side of the fence we’re able to elevate every agent to manager as they run a robot army that focuses on finding answers to problems. This frees up the agents to focus on what we know works for everyone - Improving the customer experience and adding real value to every interaction we get. Sales and Marketing are aching for more interactions with the customer - let’s not waste them by telling them to go away the moment they land on our front door. Let’s educate them and give them a great experience and add value that will help to land and keep them for years to come.