We supply software to customer support teams so we need to understand the likely impact of AI on customer support now, in 6 months, and even 2 years from now.
Below is how I've been investigating it and what I've found so far, including how customer support jobs might just increase overall, even as some larger companies appear to cut teams. Why? Software is about to enter the teenage rapid growth phase.
Here's how I've been investigating it.
I've built apps using the chatGPT API.
I've tested chatGPT answer bots on our help files to see if it can be our customer support agent.
I've attended Intercom's launch of their AI bot and spoke to their team members about how their bot will handle issues.
I've run these AI bots as part of our support process.
I've read, listened to and watched every day all the latest news on AI.
I've installed and tried every type of app.
The most important method out of the above was building apps using that API because you come face to face with how amazing it is and how flawed it can be.
Here's what I've figured out so far.
AI will answer a larger % of questions.
It's fairly obvious that AI will transform the industry and bots will answer a significant % of questions. The probability is that the % it answers will rise over time but new features and changes to products require retraining of that bot because you can't afford to let it make up stuff.
AI makes up stuff and bot providers like Intercom and Intercom's users can't allow that. That means customer support teams matter.
It's very hard to stop the AI hallucinating (making up stuff) and so Intercom, as an example, have built their bot to answer questions but redirect anything tricky to customer support teams.
When I ran my own bot and tried to 'train' it, it made up a money back guarantee we did not offer, which would have got us into deep trouble. We can't take that risk and that certainly goes for the Fortune 500.
When I ran an AI bot with live customers, it repeated that mistake and suggested to a customer they can downgrade their plan but keep the features of the previous plan. That one error was an instant $700 mistake for us, so I had to switch it off.
Maybe in time it will get better but so far, chatGPT is great for guidance but dangerous for exact answers as it's hard to control it's output for those moments.
Customer support jobs might increase because the number of apps is about to increase 100X.
Customer support jobs might increase overall. An app that took me 2 months to build, now takes me only 2 weeks.
That's going to get faster as, right now, I'm copying and pasting questions into chatGPT. In the future it will read my whole code base. It's not yet ready to write the entire app but even now it's writing 50% or more of it.
This means that the number of apps and features being built might massively increase. In addition, companies that did not have live chat might install bots and discover the need for employing someone to check on them and support them live.
Existing software platforms with a backlog of features to build are now going to get them built at a rapid pace. More features mean more training and support requests and the bots would need to learn from human support team responses before they can answer by themselves.
Consequently, predictions that this will make people unemployed might appear to be correct when some companies start reducing customer support numbers and yet 1000s more positions could open up.
Right now there is a shortage of apps because the cost of building them is so high, as developer wages are very high. If a developer who produced 1 app a year before can now produce 10, things are going to get wild.
Brand, and therefore customer support, matters more than ever now because more competitors will be built using AI.
Brand matters more now. As someone said recently, if the price of invention (the time and cost to build products) goes to zero, then the price of distribution shoots up.
If the number of apps increases 100X then all of them are going to need to market their product to get attention. Getting attention and converting visitors into customers requires good distribution and a great customer experience, which of course includes customer support. These are all part of a company's brand i.e. the customer's feeling or experience of using that product.
So anyone cutting costs on customer support and letting a bot mess it up will likely pay the price. There will be more competition to develop brands, CEOs will be judged on it. Board meetings will discuss brand, NPS scores, customer support, CX, and digital transformation even more so than they already are.
Pro-active customer support or evolution is likely.
One way to view any company is to consider it 'minimalist'. Each company is doing the absolute minimum customer support it needs to.
A customer asks for help and you go help them.
Why didn't you help them before they asked for help?
Because of resources and intelligence.
You didn't know they needed help.
Even if you did, you're not measured on pro-active help anyway so you wait until they have a problem and ask you.
With our customer support team we'd like to be able to help a customer before they ask for help. That would require using AI to see if they're struggling and drop some advice to them. We want to delight them with the answer before they've asked the question. This was tricky to do before but we think AI might just give us the opportunity to do that. We'll be building products in this direction to test out what is achievable.
What else is likely changing in customer support? Avatars, automated browsing and more.
chatGPT and generative AI in general will create amazing fun avatars, real life looking AI generated people and more. All of them will be able to speak fluently and we won't be able to tell the difference between them and real humans in most cases.
AI apps will be able to transcribe and summarise what we say perfectly.
What does this all mean?
I'd expected automated browsing and avatars or AI generated people to be guides that appear on most websites and show you what to do or even click for you.
It's entirely possible that the AI will do the complete sign up, configuration, initial setup of an app for you.
We run co-browsing software (whereby a customer support agent can see what a customer can see and even click for them remotely, without downloads) and this will ironically both expand in market size and also be disrupted at the same time, as apps increase in number and yet standardise in the latest tech they use.
All bets are off after 2 years
AI technology will improve at some exponential rate so it's hard to predict anything beyond around 2 years from now, depending on government regulation, depending on if AI companies slow down or not. The cat is out of the bag, it's open source already and that means it's a race so the only thing I can predict is that the speed of everything is about to increase, including app development.
In conclusion, it's going to be wild and the key thing I remember is that it's not a zero sum game. There's an incredible amount to build and people can collectively become wealthier together.
The world has been on slow mode so far and that's not a good thing. Low productivity is low average wealth. If it takes 1 month of working to cover all your energy bills for the year, what if it only took 1 day or 1 hour of work.
That's what I keep in mind about an AI future. I hope we all have more time for creative fun pursuits in that future because we work less to cover our basic needs. However, getting there is going to be bumpy so I want to learn quickly to be able to enjoy the ride.
Upscope's view on how important context will be in customer support
No matter what happens, you'll always have someone who will ask a question that needs a human being.
The AI is not yet set up to figure out whether Jack Smith should get his refund just because he asked or understand that Jack Smith is angry and wants to vent because he misunderstood the product he purchased.
Those situations need 'context'. We try and make our customer support as efficient as possible and we'd happily use every automation possible but having done 15,000+ support requests, I know how hard it is to automate these things and how dangerous it is to give flippant replies without context.
In order to quickly and effectively answer some people you need context on who they are, how far they've come along with your product, what they did just prior to making a support request.
This sort of context is only provided by seeing what the customer sees and that's why we're building Upscope Userview.