Don't outsource your support to a bot!
At least not if you want to provide good support
My wife had a recent tech support nightmare that I’d like to relate with her permission. She was trying to sign up with a subscription-based app that was provided as a benefit via her job. More specifically, and this becomes important later, this access was provided via her job’s payroll/HR provider. So she followed the instructions, downloaded the app, and signed in with her email that she’s registered under with the payroll system. “Sorry, there’s a problem,” the app told her, helpfully. She tried a few more times with the same result.
Next, as an intelligent and motivated person might do, she checked the online documentation for this app. There were a few documents related to signing up for a subscription, but none that were relevant to this situation or that explained what that particular error message might mean. So she turned to the handy Support chat which was prominently displayed throughout this particular service’s website.
“Sure, I can help. Have you read these documents?” It spit out, you guessed it, the exact docs my wife had already read and found unhelpful. She informed the support agent of this, and it suggested one more document. This one was, to be more than fair, at least mildly related to subscriptions. But it still wasn’t helpful. My wife asked again for help, at which point it was clear she was dealing with a bot, not a person. Was there an option to speak to a person? There was not. The bot was the only support option available, at least for individual customers.
So where did that leave my wife? She was separately having trouble with logging into her payroll tool, so she couldn’t investigate the issue from that side. Her only recourse here was to speak to her office admin who could speak to the corporate benefits people, so she threw her hands up and gave up, as the stress wasn’t worth it. Ironically, this subscription was for a mindfulness/meditation program.
Just yesterday, she was finally able to regain access to her payroll portal, and figured she’d try again with the subscription app. “Sorry, there’s a problem.”
-FIN-
A few months back, I talked about gating support behind various hurdles, including bots. Now, this isn’t quite the same thing. As annoying as it is to make your customers jump through hoops to get to a real support person, at least there’s real support to be had somewhere down the line. The trend of providing support entirely via a bot, recently accelerated by the swift adoption and training of LLMs (large language models), is understandable as a cost-control measure, but inimical to providing actual support to your customers.
Where it makes sense
Let me start off by saying: I understand why this is a trend! Support engineers are expensive, and as your customer base grows, it gets more and more expensive to provide support to everyone who’s having trouble, even if you’re limiting that support to a narrower definition of ‘customer service’. If you want your customers to have access to highly technical troubleshooting, the price goes a lot higher, as we discussed a while back. So if there’s a way to fill that need without having to hire an ever-growing team of support staff, a lot of businesses are going to jump at that.
In some cases, this tradeoff makes sense. What I have in mind is B2C (business to consumer) products like many apps, where you might be selling an inexpensive product or subscription to millions of individuals. Any one of them might have problems, so are you going to staff your support for a million possible customers? It seems unlikely. Instead, you can provide something, like a chatbot or AI assistant, that’s able to resolve some commonly reported problems. This will take care of the majority of likely support issues. So what’s the problem?
A bot can’t solve all your customers’ issues
This is the problem: you’re resolving most of your customer issues, but at the cost of completely abdicating responsibility for the rest of them. Fully automated support is a calculated risk that you’ll save more money in support costs than you’ll lose in disgruntled customers. When you have a million customers, mostly paying individually, this trade-off may be worth it to your business. And in my wife’s situation, where a company is paying for a larger subscription but problems only arise on the individual level, leaving some of those individuals dissatisfied will probably not lead to losing the business customer.
In a B2B (business to business) environment, however, this calculation becomes a lot less attractive. If you’re upsetting 10% of your customers, that can lead to a disproportionate loss of business. What if the one person you’re leaving high and dry is the buyer? And if 10% of the users at one of your largest customers are unhappy, that’s a very strong signal that you’re going to run into a lot of issues when it’s time to renew the contract. While it may be impossible to keep all of your customers happy all of the time, that doesn’t mean that you can afford to stop trying entirely.
As I alluded to before, the landscape of bot-based ‘support’ has been upended recently with the advent of well-trained LLMs. By training these models on a large corpus of support issues, product documentation, and other product-related text, a company can provide quick and accurate answers to a wide variety of common support questions and troubleshooting issues. But while this broadens the range of customer issues that the automated support can address, it is a matter of difference, not of kind.
By their very nature LLMs are good at responding to situations that are like the ones they were trained on, but quickly fall apart in novel circumstances. Suppose, for instance, that a bug was introduced in last night’s customer build, and now your customers are reporting it to your support team. The LLM will have no knowledge of this—at best it will be able to point the customer to a bug report system, and at worst it may provide inaccurate or dangerous advice. If there’s no human support to fall back on, you once again end up in a situation where you can’t help a certain percentage of customers experiencing problems. While this may change in the future—this is Andy’s Support Notes, not Andy’s GenAI Predictions—LLM-based support tools are today a useful adjunct to human support expertise, not a replacement. There’s a lot to say about how these tools can help take a lot of the more repetitive work off your team’s shoulders, but that’s a discussion for another time.
Don’t do it
By now it should be clear why I’m saying that you shouldn’t trade your support team for a bot, so I won’t belabor the point. The best way to support your customers, especially when they’re relying on your product to run your business, is with a well-trained, knowledgeable team of troubleshooters. A bot (or an LLM) can’t help investigate new or unusual situations, can’t empathize with your customers who are running into issues, and put your organization at risk of losing much more revenue than you’re likely to save by using them instead of human support engineers.
Thanks for reading Andy's Support Notes 💻💥📝!