Hey folks,
A common question I get from support teams is how to decide when to involve an AI agent. There’s an outdated assumption that AI agent = chatbot and therefore it can only handle simple, repetitive issues, while humans do the rest of the “hard stuff.”
But AI agents are a completely different technology than chatbots. And today’s AI agents are much better than the AI technology of even 12-18 months ago – and improving rapidly. They’re more powerful and capable of handling complex, diverse, multi-modal queries. (If you want to see what this looks like in action, we’ll be showcasing Fin’s advanced new features at Built For You on March 19.)
The lines between AI and humans are blurring, and we need a new way to think about when we use AI or humans. Here’s a framework to help you assess different queries and decide whether AI or human support is best.
1. What type of knowledge or tooling is needed?
In this new world, support teams are investing in knowledge and content more than ever. But there will still be parts of your business that lack documentation – like in the case of a new and quickly-changing product, or complex tooling that needs a lot of developer resources.
- Good for AI: Queries that require pulling information from multiple documented sources and following procedures, even if they're complex.
- Better for humans: Queries that rely on undocumented institutional knowledge, intuition developed through experience, or specialized expertise and tooling.
2. Is there a consistent problem-solving pattern?
Complex doesn't always mean unpredictable. For example, I’m filing my taxes for the first time while living abroad and while it's complex, the information and processes are well-established. It can be convoluted and intensive, but all expats and accountants are basically doing the same 50+ steps.
- Good for AI: Queries where the troubleshooting follows a consistent pattern, even if that pattern has many steps or decision points.
- Better for humans: Queries requiring creative thinking, where each instance needs a unique approach or where team members frequently need to think outside the box.
3. How emotional is the situation?
This is an interesting one – I think we’re still figuring out where and how AI can support emotional needs, and that it’s surprisingly good at offering thoughtful, balanced advice and care. But at the same time, there’s still something uniquely special about talking with another human, and moments when your customers (and your business) benefit from a genuine emotional connection.
- Good for AI: Queries where emotional needs are straightforward – for example acknowledging frustration or providing reassurance – and that follow predictable patterns.
- Better for humans: Queries where customers are really upset, need someone to build trust with them, or where the emotional tone keeps changing throughout the conversation.
4. What's the risk level if something goes wrong?
Not all queries are created equal, and some have a higher level of risk. If there’s potentially big fallout from a mistake, you probably want a human owning it, making the final judgement call, and being directly responsible for the outcome.
- Good for AI: Queries where mistakes have smaller consequences, where safeguards can be built in, or where AI can do 80% of the work and then involve a human to make the final decision.
- Better for humans: High-stakes queries where errors could significantly impact customer relationships, revenue, or compliance issues.
Start small, learn fast
It's exciting that we've reached a point where we're having to think about how AI can take over handling more queries. But this doesn't mean you don't have to make all-or-nothing decisions. Start by identifying just one or two query types that score well across these factors, and experiment.
Remember, the goal doesn’t have to be to automate everything possible – it's about creating the best customer experience while making the most of your team's valuable human skills.
The right balance will look different for every support org, but this framework can hopefully give you a starting point to find yours.
Good luck, and get in touch if you need help!
