Most discussions about AI in customer interactions revolve around one question: How much can we automate? Thirty percent? Seventy? Everything? That’s the wrong question. Automation isn’t a knob you can turn up or down. The real question is: Which interaction warrants which approach?
Expectations are rising, but trust isn’t keeping pace
Customers no longer compare you just to your direct competitors, but also to the most recent seamless experience they’ve had anywhere—with Spotify, Uber, Booking, and so on. AI is driving those expectations even higher. But trust isn’t growing at the same pace. 73% of consumers use AI daily, and yet only 29% believe that organisations use it responsibly.
You don’t earn that trust with better algorithms. You earn it by doing what you promise, especially when the going gets tough. And that starts with one choice: where do you let AI do the work, and where do you keep people in the picture?
Two questions that determine the choice
How complex is the interaction? High volume, a fixed process, and few exceptions make a request simple. Many systems, rules, and customization make it complex.
How significant is the moment for the customer? This is the interaction’s “trust sensitivity.” It’s not about how difficult the problem is, but about how much is at stake for the customer. Is money involved? Emotion? Time pressure? Can tone and timing make or break trust?
Combine those two questions, and you get four types of interactions. Each type requires a different approach.
Four types, four approaches
1. Simple, low trust sensitivity: fully automate
This is the world of routine and repetition. Tracking a package, resetting a password, requesting an invoice, or changing an address. The customer doesn’t want a conversation here—they want speed. Don’t slow me down, don’t make me repeat myself—just fix it. That’s exactly why AI excels here. It handles the request from start to finish, day and night, in just a few seconds. A good chatbot that immediately displays the correct package status often does more for the customer than an employee who gives the same answer five minutes later.
2. Simple, high trust sensitivity: AI prepares, people connect
Sometimes the problem is simple, but the moment isn’t. Think of an order that arrived damaged just before a birthday, or an incorrect invoice that makes the customer feel like they’re paying too much. The question then isn’t just whether you can solve it, but whether you understand what it means to them. AI can do a lot of the groundwork here: it gathers the context, summarizes what happened earlier, and immediately gives the employee the full picture. But the real connection comes from a human. Because in these moments, you build trust not with speed, but with acknowledgment.
3. Complex, low trust sensitivity: AI supports, people decide
Here, the challenge lies not in emotion, but in complexity. Think of a customer comparing different insurance policies, or a company that needs a customized quote. Many variables, a lot of information, and many interrelated factors. In this case, AI is a strong co-pilot. It organizes the information, brings the data together, and highlights what a human might easily overlook. But it doesn’t take over the decision-making. The employee maintains the big picture and makes the final choice, with AI as an assistant—not a boss.
4. Complex, high trust sensitivity: humans lead, AI assists at most
This is the most delicate category: both complex and important. Think of a suspected case of fraud, a major financial loss, or a sensitive case following a death. For the customer, everything is at stake, and what matters most here is not speed but presence. Does the customer feel taken seriously, safe, and understood? AI may help in the background, but it must never dictate the conversation. The wrong tone can cause more damage here than a slow response, and once trust is broken, it’s hard to repair. So always keep this in mind: every answer your AI gives is an answer you give.
Getting Started in Three Steps
This framework isn’t a theoretical model, but a practical tool. You don’t have to roll it out across your entire organisation all at once: three steps will get you a long way.
- Map out your reality. Don’t start with your processes, but with the real reasons why customers reach out.
- Plot each interaction on the two axes: complexity and trust sensitivity. A rough estimate is sufficient.
- Turn that into action steps. Don’t try to doeverything at once. Choose one interaction and start with it this week.
Start with the interaction, not the technology
The companies that make a difference aren’t the ones with the most advanced AI. They’re the ones that consciously choose where AI belongs and where humans remain indispensable. So: don’t start with the technology, but with the interaction. And design from there.
