There’s a strange contradiction in how customers feel about AI. On one hand, they’re getting used to it—nearly half (48%) of consumers say AI has made customer service experiences more helpful. But they’re not ready to give up on humans entirely. 51% of people prefer automated agents for quick answers, but a solid 67% still want a human when the question is complicated or the situation calls for empathy.
This lands you in quite the balancing act: using AI to make things faster, but not at the expense of the human touch customers still value. The goal isn’t to trick anyone into thinking AI is human—it’s to create AI that works alongside your team, with humanity and empathy at its core.
1. Not all AI is created equally human
Customers get frustrated when they can tell an automated agent clearly doesn’t understand what they’re trying to say. They know a human could solve their issue in seconds, so they find themselves shouting “Speak with a representative!” at their phone, desperate to escape the automated menu. This frustration isn’t new; it’s the same problem decision-tree-based bots have always had. They’re rigid, easily confused, and leave customers feeling unheard.
The first step to avoiding this experience is using the right AI. Systems like RAG-based models simply retrieve and spit out information from your knowledge base and can feel just as clunky as decision trees. These systems don’t actually understand the customer’s request—they just know how to deliver a pre-existing answer. When the issue doesn’t match perfectly, the response falls apart.
On the other hand, more advanced automated agents use natural language understanding (NLU) and natural language processing (NLP) to analyze and interpret a customer’s intent. Instead of regurgitating information, they can understand the nuance behind a request and respond in a much more human way. Systems powered by agentic AI go even further, understanding the request and taking action, much like a human would.
Take wholesale retail distributor Motel, for example. Their system requires customers to navigate a strict self-service menu. If the solution doesn’t fit neatly into one of the options, the customer has no choice but to open a ticket and wait for a human to help (kind of like those frustrating phone menus).
Start by picking a tool that can approach a system with humanity—one that combines NLP, NLU, and agentic AI features. Our flagship solutions, Forethought Solve and Autoflows for Solve, don’t just process information; they understand, reason, and act on behalf of the customer.
Once you have the right system in place, you can focus on deciding what tasks the AI will handle and make sure it handles them with the care and precision customers expect.
2. Draw the line between AI and humans with care
Before you dive into implementing or fine-tuning your AI, take a step back and ask a critical question: What’s the best way for us to approach our customers with empathy and humanity? The answer isn’t the same for every business, but it usually comes down to balance—letting AI handle what it’s good at and leaving the rest to humans.
For some companies, that means using AI for quick, repetitive tasks like resetting passwords, processing returns, or answering account questions. Customers want these kinds of issues solved quickly, and AI can deliver. But humans are still the best option for complex or sensitive situations that call for emotional understanding.
At Forma, a benefits platform, they take empathy in customer interactions seriously and have seen firsthand how AI can elevate their team’s work:
“The things you have to do 30 times a day will go away,” Kara, a Forma leader, told her team. “You’re here because I want you to spend time on what AI can’t do.”
While there was initial hesitation, the team quickly realized how much AI helped them focus on customers who truly needed their attention, making their work more impactful and rewarding. One way they do this is by always displaying the option for customers to connect directly with a person. There’s never a sense that they have to fight to speak with a human.
You could do the same or configure your AI agent so this happens automatically in specific scenarios. Perhaps you want AI to take the first crack at every request but seamlessly hand it off to a human if the customer isn’t feeling satisfied.
Here’s what a conditional split looks like in Forethought’s platform. In this case, the AI agent will provide a different response based on whether or not a customer has viewed a specific article.
Once your division of labor is defined, the next step is to make sure AI does its part thoughtfully and empathetically. AI isn’t just a tool for efficiency—it’s an extension of your team.
3. Train your AI to respond as thoughtfully as your team
For AI to work with empathy, it has to have the same information as your human team. Clean, up-to-date knowledge bases, historical tickets, and strong integrations with tools that hold contextual data are essential. But good information alone isn’t enough—your AI must also be trained to respond well.
Customers write into chatbots with intention, and AI has to understand what they’re asking to respond thoughtfully, even when the phrasing isn’t perfect. Unlike the old decision-tree bots, where you had to predict and hard-code every scenario, modern AI doesn’t need that level of rigidity. But, guiding it with a list of main intents can help it succeed right from the start.
Intents allow AI to interpret a customer’s language and deliver accurate responses. For example, if someone says, “I can’t log in,” the AI should connect that to an intent like “Forgot Password” and guide the user through resetting their account. Customers of Forethought can customize intents and training phrases to help AI catch customer language that might not be obvious.
This becomes even more important for complex situations. Take a patient scheduling system, for example. A customer might not say, “I want to change my appointment,” directly. Instead, they might ask, “Can I move my doctor’s visit?” or “I won’t make it today—can we find another time?” Training the AI with variations of these phrases can help it confidently handle the request, even if the customer doesn’t use the exact words the system expects.
Forethought customers can also use instructional notes, which are guidelines that provide context or rules to help an AI agent deliver clear, accurate, and consistent responses. For example, an investment platform might not allow automated trades for cryptocurrencies, so they create an instructional note to clarify this when customers ask automation-related questions: “Let customers know that automated trades are not available for cryptocurrencies and guide them to place a manual order through the trading dashboard.”
When combined, intents and instructional notes give AI agents the proper context to sound more human. Regularly review your chatbot’s chat insights and transcripts. Look for patterns where the AI misunderstands or struggles, then add new intents or training phrases to close the gap.
4. Control the conversation with brand and tone of voice guidelines
Whether your tone is formal and professional, like a research think tank focused on clarity, or casual and upbeat, like a company that hosts team-building adventures, your AI should adapt to sound like you if it will provide a human-like experience for customers.
Forethought lets you quickly set this voice using our Tone of Voice widget. Simply describe the tone you want, and you’ll get examples to confirm the output matches your style.
If you’re not sure where to start, don’t worry. You can choose from suggestions like “friendly,” “professional,” or “straightforward” to get your tone just right. Beyond words, visuals matter, too. Customizing the chatbot’s design to align with your brand—as Uncommon Goods did—ensures the experience feels cohesive, polished, and intentional.
These small adjustments aren’t just cosmetic; they make customers feel seen and respected. When your AI sounds and looks like an extension of your team, it communicates care, builds trust, and reinforces that their questions and experiences matter to your brand.
5. Empower AI to act, not just answer
If you called your cable company and spoke to a human, you’d expect them to actually solve your problem—fix the bill, schedule a repair, or make whatever change you need. Interactions with AI feel more human when it doesn’t just understand a request but takes action to resolve it, just like a human would.
For that, you need the most advanced form of AI: agentic AI. Unlike generative AI, which uses NLU and NLP to comprehend and communicate, agentic AI goes further—it reasons, makes decisions, and acts on those decisions. Generative AI can help you understand why your bill is wrong, but agentic AI can fix it. Without this capability, systems often pass customers to humans often.
Forethought’s Autoflows for Solve is one of the few solutions powered by agentic AI. It works within Solve, combining NLU, NLP, and generative AI with the ability to take action across your systems. The example below shows how the system would retrieve an order ID from an ERP system and process a return.
Gather uses Autoflows to automatically provide customers with quotes, turning a simple inquiry into an immediate resolution without any human intervention required.
At its best, agentic AI bridges the gap between understanding and action, giving customers the speed and results they expect.
The future of AI in CX is a partnership, not a replacement
While AI continues to evolve, delivering faster, smarter, and more human-like interactions, its true value doesn’t lie in replacing humans. It lies in amplifying what humans do best.
By carefully designing AI to act thoughtfully and empathetically, businesses can give their teams the freedom to focus on building connections, solving complex problems, and delivering experiences that AI can’t replicate. The future of customer experience isn’t about making AI more human—it’s about creating a partnership where humans and AI, together, make the customer experience truly exceptional.
