Why the Companies Winning on Customer Experience Aren’t Choosing Between AI and Humans
Technology cycles come in waves. They arrive with fanfare and big promises, and the adoption pattern is always the same. Some applications are brilliant. Others are misapplied.
In 15 years at JDA TSG, our SVP of Hospitality Jon Joseph has watched this cycle repeat again and again. But he’s not skeptical of new technology: He just wants organizations to adopt it for the right reasons.
They shouldn’t be asking, “How much can AI replace?” The real and fundamentally different question is, “Where does human judgment change the outcome?”
Jon’s seen that when companies can’t answer that second question, they end up making irreversible decisions about their service delivery model based on hope instead of evidence. They trade service quality for efficiency gains, then have to navigate the reputational and financial fallout.
That’s the core difference between a well-designed human interaction and a poorly automated one, and it shows up in measurable client outcomes.
Here, Jon shares how successful companies decide which interactions belong to AI, which belong to humans, and what happens when organizations get that wrong.
Key takeaways:
|
The business case for AI only tells half the story
You’re probably getting pressure from above to integrate AI into customer service and workforce programs. There’s a clear argument for the business case: lower costs, faster resolution on high-volume queries, reduced headcount.
“There are a ton of organizations that couldn’t really tell you how AI should fit into what they do, but their bosses, their boards, even their Instagram feed, their kids, their TV is saying, ‘You should be doing this,’” Jon says.
But following the urge to deploy AI is only half the battle; you need to think about the deployment more specifically for your company. When you treat all customer interactions like they’re the same, for example, assuming that what works for routing a flight rebooking works equally well for explaining a health insurance claim denial, that’s an issue.
That’s the cost behind misapplied AI: the degraded service quality when AI falls short.
When organizations miscategorize high-stakes interactions as automatable, they deliver a worse experience, but more importantly, they erode the trust that client relationships are built on. And that’s not always clear until a client relationship is already at risk.
“People are going to be very quick to sacrifice high-quality, expert-driven human interactions for more aspirational technologies where they’re using hope as a strategy.” Jon says, I believe customers will not appreciate that approach and will force companies to rethink how AI is a complement to delivering great service, not the solution unto itself.
You need to be deliberate about where technology gets applied in the first place. Rather than defaulting to automation and hoping it holds up, Jon says companies need to map out where human judgment actually changes the outcome, and build their tech adoption around that.
The interaction map that matters most
There’s an important distinction to operationalize: not all customer interactions are created equal, and the dimension that matters is stakes.
There are two types of customer interaction:
- Transactional: High-volume, low-stakes queries where the customer wants a fast, accurate answer. These might include checking a flight status, rebooking a reservation, or checking an account balance. This is where AI performs well; customers are increasingly comfortable with it.
- Consequential: Lower-volume, high-stakes interactions where the emotional exchange matters as much as the information’s accuracy. These might include health insurance questions, complex financial guidance, service escalations. The customer might be nervous, or the situation might be complicated, with serious consequences attached. They would want to speak to a live expert..
Query complexity is a factor in drawing the line between tiers, but it isn’t the only one. Complexity often tracks with stakes. The two tend to rise together, but stakes are the real determining factor. A query can be simple and still be high-stakes, and that’s where automation breaks down no matter how easy the request seems to be.
AI’s best-suited role: a force multiplier for human experts, not a replacement. Jon says to let AI handle information retrieval, pattern recognition, and routing so human experts have additional time to apply the judgment, empathy, and contextual reasoning that steer a high-stakes conversation.
As a result, you’ll see more capable experts: “If resolving a customer issue takes 20 minutes today, AI as a force multiplier will allow that call next year to be concluded in 15 minutes,” Jon says. “Maybe the year after that, down to 10, where the customer walks away saying, ‘S/He fired off everything I needed to know.’”
After all, AI empathy has a ceiling. While it’s often trained to sound warm, that scripted warmth and genuine attunement aren’t the same thing. Customers in high-stakes situations can feel the difference.
“The manner in which you are led through an experience, the way it is navigated and hosted, the exchanges you have, these will result in you having different emotional responses and reactions to that brand,” Jon says.
How the trust equation is shifting
Right now, AI capabilities in natural language and emotional simulation keep improving. “AI can’t do this” is continuing to become a less common answer. Organizations need to build a human-in-the-lead model now to stay ahead of the curve.
The managed services and BPO space is entering a trust-differentiation moment. As more firms automate indiscriminately, the firms that hold on to human expertise as a key pillar of consequential interactions will rise above the rest. Client retention data will prove it.
“The potential for trust erosion in the BPO space will blow back on a lot of companies who try to substitute AI-driven solutioning ahead of the value of a really thoughtful human experience,” Jon says.
Five principles for leaders navigating AI adoption
1. Map your interactions by stakes, not by volume or complexity
Before you start automating, build a map of which interactions carry emotional or financial stakes for the customer. That map, not just cost-per-interaction, should determine where AI gets deployed and where humans stay in the loop.
2. Treat AI as a force multiplier, not a headcount target
Organizations that deploy AI most effectively use it to make their human experts faster and better-informed. They know the expert layer matters, and that model produces better outcomes and carries far less trust risk.
3. Recognize that emotional attunement is not yet automatable and plan accordingly
When AI is trained to sound empathetic, it doesn’t carry the same weight and empathy as a customer gets from a human expert who understands why they’re upset and responds to the actual situation. When trust is the product, that distinction matters; customers notice it.
4. Watch for the blowback before it becomes a client retention problem
You won’t know immediately when automating high-stakes interactions are having a negative impact. You’ll find out through client satisfaction scores, escalation rates, and renewal conversations, by which point the damage is already done.
Build the measurement framework before you need it, not after a client relationship is already at risk.
Jon points to a pattern he’s seen work in a different context: employee onboarding. If new hires are sending fewer tickets to HR or IT; fewer “I don’t know what two plus two is” types of questions, that’s a faster signal that something is working, long before performance reviews or employee attrition rates would tell you the same thing.
The reverse is also a diagnostic: a spike in those tickets means something isn’t working the way it’s supposed to, or something got automated way too soon.
The same principle applies to client trust, even though the data looks different. Jon is candid that some of this can’t be proven with clean causation; “I can’t prove it, but I know it’s true”, but that’s exactly why the tracking has to start early.
You’re looking for a correlation to emerge over months, not a single data point that confirms the problem after a client has already walked. Waiting for the renewal conversation to reveal the erosion means you’re finding out from the client instead of ahead of them.
5. The more a relationship depends on trust, the harder it is to automate
Trust is important from interaction to interaction because it compounds at the relationship level. A program built on a handful of low-stakes, transactional touchpoints can be automated aggressively with little to lose. A program where the client relationship itself runs on trust, where every interaction reinforces or erodes it, can’t be automated at the same pace without putting the whole relationship at risk.
That means automation isn’t a single decision to apply evenly across a program. The right question isn’t “how much of this program should we automate,” but “which parts of this relationship can absorb automation, and which parts are actually carrying the trust that makes the whole relationship work.”
Leaders who get this right don’t split the difference. They automate hard where the relationship can take it, and they invest disproportionately in human infrastructure where it can’t.
FAQs
How do you decide which customer interactions should go to AI versus a human agent?
Draw the line based on stakes, the emotional weight and financial cost of a wrong answer, rather than query volume or complexity. High-volume, low-stakes interactions (like checking a flight status) are well-suited to AI, while lower-volume, high-stakes interactions (like a health insurance question) require a human expert.
Does this mean AI has no role in high-stakes customer interactions?
No. AI can still support these conversations behind the scenes by handling information retrieval and pattern recognition, so the human expert has more time to focus on judgment and empathy.
What’s the risk of automating too aggressively?
Organizations risk eroding customer trust in ways that aren’t immediately visible, and the impact often only becomes clear later through declining satisfaction scores, rising escalation rates, or lost renewals.