Your AI Chat Just Answered the Wrong Question, Faster.
Priya needs an MRI. Her orthopedist is out-of-network, so before booking, she needs one number: what will her coinsurance actually cover?
She spends forty minutes on the insurer’s website, sifting through paragraphs on “out-of-network cost-sharing” that use coinsurance and copay interchangeably. Then, she checks her plan documents on the app; the FAQ tells her she’s responsible for 30% of the allowed amount. She calls to confirm, gets transferred twice, and after twenty minutes on hold, a rep quotes her 50% with no explanation for the 20-point gap.
Finally, she opens the new AI chat — the one the insurer’s leadership is proud of, that took eight months and real budget to build — and asks directly, hoping it will resolve the conflict. It answers in seconds, with another version of the same coinsurance paragraph she’d already read on the website and no mention of the numbers discrepancy.
She asks the question differently; it links her to a 100+ page plan document. One more try; the chat provides a list of in-network orthopedists instead of what her coinsurance will cover out-of-network.
Frustrated and still not sure which number to trust, Priya gives up and delays her MRI another couple of weeks. Nothing got resolved. The AI just answered the wrong questions, faster.
Why Do AI Customer Experience Investments Keep Underperforming?
Priya’s experience is extremely common and, frankly, unacceptable.
Here’s the claim worth stating up front: AI succeeds in customer experience only when someone has already found the exact place where people get lost, and most organizations haven’t done that work yet. They’ve done a different, easier-to-schedule version of it.
Priya’s insurance company, plagued by “slow service” complaints, built the new AI chat feature to provide faster customer support. But, in the pursuit of speed, they forgot the one thing that’s more important: resolution. What the insurance company really needed was someone to find where in the journey customers were already getting lost, reconcile the multiple different points already causing friction, and fix that first before selecting and deploying any technology. Instead, the team shipped the chat on schedule, the launch metrics looked great, and the same confusion and contradiction now arrive faster.
We’ve made this case before. Cortico-X’s CEO, Sujay Saha, established it earlier this year: most AI investments fail because companies start with the technology instead of starting with the experience they want to create, which includes fixing what’s broken in your journey, workflows, and content before you automate it. The data backs him up. McKinsey found that companies seeing real financial returns from AI were twice as likely to have redesigned their workflows before selecting and deploying an AI model.
Most leaders reading this will agree. But “start with the experience” raises the next question immediately: which part of the experience? Because most organizations, even the ones that take this advice seriously, end up starting in the same place: the moment where they’re asking the customer to do something. Apply. Call. Click. Pay.
The Part of the Journey Nobody’s Mapping
That’s also where most journey maps live. And it’s a narrower slice of the customer relationship than it looks like from inside a strategy meeting. There’s a deck with swim lanes, friction points flagged in red, touchpoints scored on a five-point scale.
Look closer at what actually got mapped, though. Almost every journey map covers the moments when the business needs something from the customer: click here, call this number, choose a plan, sign this form. Those are the moments that make it into a CX + AI workshop, because they’re the moments someone in the room owns.
However, Qualtrics XM Institute found that nearly 1 in 5 consumers who used AI for customer support got no benefit from it at all, a failure rate almost four times higher than other AI use cases. A journey map built around action points is a map that was never built to show where Priya is quietly giving up on getting a straight answer, and neither, it turns out, was the AI aimed at that same narrow slice of the journey.
Sujay’s argument settles where to start: with the experience, not the technology. What follows here is the next layer down, or where in that experience to look first.
This is where AI investment usually goes, because it’s where the map points. And it’s why so much AI spending changes nothing: it’s aimed at the part of the journey that was already visible, while the part that actually erodes trust stays exactly as confusing as it was before anyone touched it.
Why the Money Keeps Going In While Trust Isn’t Coming Back Out
This is compounding right now. Gartner’s most recent customer service research revealed that service and support leaders put a median of 12% of their 2025 budget into AI — the largest share given to any of the ten business functions Gartner tracked. Only 24% of those leaders report an actual financial return. Customers, meanwhile, are turning to their own AI tools roughly three times more often than the company’s own chatbot when they need help.
The mechanism underneath that failure has more to do with ownership and org charts than technology.
Most organizations fund experience improvements one business unit at a time. Marketing owns the website. Service owns the call center. Whoever sponsored the AI budget owns the chat tool. Each team solves its own piece of the journey, and each team writes its own version of the answer to “how do we explain out-of-network cost-sharing.” Most journey improvement efforts stall for exactly this reason; no function owns the experience end-to-end, so nobody is accountable for what happens at the seams between projects.
That’s the direct line from siloed funding to a chatbot that can’t answer Priya’s question – or one that answered a different question than what she asked. Nobody fragmented the AI on purpose. The content was already fragmented, because three different teams had already written three different explanations of the same coverage rules, for three different channels, funded out of three different budgets. The AI chat didn’t create that confusion. It just gave the insurance company a faster way to hand it to Priya.
Smart Communications’ 2026 customer experience research underscores just how much this matters for leaders. Nearly half of customers have to repeat themselves when they move between channels, and 63 percent say they’d switch providers over communication failures like that one. Customers don’t experience your org chart. They experience your company, failing to communicate clearly across channels you own. And they will hold it against you.
Content Design Is the Lever Nobody’s Pulling
There’s one lever in all of this that keeps getting treated as an afterthought, and it’s the one that may have actually solved the scene we opened with: the words themselves.
An AI chat is only as good as the material it’s trained to say. For example, if the underlying content is jargon dressed up as an explanation, the chat doesn’t remove the jargon. It delivers it faster and with more confidence. Before Priya’s insurer’s AI chat can do anything useful, someone has to actually break down “out-of-network coinsurance” into language a person can act on: which of the two numbers she was quoted is actually correct, what she’ll owe with her plan compared to what an in-network alternative would cost, and whether there’s a middle option nobody mentioned. That’s a content and journey ownership problem that existed before the chat tool was ever selected, and it will still exist after the next one is.
The difference between content that works and content that doesn’t isn’t subtle once you see it side by side. A cardiologist doesn’t tell a patient’s family that “the patient experienced myocardial infarction with subsequent cardiac arrest.” She says his heart stopped. Same event. One version is accurate and useless to someone who needs to act; the other is accurate and immediately understandable. “Out-of-network cost-sharing” is the coinsurance-and-copay version of “myocardial infarction” — correct information, yet no help at all to the question Priya is actually asking.
It’s not just jargon, either. Unbounce’s Conversion Benchmark Report found that copy written at a 5th-to-7th-grade reading level converts at roughly double the rate of professional-level copy, 11.1% versus 5.3%. This is the difference between what gets acted on and what gets abandoned.
Rethinking whether the content itself is useful, not just how it’s delivered, is a slower step. It’s also the only one that works.
The Question to Ask Before Your Next AI Investment
None of this argues against starting with the experience instead of the technology. It argues for being specific about where. Before the next AI investment gets approved, the harder and more useful question isn’t “what can this do?” It’s: “Where in this journey are we actually losing people, who’s accountable for what they read or hear there, and does the answer hold up whether it’s a website, a phone rep, or an AI, saying it?”
Before your next AI investment gets approved, do this first: pull the last five customer complaints about the same issue — same plan, same policy, same product question — from three different channels. Read the website copy, the call transcript, and the chatbot log side by side. If they contradict each other, or answer the wrong question, or if all three hand her the same unhelpful answer in different words, you’ve found exactly where you’re losing people, and you found it before you wrote a single line of new AI budget.
The leaders who get this right won’t be the ones who found the perfect AI use case. They’ll be the ones who went looking and can point to the exact moment, in the exact channel, where Priya gave up on getting a straight answer — and fix everything downstream to make sure whoever answers there next, human or otherwise, actually gives her one.
Recent Insights
Healthcare Needs An Early Warning System. Experience Intelligence is Where it Starts.
Picture a CFO walking into her quarterly board meeting with an experience story she's proud to tell. HCAHPS scores are up 8 points year over year. Patient satisfaction benchmarks are green across the board. The board is pleased. She's pleased. Then, on the drive back...
The Joint Commission Just Raised the Bar. Does Your Language Access Program Clear It?
This article is the second in a series on language access as a patient safety priority. Read the first article: The $70 Million Wake-Up Call: Why Language Barriers Are Your Biggest Untracked Patient Safety Risk. If you read our last article about the $70 million...
