A recent study published in the Journal of Business Research tested how customers reacted when their service recovery was handled by voice AI instead of a human agent. Customers rated the company as less customer-oriented when AI ran the recovery. The researchers were able to trace why: the AI was reducing emotional moments into the measurable signals it could act on, and customers could feel the gap.
Forrester reached a similar conclusion, predicting that even as AI takes on a larger share of the contact center workload, organizations will continue to rely on humans for interactions requiring “empathy, expertise, and nuanced problem-solving.”
That is the contradiction every COO and CX leader is now living with. The category of work where AI delivers the largest operational gains is also the category where customers can most readily detect its presence. Empathy still matters, maybe more than ever, but scaling empathy is not the same thing as automating it. The same mechanism that makes AI useful at scale, with its ability to standardize, classify, and accelerate, is what tends to make it feel inadequate in the moments customers most need to be understood.
Treating those as the same project is how a contact center ends up operationally faster and experientially worse at the same time.
The Role of AI in CX Right Now
Generative AI is summarizing calls, drafting wrap-up notes, and pulling answers out of knowledge bases in real time. Routing models are getting better at matching inquiries to the right skill or the right person. Sentiment detection helps supervisors identify the conversations that actually need their attention. Forrester’s analysts have characterized the broader workforce shift as a move from agents handling everything to a smaller agent population overseeing AI workflows and intervening on the edge cases.
These use cases reduce the cost of routine work and shorten the time agents spend on documentation. They’re making it easier for newer agents to perform like more experienced ones. None of it requires the customer to perceive AI directly, because they’re not the ones interacting with it.
Where AI Erodes Empathy
The trouble starts when AI moves from operating behind the agents to running the customer-facing layer itself. The Journal of Business Research study found that AI’s tendency to break emotional situations down into sentiment scores, escalation triggers, and sentiment-shift signals is precisely what makes the experience feel cold to the person on the other end. The customer is experiencing something emotionally complex, while the AI is reducing it into something measurable.
It also shows up in the choices customers make in real time. A recent paper in Nature Human Behaviour, gave participants a choice between an immediate AI response and waiting various lengths of time for a human. Many chose to wait. Some chose to wait specifically to have their message read by a person, even when no reply was coming back. The instinct to be heard by another person, even one who can’t respond right away, is durable enough to override the convenience advantage AI is supposed to deliver.
This intersects with a problem on the quality side that gets less attention. AI-driven quality monitoring is often pitched as a leap forward. Instead of sampling 5% of calls, you can score 100% of them in near-real time. The reality is more complicated because what gets monitored is whatever the model can measure with confidence, such as talk ratio, adherence to script, and sentiment markers. Those proxies correlate with quality on the easy calls and diverge from it on the hard ones.
A 100%-monitored contact center optimized for the wrong measures only scales the misalignment faster. The volume of monitoring goes up while the underlying quality does not.
“If You Have a Bad Process and You Automate It, You Just Speed Up the Failure”
That is how Ali Karim, VP at DATAMARK, framed the underlying issue at Customer Contact Week Orlando earlier this year. CX Today’s coverage of the session highlighted the quote in its New Rules of CX Tech, because it locates the problem precisely where it tends to live—in what the AI is being asked to operate on top of rather than in the AI itself.
It’s not uncommon these days to see:
During the CMP Research and DATAMARK session at Customer Contact Week Orlando, the audience learned that nearly 60% of organizations are now actively rewriting their scripts and knowledge bases, specifically because they understand that AI will inherit and amplify whatever is already in them.
The implication for CX and operations leaders is straightforward but often deferred: process design, content hygiene, and the metrics that govern performance are now AI-readiness work. Skipping them and going straight to deployment just accelerates the wrong things.
Where AI Actually Scales Empathy
A Harvard Business School study analyzing more than 250,000 chat conversations found that giving human agents access to AI assistance made those agents both 20% faster and more empathetic in their replies. The empathy gain was largest among less experienced agents, who used the AI’s assistance to handle the complexity they would otherwise have rushed past.
This strongly suggests that AI does not scale empathy by simulating it. It scales empathy by protecting the conditions under which human empathy can survive at scale. This happens when AI reduces the friction (lookup time, post-call admin, repetitive triage) that erodes an agent’s bandwidth to actually be present with the person they are speaking to.
This is the design principle behind the agent-assist tools DATAMARK has built into its own operating model: DataSmart for real-time knowledge retrieval and DataScribe for transcription and call summarization. Both tools give experienced agents back the time and attention required to remain empathetic.
The Strategic Empathy Question
It’s time to be brutally honest about which problems AI is well-suited to solve, and which it tends to make worse. The leaders in this space don’t treat empathy and quality as features that can be added to a contact center stack at procurement. They treat them as outcomes that depend on how the entire operation is designed and run.
In other words, restraint and forethought still matter. Vendors always move fast, and boards will continue to push for rapid AI adoption. It’s tempting to deploy AI into the customer-facing layer to demonstrate progress. But the CX programs still succeeding in three years will be the ones that use AI to protect their agents’ capacity for empathy rather than substitute for it. Programs built primarily to look like progress through automation may already be running on borrowed time.
Organizations evaluating how to scale AI without degrading customer experience should carefully assess the operational design, workflows, QA standards, and knowledge systems supporting it.
To learn how DATAMARK helps organizations build AI-supported CX operations that balance efficiency with empathy and service quality, contact our team today.
