Embedding artificial intelligence (AI) in customer service operations is becoming commonplace, but the technology doesn’t seem to be lightening the workloads of customer service reps.
New research by Typewise, based on a survey of 207 frontline agents in the U.S., U.K., and Germany, found that nearly three out of four (72%) of them said AI improved their efficiency, but less than half (42%) acknowledged it reduced the time and effort spent on their jobs.
The 12-page report by the enterprise-grade AI customer service platform maintained that an “efficiency paradox” is emerging in customer service AI deployments.
While the majority of human agents report that AI improves efficiency, far fewer say it meaningfully reduces their workload, the report found.
In practice, AI shifts work rather than eliminating it, the report explained. Frontline representatives are still responsible for manually reviewing AI-generated outputs, monitoring automated actions and resolving inconsistencies between systems.
This challenge is worsened by the persistence of AI errors and limited visibility into when and why they occur. Many agents report regularly correcting AI-generated mistakes, and in some cases, only becoming aware of issues after customers respond.
“We are nowhere near ‘set it and forget it’ with agentic AI,” declared Anne DeSpain, founder of DeSpain Consulting, an AI operations and workflow automation consultancy in Park City, Utah.
“When you implement AI agents in a CX workflow — or any workflow — you still must be hands-on,” she told CRM Buyer. “You are still the driver of the car. The work shifts from executing tasks to guiding, directing, and refining the agents. There are real efficiency gains, but the time investment doesn’t disappear.”
The report added that its data suggest many organizations lack AI orchestration layers, human-in-the-loop feedback protocols, and the agentic specialization needed to handle the multitude of support requests.
“AI adoption in customer service has accelerated rapidly, but operational maturity hasn’t kept pace,” Typewise Co-founder and CEO David Eberle said in a statement. “Most teams today are not struggling with whether AI works; they’re struggling with how it works together. Without coordination, supervision, and clear ownership, AI systems can create as much complexity as they remove.”
AI Adoption Outpaces Impact
The widening gap between AI adoption and its impact inside CX teams exists because many organizations adopt AI tools faster than they redesign workflows around them, explained Anthony Miyazaki, a professor of marketing at Florida International University in Miami.
“Adding AI features to a CRM system often requires a rethinking of the processes within the system,” he told CRM Buyer. “Without integration, training, and clear objectives, AI becomes an add-on rather than a performance driver.”
This gap comes down to a fundamental misalignment between how AI is being deployed and what CX teams actually need, maintained Channing Ferrer, chief revenue officer and Americas CEO of Brevo, a global customer-engagement and marketing-automation platform.
“From where we sit, there are three core drivers of this gap,” he told CRM Buyer. “First, AI tools are being layered onto legacy workflows rather than redesigned around them. A drafting assistant here, a routing bot there. These are tactical additions, not systemic transformation.”
“Second, most deployments lack a unified data layer,” he said. “When AI can’t access a complete customer history — including purchase behavior, past interactions, email engagement, SMS opt-ins — it can only solve part of the problem.”
“Third, there’s an organizational readiness gap,” he continued. “Technology adoption outpaces the change management, training, and process redesign needed to realize its value.”
“CX AI only works when it’s connected to the full customer relationship, not just the open ticket,” he added.
Disconnected Tools Dominate CX AI
Typewise researchers found that 81% of customer service teams still operate AI as disconnected tools, rather than coordinated systems. Only one in five agents says multiple AI systems clearly work together.
“The findings reveal a clear pattern: AI is embedded but not yet harmonized in day-to-day customer service work,” the researchers wrote.
“Human customer service agents report efficiency gains, but also friction in using AI,” they continued. “They describe multiple AI systems operating simultaneously, but not necessarily in coordination.”
“They trust AI in narrow, repetitive scenarios, but hesitate when risk or judgment increases,” they added. “And critically, when AI makes mistakes, humans still own the consequences.”
“This aligns with a broader pattern observed across the AI landscape,” they noted. “Model capability has advanced faster than operational maturity. Like other industries, most AI initiatives in customer service fail today, not because the models are weak, but because architecture, orchestration, evaluation, and collaboration frameworks are incomplete.”
Illusion of Productivity
Akhil Verghese, CEO of Krazimo, an AI engineering and automation company in Dover, Del., explained that disconnected AI tools are a problem because it is rare for a customer issue to be resolved from a single source or system. “In practice, support teams usually need to combine information from multiple tools and then take action across more than one workflow,” he told CRM Buyer.
“If the AI only works in isolated pockets, it can solve fragments of a problem but not the problem itself,” he continued. “That means the human rep still has to stitch everything together manually, which limits the real impact.”
“AI can improve efficiency in a narrow sense without actually reducing human effort,” he added. “There is often an illusion of productivity because the AI is doing more, but the human is still responsible for validating outputs, checking whether actions were correct, or approving access to systems and data.”
“If every meaningful step still requires human review, the total time burden does not drop very much,” he said. “In some cases, it becomes almost as slow as having the human do the work directly.”
When tools operate in isolation with no coordination or shared logic between systems, there is no unified view of the customer across systems, which means AI cannot “see” the full customer journey, added Vincent DelGuercio, chief customer officer for Redpoint Global, a global customer data platform and engagement strategy provider.
“This leads to inconsistent or incorrect responses, which then creates even more work for CX teams because they have to switch between tools and then figure out what is missing or wrong,” he told CRM Buyer. “This disconnect prevents automation from truly working end-to-end.”
“There is no way around it,” he continued, “because without coordination, AI cannot act reliably or intelligently.”
Usage Doesn’t Equal Value
Zapier, a no-code automation platform based in San Francisco, has firsthand experience with AI tool fragmentation problems. “We began testing early AI pilots that performed great in demo environments,” explained Chief People and AI Transformation Officer Brandon Sammut. “But once those pilots had to interact inside the web of existing tools, data sources, and workflows that each team actually uses day to day, they stalled.”
“The AI part worked fine,” he told CRM Buyer. “The orchestration around it didn’t exist yet.”
“Efficiency without orchestration is just speed without throughput,” he added. “The model got faster. The workflow didn’t.”
Krazimo’s Verghese cautioned companies not to confuse usage with value. “It is easy to deploy AI widely and claim progress, but that does not mean customer outcomes are improving,” he said.
“The better measure is whether issues are actually getting resolved, whether customers are satisfied with the resolution, and whether the human team is meaningfully better off,” he argued.
“In some cases, using objective third-party evaluation is helpful because it gives a clearer picture of whether the system is creating real impact or just more activity,” he advised.
“With the right skill set, with dedicated time and with multiple layers of evaluation, CX teams can absolutely make AI work, but the goals must be crystal clear for both the humans and the agents,” added DeSpain.
“The models are good enough,” she said, “but the question is whether organizations are building the orchestration, oversight, and feedback loops to let them work together in a way that brings success.”
Loading ...
