When AI-powered customer service falls short, most organizations assume it’s solely a tech problem. In my experience, it almost never is.
The real issue is how contact centers were built in the first place, from the inside out, optimizing around one metric: minimizing time with human agents.
The better approach is from the outside in, starting with the outcome the customer actually needs, then building the workflows to deliver it. Done right, live agent efficiency follows naturally.
Legacy CRM and AI
The problem is that legacy CRM was never designed for that better approach. It was built to log interactions and open tickets, not to fulfill requests or drive resolutions.
Duct-taping AI on top doesn’t change that. At best, it gets customers to the same dead end faster.
Take a customer disputing a credit card charge:
• The chatbot understands exactly what they’re asking, but it restates the policy and offers no path forward.
• The customer ends up in a loop, not because the AI said no, but because they were never connected to anything that could say yes. They still have to call a live agent.
The bot didn’t just fail to add value. It actively wasted the customer’s time.
Legacy CRM promised transformation. What most organizations actually got was a digital filing cabinet: a system of record that was built to log interactions, not to drive actions.
According to ServiceNow research that surveyed more than 34,000 executives, service professionals, and customers globally, service reps spend just 45% of their time on actual customer issues.
So, what happened to most of the reps’ time? It was lost to manual work and to the overhead of disconnected systems that were never designed to work together; 80% of reps must toggle between three to five systems just to resolve a single issue.
The problem is that legacy CRM was never designed for that better approach.
Unfortunately, most companies have been deploying AI on top of legacy contact center and customer service platforms without fixing what was already broken.
But those systems had two fundamental problems. They either deflected customers without resolving their issues or when a customer did reach a live agent, that agent wasn’t empowered to resolve it because the tools, people, and systems needed to fulfill the request weren’t connected.
Automation Surfaces the Weak Links
Legacy CRM was built to log what happened, not orchestrate what happens next. When issues span departments, that gap becomes visible in painful ways.
Let’s say a customer calls about a service outage or a billing dispute, and resolving it requires work from teams across Fulfillment, Legal, Billing, and Operations. Legacy CRM can track the request. What it cannot do is route work to the right teams, trigger approvals, or ensure every task gets completed.
In such cases, human employees become the middleware, manually copying data and chasing approvals across systems. Cases fall into black holes, and no one has full visibility into status or accountability.
The data asymmetry behind this is well-documented but rarely addressed. 43% of reps cite inconsistent customer data as a top daily challenge. Yet only 28% of executives recognize it as a significant challenge.
This last datapoint is troubling. When Leadership does not see the foundation problem; instead they reach for solutions that do not address it: new dashboards, more tooling, another bolt-on.
Deploying AI on top of fragmented data, siloed applications, and broken workflows does not accelerate value. Instead, it accelerates dysfunction.
Only 34% of executives report significant progress on building a genuinely connected enterprise approach, and that number explains why so many AI investments in the contact center are underperforming.
Real resolution requires AI connected to data and workflows. The system needs to know who the customer is, what they are entitled to, and what just happened.
Let’s look at a disputed charge. Real-time data confirms the transaction, and the workflow resolves it immediately, or processes a refund, issues a replacement card, or schedules a callback, pulling in a human only when truly necessary.
Unified data, integrated applications, governed knowledge, and clean process design are now table stakes. For automation to deliver what it promises, the CRM must connect, sell, fulfill, and service on a single platform. Organizations that treat those as optional upgrades will keep hitting the same ceiling.
Human Agents Become Complex Relationship Stewards
As I noted earlier, most agents spend their days doing work that should not require a human: toggling between apps, doing data entry, and acting as middleware between departments.
Perhaps not surprisingly, only 39% of service reps say they have the tools and training needed to deliver superior customer experience (CX). The tools are failing them.
What reps actually need are AI tools that surface context and next-best actions in the moment, backed by ongoing training that builds the judgment to use them well and the confidence to know when to step in.
Duct-taping AI on top doesn’t change that. At best, it gets customers to the same dead end faster.
But as automation absorbs routine work, the interactions that remain for humans are higher-stakes and more meaningful. They require emotional intelligence, judgment, and the kind of relationship-building that AI cannot replicate.
From our research, 87% of customers say phone calls are their preferred communication channel. With that, 46% say their biggest concern with AI chatbots is their inability to fully understand their questions or concerns.
AI specialists surface context and recommended next actions so the agent can focus on the relationship, not the mechanics.
That is the shift that makes service a genuine differentiator. And when a situation calls for a human, the right platform makes that handoff with full context, so the customer never has to start over.
Predictive Service Evolves into Autonomous Resolution
The next phase of AI-powered CRM moves beyond anticipating issues to resolving them independently: before customers ever reach the contact center.
The expectation from customers is there; 53% of customers expect AI to deliver improved speed and efficiency. The question is whether the infrastructure behind the contact center can support that. But too often, it can’t.
Data siloed across disconnected systems means AI is working with an incomplete picture. Workflows that still require humans to manually pass information create bottlenecks automation can’t fix.
The result? AI that moves faster toward the wrong answer.
Predictive AI, connected systems, and automated workflows are converging to identify problems, diagnose root causes, and trigger corrective actions autonomously.
A telecom provider, for example, can detect network degradation in real time, identify the affected customer segment, and initiate remediation workflows before call volume ever spikes.
With such autonomous resolution, customers may never need to reach out at all. Organizations with the right data foundation and workflow infrastructure are already running with this today.
The contact center’s orientation shifts as a result. Instead of firefighting customer issues, teams are supervising and refining automated resolution loops. Instead of triaging incoming volume, agents are handling those cases that genuinely require human presence.
Automation Maturity Emerges as a Strategic Advantage
The organizations that are winning customers’ loyalty (and sales) are not the ones with the biggest AI budgets. They are the ones that start with the CX they want to enable.
Such companies anticipate these key questions from customers:
• What products and services do you sell?
• What do we get based on the products and services we bought?
• What requests can we make?
Successful companies design their solutions around those answers, then build the workflows to capture and fulfill every request. That outside-in clarity is what makes everything else work.
The companies pulling ahead have…started measuring whether customers actually got what they needed.
Success requires four things working together:
1. Targeted use cases with clearly defined outcomes.
2. Trusted unified data that AI can actually act on.
3. Disciplined governance that keeps automation improving rather than drifting.
4. Closed-loop workflows that continuously learn from every interaction.
None of those are technology purchases. They are organizational capabilities that have to be built deliberately.
Leaders also need to change what they measure. Activity metrics, calls handled, cases logged, and average handle time (AHT) were designed for a model where human throughput was the primary lever. But they do not capture what matters in an AI-enabled contact center.
Instead, the metrics that reflect real progress are resolution time, first contact resolution (FCR) rates, proactive issue identification, and customer satisfaction, e.g., CSAT. Organizations that continue measuring activity while deploying AI will keep optimizing for the wrong things.
The companies pulling ahead have stopped measuring how many contacts they deflect and started measuring whether customers actually got what they needed.
Those that build this maturity unlock faster resolution, higher satisfaction, and real cost efficiencies that compound over time. Those that do not will find themselves struggling to scale automation safely, with AI that surfaces problems faster than the organization can address them.
This article appears in October 2026.
