Image created by Google Gemini
Digital customer experience (CX) is still far from seamless at many companies, despite transformation efforts that gathered speed and urgency at the start of the decade. The underlying issue now is the same as then. Data fragmentation and disconnected systems subject customers to repeated requests for the same information, misrouted calls, human service agents who can’t see a complete picture of the customer’s situation, and personalization that misses the mark.
AI has tremendous potential to solve these challenges, but before it can deliver the CX improvements companies need for better CSAT and lower churn, leaders have to improve the data architecture AI agents will use. The need for unified data is widespread. In a recent survey of 366 marketing leaders, 68% said their organizations have only partially unified or fragmented marketing, sales, customer, and analytics data.
Adding AI to fragmented systems can make existing CX problems worse. For example, 74% of consumers are still frustrated by the need to repeat their information during support contacts, according to the Zendesk CXtrends 26 report. AI agents can remember customer information so the customer doesn’t need to repeat it. But if the AI agents can’t share that information with each other and human agents, the customer may have to share the same information again to get their problem solved.
AI agents embedded in unified systems can solve the repeated-information issue. AI agents can also route customers to the right human agent when necessary, to save customers the step of making a call. And both the agentic chatbot and the human agent should be able to see customers’ product details and history for faster resolution.
Leaders often focus on the agentic AI capabilities in these scenarios, but the unified system matters as much as the AI. Without access to integrated systems and data, AI agents can fail to pass along important information to human agents, send customers to the wrong support teams, or serve information that’s incomplete or incorrect. Agentic QA and predictive analytics tools running on the same foundations now also have the dual challenge of incomplete systems access and flawed engagement data coming from customer-facing AI agents.
Now consider the speed and scale at which AI agents can work. Agents running on an incomplete data foundation can not only fail to move the needle on CX quality but also move it rapidly in the wrong direction. Proper planning before agentic AI deployment is the way to ensure that AI investment translates into better CX.
When AI for CX starts with a solid data foundation
The rush to implement AI is understandable given the pressure that many CX teams and leaders are under. But the organizations that start with good data are the ones realizing the best outcomes with AI. One of the best examples I’ve seen involved a global PC manufacturer that set out to make its vast library of product support information easier for support agents to use.
The challenge the PC maker faced is a common one in tech. As product lines grow and new versions launch, knowledge bases become too large for users to navigate easily, even when they’re structured logically. So customers contact support, but support agents also struggle with the complexity of the knowledge base. The end result is long troubleshooting sessions that leave customers frustrated and employees stressed.
This company wanted to use AI agents to help its human support agents find the right information faster to reduce resolution times and repeat contacts. The solution was to custom-design agentic bots that could navigate the knowledge base in real time to give human agents fast and accurate troubleshooting support. The bot automatically joins customer interactions with human agents to pull key information about the customer’s issue, uses that information to search the knowledge base, and displays the information it finds so the human agent can talk the customer through the steps. If the information isn’t what the customer needs, the bot refines its search strategy to retrieve more relevant information.
After deploying the AI agent in a pilot program, average handle time dropped by 58%, first-time resolutions increased by 55%, and NPS and CSAT scores rose by 18%. The company has since rolled out the agentic bots to all its contact centers.
Most organizations don’t start with a single knowledge base. Many are more like a tax management firm that wanted to use AI to drive customer upsells, cross-sells, and growth but first needed to resolve a highly fragmented data landscape. Because the firm’s customer data was scattered among 20 silos, they struggled to personalize campaigns and to evaluate campaign performance.
Unifying their data required pulling all that data into a lake for identity resolution, cleaning, and standardizing. New governance protocols ensure compliance with data privacy regulations and security best practices. An intelligence layer segments customers and develops personas for more accurate personalization and more effective cross-sell and upsell messaging. The data-driven insights drawn from the new foundation led to a 12% customer lifetime value uplift in the high-potential segment alongside a 30% reduction in marketing spend.
Planning for a successful CX AI deployment
Because the data foundation makes or breaks AI ROI, start by mapping customer data across systems before evaluating AI tools. Understand where all your customer data is, whether and how it moves between systems, and what needs to be fixed before adding AI to the mix. This can help your organization avoid the common problem of discovering data gaps after AI deployment, when fixing them is harder and more expensive.
QA agents can also make your agentic AI deployment and human support team more effective. Assessing all interactions for real-time feedback and training benefits the entire customer service organization and helps deliver more consistent CX. As you plan your agentic AI KPIs, think about how you’ll measure experience as well as efficiency.
For example, if your handle time is decreasing but your CSAT scores stay flat or drop, that’s a signal to look at how your agentic AI is handling engagements. In a case like that, you might need to reduce the number of chatbot-customer interactions in favor of more human agent-customer interactions with chatbot support. Agentic AI can be efficient, but it can’t replace the empathy and judgment that good human agents provide. That’s one reason Gartner predicts that the recent trend of laying off customer service employees in favor of AI will start to reverse by 2027.
There’s no question that AI will remake CX operations. That change is already underway. The question for organizations is whether AI will improve their CX or magnify the issues they already have. The answer depends on whether companies take the time to map and improve their data foundation before bringing in transformative AI.