Francesca Roche sits down with Matt Graney, Chief Product Officer at Celigo, to examine why enterprise AI can produce unreliable outcomes even when it has access to huge volumes of business data.
As AI moves beyond answering questions and begins issuing credits, sending emails, and updating orders, the quality of its decisions depends on more than access to information.
Graney points to a growing focus on context across the enterprise technology market, following recent context-related launches from AWS and Databricks.
“When the biggest names in enterprise tech ship context products in the same month, I think that’s the industry admitting that the AI model was maybe never the bottleneck, but it’s actually understanding,”
For CX leaders, that distinction matters because customer information rarely sits in one place.
An order can move through a storefront, finance platform, support desk, and shipping provider, creating multiple handoffs where essential business meaning can be lost.
Celigo describes this issue as context decay, which Graney defines as “a gradual loss of meaning of data, business meaning as the information moves across systems and through time.”
That loss can be subtle, and it can worsen as customer records, policies, and business processes change.
“Even if you capture all that perfectly on day one, the business continues to evolve,” Graney says. “That means that AI’s picture and its understanding goes stale.”
The customer, meanwhile, sees only the result.
“Customers don’t experience the architecture of the businesses that serve them,” Graney says. “They just experience the consequences.”
A support chatbot may rely on data updated overnight while another system holds a more current view, creating a frustrating experience for someone who expects the business to recognize their circumstances.
Graney offers a particularly sharp example: an AI system could attempt to upsell a customer who is already frustrated by an unresolved issue.
The wider risk is not simply that AI makes a mistake, but that it can make one with apparent certainty.
“If AI can’t find the meaning, it doesn’t just stop, it will guess,” Graney says.
“That’s a pretty dangerous failure because it’s confident as it fails.”
The interview explores why CX teams should begin by mapping the real customer journey, including where data is created, transferred, and allowed to become outdated.
Graney also discusses the value of maintaining a current view of the customer and giving AI only the autonomy a task genuinely requires.
As AI takes action across the enterprise, a wrong response can become a wrongly issued refund, an incorrect shipment, or an avoidable hit to customer trust.
Watch the full interview for Graney’s practical view on how CX leaders can manage context decay before confident AI mistakes become costly customer experiences.
