Inflation and economic uncertainty are pushing customer service teams to rethink how they use AI, particularly when shoppers are asking tough questions about price increases and product changes. Many AI chatbots can handle high volumes of support requests, but they often struggle with the context and nuance that price-sensitive customers need, according to Ali Fazal, CMO of Gladly, an AI-powered customer service platform.
Fazal said brands should focus less on automating every interaction and more on determining which conversations AI can handle effectively. Gladly organizes interactions around customers rather than individual support tickets, maintaining conversation history across channels alongside customer information such as purchases, loyalty status, and lifetime value.
This approach challenges a long-standing customer service priority: deflecting inquiries before they reach human agents. When customers want explanations about higher prices or product changes that affect purchasing decisions, that deflection strategy can backfire.
The lifetime value metric most teams ignore
Rather than measuring AI support primarily by how many interactions it deflects, Fazal said brands should consider its effect on the customer relationship. The metric that matters most - customer lifetime value after a support interaction - rarely gets tracked.
"The metric that matters most and gets tracked least is the lifetime value of a customer who had a support interaction. If it sits below your baseline, your service organization is destroying value, and no deflection rate is going to tell you that," he told CRM Buyer.
The push to measure customer service beyond operational efficiency extends beyond Gladly. Forrester analysts Kate Leggett and Laura Ramos see AI moving customer service toward a greater focus on customer value and revenue growth as automation takes on more routine tasks. Salesforce research cited by Forrester in May found that 85% of service decision-makers expect service to contribute a larger share of revenue this year.
AI deflection struggles with price-sensitive customers
Fazal attributes growing price sensitivity to two broader changes in consumer behavior: shoppers increasingly question what they see online, while AI tools have conditioned them to expect information almost immediately. Shoppers now ask AI services questions about product pages and get answers within seconds, and they expect the same transparency from the brand itself.
Many price complaints are actually requests for information the brand already possesses. Support systems designed around deflection may prevent that information from reaching the customer.
"They expect that same level of transparency from the brand itself," Fazal said.
Price-related inquiries often involve more than a simple transactional question. Customers may also be reacting to frustration over higher prices, changed products, or diminished value. Fazal noted that brands have traditionally responded cautiously to complaints about higher prices or quality changes, often providing limited explanation.
"That was defensible when customers had no way of knowing whether the brand was hiding something. Now they do know, and a clipped non-answer feels like an insult to their intelligence," he said.
The fix typically involves providing more substantive explanations - why prices changed, why packaging is different, why shipping takes longer. "You aren't giving up trade secrets by telling a customer that a fact they took for granted has changed and why. The situation has less to do with sentiment detection than with whether the AI can reach the reason behind the decision, which most of them can't," Fazal explained.
Beyond deflection as the primary CX metric
Customer experience performance could also include outcomes such as converting a return into an exchange or turning a sizing question into an additional purchase. About 30% to 50% of what brands classify as support volume is actually pre-purchase conversation, Fazal said - and it's measured by teams whose compensation has nothing to do with revenue.
Organizational silos can complicate that measurement. Pre-sale chat may fall under marketing while customer service reports through operations, leaving the two functions using different platforms and performance metrics. "So you get large, well-known brands running two AI bots on the same website, which is disorienting for the customer. It's also humiliating for the brand, or it should be," he said.
Teams looking to restructure their AI support strategy may find practical guidance in AI Customer Service Training for Call Center Supervisors, which covers chatbot management and workforce optimization for support leaders.
Brands face pressure to deploy generative AI quickly and cut customer service costs. Moving too fast, however, increases the risk of hallucinations, privacy problems, and responses that fail to account for the customer's circumstances. Fazal traced many AI failures to generalized systems deployed without enough industry-specific context.
"The breakage shows up on the return dispute with a loyalty exception attached to it, and that's the conversation deciding whether the customer comes back," he said.
Human oversight remains essential
Customer service AI doesn't run on autopilot. It requires ongoing monitoring as policies, products, economic conditions, and customer behavior change. A Gartner survey of 321 customer service and support leaders found that 85% are expanding human agent responsibilities as AI reduces the increases contact volume and shifts work toward higher-value tasks. Eighty percent also reported pressure to make workforce changes as AI improved agent efficiency.
Fazal cautioned brands against taking claims of "self-improving" AI at face value. Without human oversight, automated adjustments can reinforce existing errors rather than correct them. "It doubles down on the tone that's slightly off, the policy it's been misreading, the escalation rule that fires too aggressively, and every conversation makes it a little worse in the same direction," he warned.
He instead favored a controlled process in which teams evaluate conversations, identify recurring problems, and propose specific changes. Human feedback needs to be retained so the same mistakes don't repeat. Updates must happen quickly because pricing, tariffs, and other policy conditions can change well before the next scheduled model release.
Setting guardrails for AI concessions
Some price-related interactions require a discount, credit, exchange, or other financial concession. Fazal sees room to automate some financial concessions when decisions follow established policies that agents already retrieve from internal systems.
A 10% goodwill credit on a delayed order for a customer above a certain tier can be treated as a rules-based decision. AI could apply such a policy within a dollar ceiling and frequency can limit, with each concession logged for review.
"Where I'd draw the line is precedent. If approving this concession creates a new one, or the amount falls outside what the brand authorized in advance, it goes to a person. Same if the customer has already had two exceptions this quarter," Fazal said.
Automated concessions also need an audit trail showing what the system authorized, whom for whom, and under what circumstances.
Why this matters for customer support teams
For support professionals building or refining AI workflows, the takeaway is concrete: measure what happens to customer value before and after service interactions, not just how many tickets per hour the system redirects. Teams that track lifetime value after a support interaction will know - with data, not intuition - whether their AI is actually protecting the relationship.
The other operational shift is deliberate human-in-the-loop monitoring. Support managers should review AI conversations, test policy changes against real past interactions, and be able to roll back changes quickly. That requires treating AI oversight as a permanent part of the workflow, not a one-time deployment task. Professionals looking to build these skills can explore AI for Customer Support Courses, which cover helpdesk optimization and customer experience strategy.
Your membership also unlocks:
