Customer resolution sets the standard for agent-led commerce
An automated conversation can finish while the customer’s problem remains unresolved. That gap matters more when customers delegate shopping and service tasks to AI agents: software that compares products, seeks offers, or exchanges information with a brand on their behalf. For leaders responsible for customer experience (CX)—how people experience their dealings with a business, the test is whether those exchanges produce an outcome the customer can use.
Agentic commerce puts these agents to work across discovery, comparison, purchasing decisions, and support, extending AI’s marketing role beyond personalisation and audience targeting. It also introduces AI-to-AI exchanges between customer and company systems. Yet Eric Williamson, chief marketing officer at CallMiner, says “the customer is still the customer,” and those exchanges “still need to deliver an experience that works for the human on the other end.” CallMiner has a commercial stake in businesses investing in customer-experience AI and conversation intelligence, so readers should weigh its research and recommendations with that interest in mind.
Brands must be discoverable to agents before customers can choose them
Before a business can deliver that outcome, an agent must be able to find and understand its offer. To research a purchase, the agent needs information that helps it identify relevant products and compare them. Williamson argues that marketing influence will shift as agents take on this work, because the information available to an agent can determine whether a brand enters the customer’s consideration.
Williamson puts the requirement directly: “If consumers increasingly rely on AI agents to research products, compare options, find offers, or resolve service issues, brands need to make sure those agents have access to accurate information about what they offer and why it is relevant to the customer.” His recommendation connects acquisition with service. An agent seeking an offer needs dependable commercial information, while one handling a service issue needs information that helps resolve it.
For marketing teams, making that information usable becomes part of discovery. Product descriptions, pricing, and availability need to be accurate and organised so a system can identify what each detail means and which offer it describes. A price that an agent cannot reliably connect to the relevant product cannot support a dependable comparison. If the agent excludes an unclear offer, the brand’s usual customer-facing messages may never reach the person making the purchase.
To improve visibility, Williamson recommends generative engine optimisation, or GEO, which concerns how brands appear in generative AI responses, alongside search engine optimisation, or SEO, which supports visibility in conventional search. Large language models, or LLMs, generate language-based responses. His objective is that a brand’s “products and services are visible in LLMs and AI-generated search results.” These routes expand marketing’s discovery work, while product quality, the experience after purchase, and trust determine whether the choice serves the customer.
A project in mind?
Schedule a 30-minute meeting with us.
Senior experts helping you move faster across product, engineering, cloud & AI.
Customer value requires evidence beyond AI adoption
Discovery creates an opportunity to serve a customer, but businesses need outcome measures to assess what follows. CallMiner’s The Customer Experience (CX) Landscape Report 2026 describes substantial AI use across customer experience: nearly half of organisations use it for real-time agent assistance, which supports employees during customer interactions. The report also finds that 43% use AI to release frontline teams for higher-value work. Both applications change how service work gets done.
Alongside that adoption, the report finds that 62% of organisations believe they are failing to make full use of their customer-experience data. Businesses can therefore have AI operating in service while leaving much of their customer knowledge unused. CallMiner proposes a more cohesive AI strategy to connect implementation with meaningful results, warning that poorly directed automation can add friction and weaken trust. For leaders following that advice, the practical question is how to measure the customer result.
Williamson’s answer is to “measure whether the customer actually received the right outcome, not simply whether the technology completed the conversation.” A conversation’s end is easy to observe and count. Assessing whether the answer was correct and usable requires examining what happened during the exchange and what the customer had to do afterward. So the assessment must cover the customer’s whole attempt to resolve the issue.
Containment measures whether an interaction stays within automation instead of moving to a person. It shows how much work automated systems handle, but needs other measures to explain the quality of that work. Williamson says: “containment can be useful, but it does not tell the full story. Look at accuracy, resolution, repeat contacts, escalation patterns, customer effort, sentiment, and whether the underlying issue was actually solved. If the customer received the wrong answer, has to come back later, or ultimately needs a human to repair the experience, the automation did not succeed.”
Within that assessment, accuracy concerns whether the information was correct, while resolution concerns whether the underlying issue was settled. In Williamson’s wrong-answer example, keeping the exchange inside automation would record completion while hiding a failed customer outcome. Measuring each conversation separately can also hide the connection when a customer returns about the same unresolved issue. If an employee must then correct the automated answer, the eventual resolution depends on that repair.
The remaining measures help explain why resolution was difficult. Escalation patterns show where people become involved, customer effort captures how much work the customer must do, and sentiment indicates how the customer feels about the experience. A completed interaction can still leave someone dissatisfied, confused, or responsible for seeking further help. These measures help diagnose the experience while keeping the central question clear: was the issue solved?
Agent-led exchanges require the same assessment because their consequences reach the person behind the software. The immediate participant may be an agent handling a purchase or service request, but return contacts and corrective work still affect its customer. Measuring those consequences keeps unresolved needs visible even when the systems involved report that their conversation has finished.
Match automation to the service decision
A resolution-based standard gives routine automation a clear role. CallMiner describes customers as expecting immediate service through chat, email, and messaging platforms, “regardless of time or channel.” Meeting those expectations requires responsive service and consistent answers. Predictable, low-risk enquiries are therefore reasonable candidates for independent automation, with containment helping teams understand how much of that workload the system handles successfully.
More complex or sensitive situations require a different deployment choice because empathy, understanding, and judgment can affect the outcome. AI can assist an employee while the person remains responsible for the interaction. Planned employee involvement is part of the service design; corrective work after an inaccurate automated answer is a recovery task. Measuring these separately helps leaders judge both the original deployment decision and the quality of the eventual service.
CallMiner’s research shows why leaders are concerned about those decisions: 91% of organisations express concern about implementing AI in customer experience. Among respondents, 52% worry about AI communicating misinformation to customers, while 47% are apprehensive about security and compliance risks. These concerns span AI use in CX. They make deployment boundaries a practical management priority alongside speed and capacity.
Customer-side agents add the question of whose instructions a system is carrying out. Williamson warns that “the first interaction may not always be directly with the customer.” A business may need to respond first to the customer’s software representative, so it must consider whether that agent is legitimate and faithfully conveys the person’s intent. As agents participate in more stages, authentication, checking who or what is making a request, becomes more closely connected to accountability, security, misinformation, and decisions about human involvement.
Those questions become more pressing as customer behaviour changes with improving reliability. CallMiner’s report says the future of CX “will be shaped by organisations that apply AI with discipline and intent,” making deliberate limits part of preparation. Technical ability establishes what a system can automate; service policy determines which requests it should handle independently. That policy must also specify when an employee should become involved to understand the request or resolve it responsibly.
Customer information should improve business decisions
Putting those deployment choices into practice requires dependable information across the business. Product information helps an agent discover an offer, while customer information supports decisions during service. Williamson’s preparation advice begins with that dependency: “the first step is making sure the information these systems will rely on is accurate, consistent, and accessible. Most companies have customer information spread across several different platforms, and AI is only as effective as the knowledge it can actually access.”
Information spread across platforms can leave the system handling an interaction without knowledge the organisation already holds. Teams therefore need to establish which information is available when a service decision is made and whether it is reliable and consistent. This gives data management a direct operational purpose: enabling the system or employee to respond using relevant customer knowledge.
Once that knowledge is accessible, customer conversations can inform decisions beyond the immediate reply. AI-powered conversation intelligence uses AI to analyse interactions and identify patterns people can act on. CallMiner’s report describes its application to employee development, where conversation analysis can reveal strengths and areas for improvement. Service leaders can use those findings to support staff alongside their decisions about which enquiries AI should handle independently.
The same analysis can help marketing teams identify customer needs and potential opportunities. CallMiner calls this augmented decision support: analysis that informs people’s choices, with the aim of turning “raw conversation data into actionable insight”. Its report proposes using those insights for “smarter decisions across product, marketing, operations, and compliance, strengthening customer loyalty while reducing reputational risk.” Achieving those benefits requires the relevant teams to turn the findings into changes in their work.
To make informed choices about those changes, marketing leaders also need direct experience of AI tools. CallMiner urges them to use the technology themselves so they understand its capabilities and shortcomings. Alongside knowledge of their customer data, that experience gives leaders a practical basis for deciding which tasks to automate, where employees need assistance, and whether the resulting exchanges improve their understanding of customers.
Key highlights
- Make offers readable to agents: Marketing and product teams need accurate, clearly organised product descriptions, prices, and availability so shopping agents can compare offers reliably. Include AI-generated search visibility alongside conventional SEO in discovery planning.
- Measure customer resolution: CX leaders need accuracy, repeat contacts, escalation patterns, customer effort, and sentiment alongside containment metrics. Connect interactions across the customer’s service journey to identify unresolved issues and employee repair work.
- Set clear automation boundaries: Service leaders can assign predictable, low-risk enquiries to independent automation and define when employees take responsibility. Customer-side agents also require checks on identity and authority to act on a person’s behalf.
- Connect customer information to decisions: Data owners need to make customer knowledge accurate, consistent, and accessible across service systems. Product, marketing, operations, and compliance teams can then use conversation analysis to identify recurring problems and prioritise changes.
A project in mind?
Schedule a 30-minute meeting with us.
Senior experts helping you move faster across product, engineering, cloud & AI.


