Authored by Victor Martino | Executive Vice President, Services & Strategy, Pinnacle Technology Solutions
An AI agent completes its workflow. Every API call succeeds. The system records the task as resolved.
The customer still needs help.
That gap between technical completion and customer success deserves far more attention as organizations develop agentic AI.
Development conversations can quickly center on models, tools, integrations, and the actions an agent can perform. Those decisions matter. But the customer’s experience must shape those decisions from the beginning.
What does the customer actually need? How much effort will the process require? Will they understand what is happening? Can they correct the agent or reach someone who can help?
When the workflow ends, will their problem truly be resolved?
Customer experience is the ultimate test of delivered customer value. Accuracy, security, and responsible operation are essential requirements for achieving it.
From my perspective, this is a shared responsibility for organizations deploying AI and the technology partners helping them.
When development moves quickly without that foundation, teams risk building around an incomplete definition of success. Customer feedback then exposes missing context, confusing decisions, unnecessary steps, or inadequate escalation paths. The workflow returns for revision because the experience was never fully designed.
Drawing on established principles of human-centered AI, the following 10 Customer Experience Imperatives offer a practical framework for designing and evaluating agentic workflows.
1. Define success in the customer’s terms.
Every use case should begin with a clear statement of what the customer needs to accomplish and how resolution will be verified.
“Close the support ticket” describes an internal action. “Restore the customer’s ability to work and confirm that the issue is resolved” describes a customer outcome.
Document the practical result, acceptable effort, and information the customer needs when the interaction ends.
Design imperative: Establish customer-defined acceptance criteria before selecting the technology.
2. Design the complete customer journey.
An agent may perform one task within a larger experience. The customer experiences the entire journey across systems, channels, departments, and people.
A well-executed step can still create frustration if the next team lacks context, a confirmation never arrives, or the customer must start over.
Map the journey from the initial need through resolution and follow-up. Identify waiting points, repeated requests, conflicting instructions, and gaps in ownership.
Design imperative: Make every handoff preserve context, accountability, and progress toward resolution.
3. Understand intent and context before taking action.
The same request can represent very different needs. “Replace this device” might describe a routine refresh or an urgent failure affecting a critical business function. The appropriate response depends on the circumstances.
Use relevant, authorized information to establish intent, urgency, and constraints. Ask focused questions when uncertainty could materially change the outcome, and allow customers to correct assumptions.
Design imperative: Define what the agent must understand before each consequential action.
4. Reduce the customer’s effort throughout the workflow.
Automation should make the customer’s task easier. Repeated explanations, unnecessary
confirmations, and multiple transfers can create more work than they remove. Internal efficiency gains lose value when the customer absorbs the effort.
Reuse information appropriately, simplify choices, and make the next step clear. Include the work customers must do to recover from errors when evaluating the experience.
Design imperative: Measure total customer effort alongside processing time and operating cost.
5. Make actions, progress, and limitations understandable.
Customers should know they are interacting with an AI agent, what it can do, and when a person becomes involved.
Distinguish between an action that has been requested, one that is underway, and one whose result has been verified. Communicate delays and uncertainty honestly.
Use plain language and provide the information customers need to understand their options.
Design imperative: Make status, limitations, and next steps clear at meaningful points in the journey.
6. Match autonomy to authorization and consequences.
The appropriate level of autonomy depends on the action, its impact, and the authority granted.
Retrieving information may require little intervention. Making a purchase, changing a service, or modifying a production environment may require explicit approval and stronger controls.
Define what the agent can do independently, what requires confirmation, and what must be escalated. Customers and authorized operators need practical ways to pause or correct its activity.
Design imperative: Establish and enforce clear boundaries around autonomous action.
7. Make human assistance dependable.
Human support belongs in the design from the start. An agent should recognize when a request exceeds its authority, repeated attempts are failing, or the customer wants help from a person.
A useful handoff includes the customer’s objective, relevant history, attempted actions, and unresolved questions. It also identifies who owns the next step and sets an honest expectation about response time.
Design imperative: Test the human handoff as carefully as the automated path.
8. Design for failure, recovery, and accountability.
Failures need a defined customer experience. A system may be unavailable. Information may be incomplete. An action may partially succeed. Repeating it automatically could create a duplicate transaction or another unintended consequence.
Detect those conditions, stop unproductive loops, explain the situation, and recover safely.
Establish who is responsible for correcting the result and communicating with the customer.
Design imperative: Give every material failure scenario a recovery path and an accountable owner.
9. Validate with the customers the workflow must serve.
Customer participation should influence development while meaningful design choices remain open.
Test with people representing different abilities, language needs, levels of technical confidence, and operating conditions. Include incomplete requests, urgent situations, changes of mind, and exceptions.
Evaluate whether customers can understand the interaction, achieve their objectives, and obtain help when needed. Look for differences in performance across customer groups.
Design imperative: Require representative customer testing and evidence of an accessible, fair experience before release.
10. Measure customer outcomes and improve continuously.
A dashboard can show high automation rates while customers repeatedly return with unresolved issues.
Measure verified resolution, repeat contacts, customer effort, satisfaction, and escalation quality alongside accuracy, reliability, and cost. Investigate abandonment and silence before interpreting either as success.
Assign ownership for reviewing feedback and correcting recurring problems. Reassess performance when models, tools, policies, or customer needs change.
Design imperative: Make customer experience a continuing operating responsibility with measurable outcomes.
Safeguards that apply to every imperative
Privacy, security, and fairness must influence every design decision.
Use only the customer information needed for the task, enforce access permissions, and establish appropriate retention practices. Control which tools and actions the agent can access.
Maintain sufficient records to investigate decisions and correct mistakes.
Test whether the experience works equitably across the customers it serves, including people who need accessible interfaces or alternative support channels.
A positive interaction cannot compensate for an unauthorized action, inappropriate data use, or an unreliable result. These requirements must be built into the workflow and its release criteria.
Turn the principles into operating decisions
For each imperative, document the design decision, the accountable owner, the evidence that demonstrates it works, and any remaining gap that must be addressed before release.
Review those decisions when selecting a use case, approving its design, authorizing deployment, and expanding its autonomy. This makes customer experience part of the implementation process and gives business and technical teams a shared basis for evaluating readiness.
Demonstrate the value through real deployments
The framework’s value should be demonstrated through measured results.
Begin with a defined use case and establish a baseline for the current experience. Agree on success measures before launching a limited pilot. Compare verified resolution, repeat contacts, customer effort, escalation quality, and unintended actions against that baseline.
Include customer feedback and assess performance across relevant customer groups. Set conditions for expanding, revising, or pausing the workflow.
Publish findings with their context and limitations. An improvement in one setting is evidence to learn from; broader claims require broader testing.
Consider an agent handling an equipment failure. Submitting a replacement order is one milestone. The customer’s experience also depends on whether the agent understood the urgency, selected an appropriate replacement, secured authorization, communicated delivery expectations, and provided help if delivery failed.
That broader definition changes the workflow itself. It determines which information matters, where approvals belong, when a person becomes involved, and what the organization measures.
For business leaders and technology partners, this is a leadership responsibility. Customer-facing teams, service owners, designers, security specialists, and engineers need to establish the experience together.
The discipline begins with one question:
What must the customer experience for this outcome to be considered successful?
Answer that first. Let it guide the workflow, the technology, and the evidence required to demonstrate success.
Every agent acting on behalf of an organization becomes part of its relationship with the customer.
Build the agent to earn the customer’s trust, and measure success through the experience it delivers.
Agentic AI presents tremendous opportunities to improve efficiency, simplify operations, and deliver better customer experiences. But realizing that potential requires more than deploying new technology. It takes the right strategy, infrastructure, security, and expertise.
At Network Solutions, we help organizations navigate emerging technologies and turn innovation into measurable business outcomes. Whether you're exploring AI opportunities or preparing to implement agentic workflows, our team can help you take the next step with confidence.
Let's talk about what AI can do for your business. Complete the form below to start the conversation.
Victor Martino serves as Managing Director of Network Solutions and Executive Vice President of Services and Strategy at Pinnacle Technology Solutions. With over 20 years of technology leadership, he specializes in strategic partnerships, business growth, technology innovation, and customer-focused solutions.