• Skip to primary navigation
  • Skip to main content
  • Skip to footer
Metric Sherpa

Metric Sherpa

Turn Purpose Into Practice

  • Services
  • Partnerships
  • Resources
  • About
  • Contact Us
Published in
Customer ServiceNews
Justin Robbins
Founder & Principal Analyst
Share this:

The Smallest Unit in Customer Service Is the Wrong One

What Cisco’s WebexOne 2026 announcements reveal about the blind spots between customer interactions

Cisco’s WebexOne announcement this month contained a word I rarely see in a vendor press release. The company wrote that “customer experience today is constrained by treating engagements as isolated, atomic interactions”.

Atomic fits, and it describes a choice most service organizations made decades ago without noticing. An atom is the smallest unit a system recognizes. In customer operations, that unit is the interaction. Leaders forecast interactions, staff for them, score them, price them, and close them. The interaction holds the record, the survey, and the handle time. Everything that happens between interactions falls outside the system’s field of view.

Customers live in that space. The refund that needs a manager’s approval, the replacement part waiting on a warehouse, the address change that never reaches shipping: each one starts in a conversation and finishes (or stalls) somewhere the contact center can’t see. When it stalls, the customer reaches out again, and the system logs a fresh atom with no memory of the last one. Repeat contacts, surprise escalations, and churn that no survey predicted all trace back to that gap.

At the press and analyst event in Austin, Anurag Dhingra, Cisco’s senior vice president and general manager of networking and collaboration, framed the consequence on a single slide: “Execution capacity is no longer the constraint. Judgment is.” AI agents already answer questions faster than any team can staff for. The scarce resource sits in deciding what the customer needs next, who owns each piece of the work, and whether anyone followed through. Those decisions live between the atoms, and Cisco aimed its most consequential CX announcement squarely at them.

Dialog Goes to Work After Goodbye

Cisco’s answer is Dialog, which the company calls an agentic harness for the Webex AI Agent platform. In plain terms, Dialog sits between the AI agents a company deploys and the enterprise systems those agents depend on, and it keeps the customer’s situation intact as work moves among people, agents, and software. Cisco lists Dialog in beta for the first quarter of 2027. Its four core capabilities line up with four blind spots I see in nearly every service operation.

The work after the conversation

Long-horizon execution keeps an AI agent working on the customer’s behalf after the conversation ends, coordinating with people, other agents, and back-end systems until the customer reaches the outcome, according to Cisco.

Drummers learn early that the rests count. The silence between notes carries the tempo forward, and a player who loses count during a rest comes back in at the wrong moment. Customer operations treat the rest as empty space. The call ends, the case closes, and nobody keeps count until the customer calls back angry. Long-horizon execution counts through the rest. Take the replacement part. The AI agent confirms the order, then watches the warehouse pick, sees the shipment miss its carrier window, rebooks the pickup or flags a person in logistics, and gives the customer the new delivery date before she thinks to ask.

Context scattered across a dozen systems

Dhingra’s slides listed where customer and company context lives today: CRM, ERP, tickets, email, calendars, documents, calls, meetings, messages, and project tools. Each team sees its own slice and makes decisions from it. Dialog’s relationship context carries one persistent understanding of the customer across channels, teams, systems, and time by joining customer history with enterprise data.

The payoff for leaders reaches well past personalization. When every participant works from the same picture, the person who handles the follow-up sees an open commitment where today that person sees a closed case. Shared context turns a handoff into a continuation.

Lessons that leave with the shift

Every contact center has veteran agents who know when to bend a policy, which exceptions keep a customer, and which fixes actually stick. That judgment rarely makes it into the knowledge base. It leaves with the agent at the end of every shift and for good when the agent resigns. Dialog’s self-learning capability refines agent performance from customer journeys, human judgment, and exceptions while adhering to enterprise policy.

This capability carries the biggest governance question of the four. A system that learns from exceptions needs a named owner who decides which exceptions become standard practice. Leaders should settle that ownership before the first deployment and should expect any vendor to show them exactly how learned behavior surfaces for human review.

The handbook problem

Rapid onboarding lets a company bring an AI agent up to speed the way it onboards a new hire, by indexing the employee handbooks, knowledge bases, and operating procedures the enterprise already has. That approach shortens deployment, and it also exposes a truth most organizations avoid. Handbooks describe the company leadership believes it runs. Frontline teams run a different company, built from the workarounds that keep customers whole when the official process breaks.

An AI agent trained on the handbook will follow the handbook faithfully, including the steps everyone else quietly skips. Teams that audit their procedures against actual practice before onboarding an AI agent will get a better agent and a clearer view of their own operation. I consider that audit the most valuable byproduct of the entire exercise.

Watching the Agents Work

An AI agent that keeps working for days across several systems raises the stakes on oversight. Most leaders today see a containment rate and a handful of transcripts. They don’t see where AI agents stall, which customer requests trigger escalation, or how a slow back-end system degrades the answer a customer receives.

AI Agent 360 Observability with Splunk targets that blind spot. It unifies agent telemetry with enterprise data, gives teams configurable Splunk dashboards for agent health, and highlights escalation patterns and operational inefficiencies. Cisco plans general availability for the first quarter of 2027. The company also runs Webex AI Agents inside guardrails from Cisco AI Defense, with live telemetry and operational observability from Splunk.

The practical value sits in the link between AI behavior and the business systems underneath it. A dashboard that ties an escalation spike to an outage in the order management system tells an operations leader exactly where to send the engineers. Without that link, the same spike reads as an AI problem, and the team spends a week re-tuning prompts for a failure that lives somewhere else.

One Workforce, One Scorecard

The third announcement addresses a blind spot that widens every quarter AI handles more volume. Workforce engagement management tools grew up counting humans. Forecasts predicted human contacts, schedules filled human seats, and quality teams scored a small sample of human conversations. Meanwhile, AI agents absorb thousands of interactions that no forecast includes and no quality program reviews.

Cisco built its AI Workforce Engagement Management offering as an AI-first platform that rethinks quality management, forecasting, scheduling, coaching, and performance management to cover human and AI agents as one workforce.

The people still on the phones feel this shift first. When AI takes the simple contacts, the human queue fills with the complicated, emotional, and urgent ones. Handle times climb, schedules built on old averages break, and the agents doing the hardest work start to look like the weakest performers. A unified view of the workforce lets leaders see that change as it happens and staff, coach, and measure for it.

Cisco shared one example at the analyst event. Uniting NSW.ACT, one of Australia’s largest networks of social services providers with 5.8 million interactions a year, deployed a Webex AI voice agent its team calls Jeanie. Pilot clients aged sixty-six to ninety-one rated their willingness to use it 4.06 out of 5. The AI agent handled 10% of inbound calls, the ones tied to appointment cancellations, moved the operation from business-hours coverage to around-the-clock service, and saved $100,000 a year in its first use case. Those cancellations once filled a human queue. The work left for human staff changed shape, and any workforce plan that ignores Jeanie misreads it.

The Wider WebexOne Picture

Dialog, AI Agent 360, and AI WEM sit within a broader set of WebexOne announcements. Cisco introduced Claude Managed Agents in Webex, a Model Context Protocol integration that brings Webex context into OpenAI’s dots, collaborative AI agents that join spaces, meetings, and calls, and RoomOS 27 with native Google Meet support alongside Webex, Microsoft Teams, and Zoom. One thread runs through all of it, and it runs straight through the customer story: AI performs only as well as the context it receives, and most companies scatter that context across tools that never talk to each other.

What to Do Before the Beta Arrives

The questions Dialog raises apply to every vendor selling agentic customer service, and leaders can start answering them now with tools they already own.

Start with one journey that crosses at least three teams, such as a warranty claim, a disputed charge, or a service move that touches scheduling and billing. Walk it from first contact to final resolution and mark every point where a promise leaves the conversation and enters another system. Each mark shows you a place where your current metrics go dark.

Next, pull the procedures an AI agent would learn from and put them in front of your best frontline people. Ask them which steps they skip and why. Their answers will show you where the handbook and the operation disagree, and you want to close that gap before an AI agent memorizes it.

Then measure what customers care about: whether they reached the outcome, how many days it took, and how many contacts it required. Contact-level metrics will keep reporting that the first call went well, right up until the customer leaves.

Questions to ask any agentic CX vendor

  • Who in my organization approves what the AI agent learns, and where do we review it?
  • When an AI agent works a case for days, where do I see its status, and who gets an alert when it stalls?
  • How does the platform measure AI agents and human agents against the same customer outcomes?
  • What happens to customer context when the work crosses into a system your platform doesn’t own?

Counting What Happens Between

Every operation I know measures interactions with precision. Few can say how many commitments their teams made last week, how many they kept, or how long customers waited between the promise and the result. That number describes the health of a customer relationship more accurately than any handle time, and the technology to track it now sits on vendor roadmaps with dates attached.

The atom served customer service for decades because it made work easy to count. Customers judge something harder to count: whether the company finishes what it starts.

Subscribe to get stories like this in your inbox

Subscribe

Share this:
About this author
Justin Robbins
Founder & Principal Analyst
  • Twitter
  • Facebook
  • LinkedIn
  • Instagram
  • YouTube

Footer

Turn Purpose Into Practice

Proudly based in Wilmington, NC.

Contact Us:‭
hello@metricsherpa.com | ‭+1 (910) 565-9623

  • About
  • Services
  • Partnerships
  • Resources
  • Contact Us
  • LinkedIn
  • YouTube
  • Privacy Policy
  • Terms of Service
  • Disclaimer
  • Cookie Policy

Copyright ©2026 Metric Sherpa. All Rights Reserved. </>
Privacy Settings

Kalley Niebuhr
Head of Brand & Content Strategy

Kalley Niebuhr blends storytelling, social strategy, and creative leadership to help brands show up with clarity, purpose, and authenticity.

With a background in television writing, brand development, and digital content creation, Kalley has shaped impactful messaging and community-first strategies for entrepreneurs, small businesses, and educational brands.

A lifelong creative and community builder, Kalley thrives at the intersection of analytics and emotion—crafting content that connects while delivering results.

She lives in Wilmington, NC with her husband, young daughter, and two dogs. When she’s not creating, you’ll find her in the surf, running community art socials, or researching her next script.

Justin Robbins
Founder & Principal Analyst

With more than 20 years of experience, Justin Robbins has helped organizations worldwide strengthen their customer experience strategies, optimize operations, and achieve measurable results.

His expertise spans contact center operations, in-person service delivery, multimodal interaction design, quality assurance, workforce training, and global CX certification standards. Beyond operations, Justin has advised SaaS companies on content strategy, community engagement, customer marketing, and corporate communications.

As Founder and Principal Analyst at Metric Sherpa, Justin focuses on the intersection of human connection and technology in customer interactions. He is a trusted industry voice, frequently cited by the media, the author of numerous research studies, and recognized for his ability to make complex topics clear, actionable, and relevant.

When he’s not working, Justin is based in Wilmington, NC, where you’ll often find him cooking BBQ, out on the water, cheering at a game, or on adventures with his wife and four kids.