This article is sponsored by Dialpad and was written, edited, and revealed in alignment with our Emerj sponsored content guidelines. Learn extra about our thought management and content material creation providers on our Emerj Media Services page.

Enterprises are usually not struggling to deploy agentic AI as a result of the expertise is immature — they’re struggling as a result of their workflows, governance, and information infrastructure had been by no means constructed for it. MIT’s NANDA initiative found that regardless of $30–40 billion in enterprise generative AI funding, solely 5% of AI pilots attain manufacturing, with most failing as a result of brittle workflows and a scarcity of contextual studying.

Adoption has outpaced deployment self-discipline. Stanford HAI’s 2026 AI Index reports that generative AI reached 53% adoption amongst the inhabitants inside three years, quicker than PCs or the web — but most organizations nonetheless lack the workflow structure to maneuver past pilots. Governance hasn’t caught up both: the identical Index documents that hallucination charges throughout 26 prime fashions vary from 22% to 94%, whilst the share of companies with no responsible-AI coverage in place fell from 24% to 11%.

Public belief compounds the strain. Pew Research Center found that 67% of Americans have little to no confidence in the U.S. authorities to manage AI successfully, and individually reported that 71% of U.S. adults count on elevated AI use to make their private info much less safe.

Even regulated federal companies are behind. The GAO determined that greater than 25% of the IRS’s AI use circumstances lacked info on how they’d profit the company, and located no workforce plan to shut AI expertise gaps. A separate GAO evaluate concluded that the National Credit Union Administration’s mannequin danger administration steerage is proscribed in scope and doesn’t give workers or credit score unions ample element on managing AI-related mannequin danger.

Emerj just lately hosted Craig Walker, Co-Founder and CEO, Dialpad., Shezan Kazi, Head of AI Transformation and AI merchandise at Dialpad, and Shri Nandan, VP of AI Products and Experiences at Comcast to unpack the system‑stage mechanics that decide the place agentic AI creates actual operational worth — from discovery and triage to integration and controlled‑workflow execution.

This article examines three insights that make clear why agentic AI is turning into the defining functionality in trendy customer support, notably in regulated industries the place accuracy, belief, and integration decide enterprise outcomes.

• Conversation information as the supply of excessive‑worth automation: Large‑scale evaluation of historic interactions reveals repeatable, compliance‑delicate workflows the place agentic methods can ship quick ROI by changing guesswork with proof.
• AI‑led triage as the catalyst for human augmentation: Placing agentic methods at the entrance of buyer interactions removes remedial duties from human brokers, enabling quicker resolutions and better‑high quality work whereas preserving human judgment for ambiguous or excessive‑stakes circumstances.
• Integrated platforms as the antidote to fragmented CX: Unified methods remove re‑verification, repeated explanations, and damaged handoffs, permitting agentic AI, analytics, and human brokers to function in one steady loop that improves buyer satisfaction and operational effectivity.
• Workflow redesign as the unlock for vertical‑particular accuracy: Rebuilding processes so AI operates inside the dwell workflow — enriched with area‑particular fashions — produces the accuracy, empathy, and compliance required for regulated environments to undertake agentic automation at scale.

Listen to the Episodes under:

Episode:  Operationalizing Customer Service at Scale with Outcome-Driven Agentic AI – with Craig Walker of Dialpad

Guest: Craig Walker,Co-Founder and CEO, Dialpad

Episode:  Scaling Customer Experience with Operationalized Agentic AI – with Shezan Kazi of Dialpad

​Guest: Shezan Kazi, Head of AI Transformation and AI merchandise at Dialpad

Episode:  AI-Empowered Customer Service, From Hype to Scalable Operations – with Shri Nandan of Comcast

​Guest: Shri Nandan, VP of AI Products and Experiences at Comcast

Conversation Data as the Real Map of Automation Value

Enterprises start their automation journey with a assured image of the place buyer friction lives — and dialog information instantly proves them incorrect. Craig Walker argues that leaders routinely misjudge their very own workflows as a result of they depend on instinct somewhat than proof. Six months of historic interactions reveal patterns executives by no means anticipate: surprising spikes in frustration, workflows that repeat much more usually than assumed, and points that dominate quantity regardless of by no means showing on management’s “prime issues” record.

Shezan Kazi notes that this course of routinely overturns enterprise instinct. CX leaders usually request automation for password resets, flight adjustments, or lead qualification — till the information reveals these aren’t the actual drivers of quantity or friction. The highest‑impression workflows are steadily unrelated to what leaders believed mattered.

The interplay information usually reveals a very completely different actuality:  

“When we take a look at six months of previous conversations, we will see precisely which issues weren’t resolved and the place buyer frustration constantly spiked. Those patterns virtually by no means match the record of points leaders stroll in assuming are their greatest ache factors. In one case, a journey firm anticipated flight adjustments to dominate their quantity, however the information confirmed their younger prospects had been principally calling about one thing so simple as the best way to do their laundry.”

— Shezan Kazi, Head of AI Transformation and AI merchandise at Dialpad

Shri Nandan provides that enterprises underestimate the operational complexity hidden inside their interactions. Only a full-spectrum view — throughout channels, buyer varieties, and subject classes — reveals the place automation might be deployed precisely, particularly in regulated environments the place precision is non‑negotiable.

The result’s a transparent diagnostic: interplay information, not govt instinct, determines the place agentic AI can create actual operational worth. Discovery shouldn’t be a preliminary step — it’s the basis of the whole automation loop.

AI‑Led Triage as the Catalyst for Human Augmentation

Once interplay information exposes the place friction truly lives, the subsequent problem turns into executing these insights in actual time. AI‑led triage is the mechanism that turns discovery into motion — not by changing brokers, however by deciding how every interplay ought to unfold.

Shri Nandan argues that triage begins with AI taking the first move: capturing identification, detecting intent, and resolving deterministic duties instantly. For him, triage is basically a compliance safeguard — ambiguity, noise, or regulatory sensitivity should set off a handoff to a human whereas preserving full context.

Craig Walker sees a special profit in the identical construction. He frames triage as a solution to elevate human work: when AI handles verification steps and repetitive questions, brokers can give attention to complicated drawback‑fixing, empathy, and escalation administration. The human function turns into focused on nuance somewhat than routine.

Shezan Kazi introduces a 3rd angle: triage as a management system. He describes how Dialpad makes use of confidence scoring to find out whether or not the AI ought to proceed or escalate. Deterministic duties are resolved immediately; ambiguous ones path to people. The system behaves like a routing engine — not a conversational interface.

Triage solely works when the system preserves context finish‑to‑finish:  

“If the platform isn’t unified, the buyer finally ends up repeating themselves each time the interplay strikes between methods. The agent then has to rebuild context from scratch, which breaks the whole augmentation mannequin. AI can’t meaningfully assist anybody in that setting, as a result of it by no means sees the full interplay loop the method a human would.”

— Shri Nandan, VP of AI Products and Experiences at Comcast

In apply, triage unfolds in sequence: the AI captures identification and intent, resolves deterministic points, after which — when ambiguity or compliance danger arises — palms off the interplay with full context intact. This is how discovery turns into executable inside the dwell workflow.

Integrated Platforms as the Antidote to Fragmented CX

Triage can solely operate if the underlying structure preserves context. This is the place fragmentation turns into a structural failure somewhat than a tooling inconvenience. Shezan Kazi factors out that when chatbots, IVRs, CRMs, and analytics instruments function as separate methods, the interplay turns into disjointed. AI loses grounding. Agents lose context. Customers repeat themselves. Every handoff turns into a reset.

Craig Walker argues that unified platforms basically change system habits. When each a part of the workflow runs inside a single setting, the AI can observe context throughout turns, preserve grounding, and resolve whether or not to resolve or escalate. Human brokers obtain the full interplay historical past as an alternative of reconstructing it. Architecture turns into an enabler of triage, not an afterthought.

Unified structure eliminates the reset impact:  

“When you sew collectively 5 completely different methods, the buyer successfully begins over each time the dialog shifts. The AI loses grounding, the agent loses context, and the whole workflow turns into fragmented. A unified platform creates one steady loop the place each a part of the system sees the identical interplay historical past, which is the solely method AI can behave reliably.”

—Co-Founder and CEO, Dialpad.

Shri Nandan provides that built-in platforms additionally allow compliance, auditability, and constant decisioning — particularly in regulated industries the place accuracy is crucial. When AI sees the whole interplay loop, it may act with precision somewhat than guessing.

Three design rules emerge:

• Context continuity: Every system sees the identical interplay historical past.
• Unified decisioning: Routing, triage, and automation draw on a single supply of fact.
• Seamless escalation: AI palms off to people with out dropping state.

Integration isn’t a CX choice — it’s the architectural requirement that makes triage viable at scale.

Workflow Redesign as the Unlock for Vertical‑Specific Accuracy

Integrated platforms clear up continuity, however regulated industries introduce a special constraint: workflows that machines can’t but execute. Vertical accuracy comes from AI performing every step of the regulated workflow with embedded area logic — not producing responses round it.

Shri Nandan argues that industries reminiscent of healthcare, insurance coverage, and monetary providers depend on processes characterised by implicit guidelines, compliance gates, and historic determination patterns. These workflows solely make sense when executed in sequence. When AI is bolted onto legacy flows, it behaves like an observer somewhat than an operator.

Craig Walker describes how Dialpad addresses this by decomposing regulated workflows into machine‑executable items. Eligibility checks, formulary guidelines, credential validation, coverage lookups, case‑observe retrieval — every turns into a discrete motion the AI can carry out deterministically. The workflow shifts from human interpretation to step‑stage execution.

Shezan Kazi provides that area‑particular fashions behave appropriately solely when embedded immediately into this redesigned workflow. Vertical fashions educated on regulatory nuance and historic selections act like practitioners when grounded in the operational context.

Vertical accuracy requires AI to execute the workflow itself, step-by-step:  

“You get accuracy in regulated workflows solely when the AI is performing the precise operational steps somewhat than guessing at them. The area logic must be current at the precise second every motion happens, not bolted on afterward. When the workflow is rebuilt into machine‑executable items, the AI can behave with the identical precision a educated specialist would.”

— Co-Founder and CEO, Dialpad.

Regulated environments impose three constraints on agentic automation:

• Machine‑executable steps: Each motion should be deterministic and auditable.
• Domain‑grounded fashions: Regulatory nuance and terminology should be embedded at the step stage.
• Compliance gates: The workflow should embrace specific checkpoints the place AI both meets accuracy thresholds or escalates.

In this sense, agentic AI doesn’t fail as a result of fashions fall brief — it fails as a result of the enterprise by no means transformed its operations into one thing a machine can truly execute.

​