But for customer experience (CX) leaders, the bigger question isn’t simply what Voice AI can do. It’s where the technology creates meaningful value for customers and the business, and how leaders can prove it. 

That question was at the center of our latest Virtual Executive Roundtable, where CX leaders shared what they’re learning as they test, implement, and evolve Voice AI strategies. While approaches varied, several themes emerged around measurement, readiness, use cases, and the continued importance of the human experience. 

Deflection Isn’t the Finish Line 

For years, contact center leaders have looked at deflection and containment as indicators of automation success. But the conversation is shifting. 

Several leaders pointed to metrics like end-to-end resolution, customer effort, CSAT, cost per contact, assisted conversion, and recontact rate as more meaningful measures of whether Voice AI is delivering value. 

The distinction matters. An interaction that never reaches a human isn’t necessarily successful. If the customer has to call back, struggles to get an answer, or abandons a purchase, a high containment rate can disguise a poor experience. 

The more useful question is: Did the customer get what they needed, and did the interaction create value for the business? 

For CX leaders, that means building measurement strategies around outcomes, not simply how many conversations AI handles. 

Voice AI Is Moving into More Complex Territory 

The capabilities being discussed today extend well beyond the straightforward transactions that initially made Voice AI attractive. 

Leaders shared examples of AI evolving from FAQs and transfers into reservations, ordering, and other customer-facing transactions. Others pointed to an emerging ability to handle more ambiguous requests, frustrated customers, and interactions that traditionally were considered too nuanced for automation. 

That evolution creates a new strategic challenge. 

As AI becomes capable of handling more complexity, organizations need to be increasingly deliberate about where AI should be used, and where it shouldn’t. 

The goal isn’t to automate as much of the customer journey as possible. It’s to determine where technology can reduce effort, improve outcomes, and create a better experience. 

AI Readiness Starts Before the AI 

Some of the biggest barriers discussed weren’t about the AI’s sophistication. They were about everything surrounding it. 

Knowledge source cleanup, access to backend technology, and connections to systems of truth can all become hold points. An AI system can only provide a reliable experience if it has access to reliable information and the systems needed to take action. 

That makes Voice AI as much an organizational readiness challenge as a technology initiative. 

Before scaling an implementation, CX leaders were urged to look closely at the fundamentals: Is our knowledge accurate? Can AI access the information it needs? Can it complete the actions customers are asking for? Are our systems connected well enough to support the experience we’re promising? 

The Right AI Depends on the Customer Need 

Another important theme was that Voice AI shouldn’t automatically be the answer simply because it’s available. 

For straightforward transactions, Voice AI may create meaningful value. For more complex interactions, agent-facing AI may be more effective at helping employees find information faster, improve accuracy, and resolve issues while preserving the human connection. 

This points to a broader principle for CX leadership: Start with the customer need, then determine where AI belongs. 

Sometimes that means a virtual agent or AI supporting an employee. And sometimes the best experience still requires a person. 

The most mature AI strategies won’t necessarily be the ones with the most automation. They’ll be the ones that make the clearest decisions about where technology can improve the experience. 

The Human Impact Matters 

Even as organizations pursue efficiency, the workforce implications of AI remain an important part of the conversation. 

Leaders discussed the importance of thinking beyond replacing roles and considering how employees can be retrained and prepared for new responsibilities as AI becomes part of the operating model. 

That requires CX leaders to think about AI transformation as a people strategy, too. 

The strongest implementations may ultimately combine technology’s ability to increase knowledge, efficiency, and personalization with the judgment, empathy, and expertise of human employees. 

The Conversation is Still Evolving 

Perhaps the clearest takeaway from the roundtable was that Voice AI is not a set-it-and-forget-it initiative. Leaders are still testing, learning, measuring, and refining their approaches. 

And that’s exactly why peer conversations matter. 

As Voice AI continues to evolve, CX leaders can learn not only from what works, but also from the questions, challenges, and lessons emerging across the industry. 

Want to stay part of the conversation? Make sure you’re on the Execs In The Know email list to receive updates on upcoming Virtual Executive Roundtables. 

A special thank you to our business partner, Vapi, for helping bring this conversation about the future of Voice AI to our CX community. We appreciate their partnership in creating space for meaningful conversations, shared insights, and continued learning across the industry. 

The post What CX Leaders Are Learning About Voice AI appeared first on Execs In The Know.

Your Performance Problem is a Leadership Problem

Two numbers every executive should sit with: 20% of job knowledge comes from formal training, and employees spend 1% of their workweek on skill development. If that’s the system you’ve funded, you didn’t buy capability — you bought a compliance event. That’s a design decision, and design decisions belong to leadership.

The Problem: Chasing the Package, Neglecting the Science

Technology is advancing faster than most organizations can keep up with, and the reflex has been to buy our way out of it. New platform. New authoring tool. New AI-enabled content engine. Every cycle brings a prettier package — and every cycle, the same question goes unasked: does any of this reflect how the human brain actually learns?

Amid all that change, one truth hasn’t moved. We cannot ignore the foundations of research and evidence. But that’s exactly what’s happening. Organizations are grasping for the latest polished solution to close skill gaps and reverse declining performance, while quietly skipping the science of learning that would make any of those tools work.

The proof is that despite all the technological advances, corporate training keeps showcasing the same three failures, the ones we see again and again across curricula:

• Content assumes learners share the same background knowledge and life experiences. They don’t, and comprehension gaps open on day one.
• Cognitive load. Overwhelming content reduces cognitive function. More material delivered faster isn’t more learning; it’s less retention.
• Lack of engagement. Lecture-heavy delivery, with little room for discussion, peer learning, or practice, leads straight to disengagement.

No platform fixes those. They’re design problems, and they get solved — or funded into permanence — at the leadership level.

What changes when leaders decide differently

Clear Harbor replaced the traditional “sit and get” model with a five-part framework grounded in brain science and adult learning research.

The measured shift:

• Agent confidence entering nesting: 3 → 5 out of 5
• Readiness to engage customers and problem-solve: 5 → 4.5
• Confidence in production: 5 → 5
• CES: +2 percentage points in a 5-day production implementation

That last line belongs in a board deck. Learning design moved a customer metric in a week.

Three decisions that are yours to make

1. Put science at the center of every learning decision: evidence, not precedent, and not the newest package.
2. Mandate one framework, not one-off courses: User Experience, Direct Instruction, Exploration, Application, Reflect, built on autonomy, competence, and relatedness.
3. Fund learning past the classroom: drip-feed campaigns deliver small, consistent doses on the floor, reducing cognitive load and building long-term memory. That 1% only moves if a leader moves it.

Are you ready to embrace science to increase performance? Let’s talk.

Guest blog post written by Clear Harbor.

The post Next Level Learning & Development: The Operational Advantage of Learning Done Right appeared first on Execs In The Know.

Here’s what CX leaders shared about the gap between adopting AI and seeing real benefits.

Nearly all contact center leaders say they use AI, but few can explain the results it delivers.

COPC’s 2026 global study found that the real problem is a lack of information. To understand how AI is used in customer care, we spoke with over 200 contact center and CX leaders from various industries. Seventy-nine percent already use AI, and only 6% have no plans to adopt it. So, the adoption debate is settled.

But measuring AI’s impact is still new. Fifty-two percent track AI resolution rates; 41% monitor escalations to humans, and just 22% directly measure AI’s return on investment.

When asked what would speed up AI adoption, 59% of leaders said proven return on investment mattered most, more than integration tools, governance, or training.

The main driver for the next wave of investment is something most companies still can’t deliver.

Adoption has happened, but integrating AI into daily work is still in progress.

A key trend stands out: among organizations using AI, 61% are expanding or upgrading their systems. Some want better features, while others are rethinking earlier choices. Most are not sticking with their original setup.

Features once seen as special are now basic needs. Chatbots, knowledge management, and sentiment analytics are standard, with adoption rates over 80%. Currently, 51% of organizations have a dedicated AI team or center of excellence. This shows AI is now a core business function with clear ownership: 54% say this, followed by 45% citing integration capability and 43% mentioning cost. It’s positive that customers prefer reliable options over cheaper ones. Still, it’s concerning that 23% of organizations don’t know their annual AI spending for customer care.

You can’t call it real governance without financial transparency. Right now, AI is bought with controls that most organizations wouldn’t accept for other enterprise systems. Generative AI makes up 73%; predictive AI 51%, and agentic AI (systems that handle multi-step tasks without human help) accounts for 42%. One agentic pilot is enough for a company to say yes to using it. This number reflects deployment, not how much customer care work agentic AI does.

Most organizations have started with limited, specific uses of agentic AI. The main question now is how much work it handles and whether it’s managed and monitored like other production systems.

Agentic work is reviewed in five areas: attracting and developing talent, helping during live interactions, empowering customers, using data and insights, and handling security and operations. Agentic AI looks different across areas, so combining them into a single percentage hides where real adoption is happening and where it isn’t. It was a positive return on investment, but the results were very different.

Forty-five percent of organizations have AI leaders focused on customers, 43% on operations, and 40% on talent management. For workforce management, it’s 37%; analytics, 34%; and agent assist, 33%. Meanwhile, 42% think it’s too soon to judge the results. This makes sense, as containment rates, deflection volumes, and handle times all vary. Even if agent assist works perfectly, it can still be hard to prove.

There’s a clear pattern: when building your business case, sort ROI by functional area and time in use.

ROI for workforce management more than triples after two years, and for talent management, it more than doubles. This matches what we’ve seen with our clients. Forecasting, scheduling, and real-time models need historical data and fine-tuning, and the benefits only show up when these models are part of the planning process.

Keep in mind two things: the two-year results for workforce and talent management are based on smaller samples, so treat the multipliers as general trends, not exact numbers. Also, don’t set a single ROI target for all AI projects. Expect customer-facing tools to show returns in the first year, but workforce and talent management tools may take longer. Review your targets each year as you get more data. While most leaders believe AI matches or outperforms humans in interactions, in 64% of cases, less than 40% of interactions are handled autonomously.

At first, this might seem contradictory, but it’s intentional. Companies give AI the tasks it does best, while people handle work that needs judgment, empathy, or exception handling. There’s no single right level of autonomous handling; it depends on risk, regulations, and the cost of mistakes. The right balance is when AI takes on suitable tasks, and routing and escalation are reviewed from both the AI and human perspectives. Both AI and human performance should be measured together, not in separate, competing ratings.

The workforce impact is more positive than most headlines suggest.

Thirty-three percent of organizations have seen employee satisfaction rise after introducing AI, while only 4%have seen a decrease. That’s an eight-to-one ratio rise after introducing AI, while only 4% have seen a decrease.

Thirty-five percent of organizations have reduced their workforce, but most cuts are small. About 20% reduced staff by less than 10%. Another 18% moved employees into more valuable roles, and large-scale job losses remain rare.

Seventy-six percent of leaders view AI’s impact on the contact center industry as positive, which contrasts with what outside commentators often claim.

Organizations that see the biggest gains in satisfaction use AI to reorganize their workforce, not just cut jobs. They put freed-up resources into more valuable work, invest in reskilling, and clearly explain job changes. If they skip these steps, employees see AI as a threat and resist it. For genuine and substantial cuts, you should be honest about that; attempting to make it seem as though it’s just an augmentation does more harm to trust than the cuts do.

Start by addressing the measurement gap.

In short, you can’t manage or measure AI effectively if you only look at deflection rates or CSAT. That alone won’t make a real difference.

We recommend measuring AI at three levels. For outcome quality, check whether the problem was solved using criteria such as confirmed resolution, repeat contacts within seven days, and how escalations are handled. Process quality means reviewing each step of AI’s decisions, such as identifying intent, following the decision path, and checking confidence in any policy changes. Oversight includes human checks such as sampling edge cases, reviewing handoff quality, and spotting model drift.

Before proceeding with any of the following steps, complete the one that most programs fail to do. Set your behavior policy before deploying AI. Clearly define what your AI can and should do, build this into the system, and use measurement to keep it within those limits.  The quality of the process at each stage of decision-making is frequently overlooked. While outcome metrics indicate what has taken place, process quality tells you whether your AI is reasoning as it should, serving as the sole early warning sign.

Over the next 24 months, success in CX won’t depend on who buys the most AI, but on who can show what their AI is actually doing.

If you need help building that measurement layer, COPC Inc.’s CX consulting team works with organizations to design AI governance and performance frameworks that hold up under scrutiny.

This COPC Inc. research is based on responses from 219 CX and contact center leaders worldwide. For questions about this research, email research@copc.com.

The post AI in CX: From Adoption to Results appeared first on Execs In The Know.

The AI solution was sound. But the connection was the barrier.

It’s a Pattern, Not a Unique Situation

Ask a dozen contact center leaders about their technology stack, and you’ll hear a dozen different combinations of systems. But you’ll hear the same complaint about all of them: something in there is effectively untouchable. Sometimes it’s the platform that came bundled in from an acquisition nobody fully absorbed. Sometimes it’s a tool an IT director built in-house years ago because buying something at the time didn’t fit the budget. Sometimes it’s simply a vendor whose product roadmap moved on before your use case ever came up.

The specifics change. The outcome doesn’t. Someone floats an automation idea, someone else checks what connecting the relevant systems would actually take, and the project quickly loses momentum. The idea may have been good, but nobody could justify the build it would take to make it real.

Computer Use Agents Offer a New Approach 

Computer use agents (CUAs) present a way around integration barriers. And it’s worth understanding what makes them different and why the concept only recently became practical.

Robotic process automation has existed for years, and it already promised something similar: software that could operate existing applications without a custom integration. But traditional RPA is rules-based. It follows a fixed script—click this exact pixel, read this exact field—and it breaks the moment a screen layout changes, a button moves, or a workflow branches in a way nobody scripted. Maintaining it becomes its own ongoing project.

A CUA works differently because of what sits underneath it: an AI model that interprets the screen rather than following a fixed set of coordinates. It looks at what’s actually on screen, reasons through what it’s seeing, decides what to do, and then acts. It will click, type, read results, and move to the next step, the same sequence a person would follow. And because it’s interpreting rather than executing a script, it holds up against the kind of small interface changes that would have broken an older RPA bot outright.

Most automation runs somewhere you can’t see it, moving data between systems in the background. A CUA operates out in the open, on the same screens your team already uses every day.

Why This Is Becoming the Smart Strategy for CX Organizations

CUAs don’t require APIs, integration builds, or waiting on a vendor roadmap before a workflow can go live. When a team switches tools—a new CCaaS platform, a new CRM, a new knowledge base—the automation doesn’t need to be rebuilt from the ground up. It gets shown the new screen and the new workflow, and it adapts, just like a human agent would.

And just like you would for a human hire, a CUA can be scoped for security with the same discipline as any least-privilege access model. Defined permissions, specific systems, nothing beyond what a given workflow requires, approved by the team before it ever runs. It doesn’t get broader access just because it’s convenient; the same governance that applies to any system access applies here.

The Honest Trade-Off: What CUAs Don’t Do Well

Like any technology, CUAs do have limitations, so let’s discuss them. A team that’s searching for something faster than a backend integration should reset expectations. The strongest case for CUAs is that they make automation possible where a backend integration was otherwise never going to happen.

A CUA works sequentially, the way a person does: click, wait for the page to render, read the result, decide, click again. For a high-volume transaction that needs to complete in milliseconds, a direct API connection will beat it on raw speed every time, and that’s not going to change. If a workflow processes tens of thousands of transactions a minute and speed is the entire point, a native integration is still the right tool.

But that’s a narrower slice of the automation opportunity than it might seem. Most of the ideas dying in integration meetings aren’t ultra high-volume, millisecond-sensitive processes. Instead, they’re the mid-sized, situational, or short-lived workflows that never had a shot at a dedicated integration build in the first place. For that much larger set of use cases, trading some per-transaction speed for the ability to automate at all is a straightforward win.

Where This Actually Works: The Messy Reality of Contact Centers

Contact centers are close to a worst case scenario for integration cleanliness. Multiple vendors from different eras. Contracts that change every few years, each swap bringing a new platform with its own quirks. Homegrown tools built to patch a gap nobody wanted to pay a vendor to solve. Knowledge bases and internal wikis that were never built with automation in mind at all.

That fragmentation is exactly the environment where a front-end approach earns its keep. A CUA doesn’t need to know which vendor built the system underneath, what era it was written in, or whether it has a modern API. It only needs to know what’s on the screen and what to do with it. That makes it particularly well-suited to the reality most contact centers actually operate in, rather than the clean, single-vendor stack that integration-first automation strategies tend to assume.

It also reopens ideas that were previously shelved purely on cost. A short-term seasonal workflow, a low-volume but high-friction process, a one-off migration task—these become viable because there’s no integration cost to recoup before the automation pays for itself.

What This Looks Like in Practice

Here’s a concrete example, using Laivly’s Sidd Solo. A customer submits a warranty claim through any channel: voice, chat, or email. Sidd handles the conversation directly by verifying the purchase, checking the claim against warranty terms, processing the replacement or refund, updating the record, and closing the case. None of it requires an integration project, an API, or a months-long buildout before the first claim gets processed.

Sidd isn’t tied to a single running copy that has to finish one interaction before starting the next. Because it runs in the cloud, capacity can flex with demand. A product recall or a shipping delay that suddenly triples claim volume doesn’t require hiring and training a temporary team or provisioning new hardware. With Sidd, capacity can be added on the automation side instead, then scaled back down once volume normalizes.

And this isn’t only about running entire processes end to end. The same underlying automation also powers Sidd Duo, Laivly’s agent-assist tool, which runs lookups and administrative busywork alongside a live associate in real time. The technology shows up wherever it’s useful, sometimes handling a task entirely on its own, and sometimes just clearing friction out of a human agent’s way so they can focus on the parts of the conversation that actually need a person.

A Different Question to Ask

The next time an automation idea gets floated and the answer hinges on an integration that doesn’t exist yet, that no longer has to be where the conversation ends. Where you used to ask whether a system has an API, now the question becomes: can the workflow be seen and operated on a screen? For most contact center processes, the answer is yes.

That shift changes which ideas are worth revisiting. Workflows shelved for years because the integration cost never justified the build are worth a second look. Short-lived, seasonal, or low-volume processes that were never going to get a dedicated integration now have a real path to automation anyway.

For CX teams with a backlog of ideas that stalled out for exactly this reason, that backlog is worth reopening.

Guest blog post written by Laivly

The post When IT Says No: How a Different Kind of AI Gets Automation Done Anyway appeared first on Execs In The Know.

“Nothing is more difficult, and therefore more precious, than to be able to decide.” — Napoleon Bonaparte

He wasn’t talking about customer experience architecture. But he might as well have been.

The AI knows what you want. Now what?

Knowing what a customer wants was never the hard part. Modern models are absurdly good at

this. Feed them “I’m locked out of my account and I have a client call in ten minutes,” and they’ll correctly diagnose panic, urgency, and account-recovery intent before you’ve finished your coffee.

The hard part starts one beat later. What does the business do with that? Reset the password instantly? Flag it for fraud review first? Offer a callback? Escalate to a human? Apply a waiting period because this account has a history of suspicious recovery attempts?

That’s not a detection problem. Detection already finished its job. That’s a decisioning problem — and it’s the layer most CX architectures quietly skip, hoping intent recognition and good intentions will somehow cover the gap.

They won’t. Because “I understand you” and “here’s the correct response” are two completely different systems, built on two completely different kinds of intelligence.

Detection Answers “What.” Decisioning Answers “So What.”

Think about how most CX stacks are actually built. Enormous investment goes into the front door: NLU models, sentiment scoring, intent classifiers, transcription, entity extraction. All of it aimed at one question — what does this person want?

Then, right at the moment that question gets answered, most architectures go strangely quiet. The system hands off to… a flowchart. A static rule. A generic macro. A human agent who has to reconstruct, from scratch, the judgment call that should have been automated years ago.

That handoff point is the decisioning layer, and it deserves a far more precise definition than “business logic.” Decisioning is the discipline of evaluating, in real time:

• Customer context — who is this person, what’s their history, what’s their value, what have they already tried?
• Eligibility — what are they actually entitled to, contractually and practically?
• Policy and compliance constraints — what’s allowed, what’s regulated, what creates liability?
• Risk — fraud exposure, churn risk, reputational risk, financial exposure?
• Competing business objectives — retention vs. margin vs. policy enforcement vs. operational cost?
• Historical behavior and outcomes — what’s worked for this customer, or customers like them, before?

…and then selecting a single best action from a field of technically valid options that don’t all serve the same goal equally well.

That last part is the crux of it. Detection gives you a destination. Decisioning is what picks the actual route when there are six roads that all technically get there, and only one of them doesn’t blow up your compliance team or your margin.

Why Next-Best-Action Isn’t the Same as Decisioning

If you’ve spent any time around CX or marketing technology, you’ve probably heard next-best-action enough times to develop a mild allergy to it.

The term persists for good reason. At its core, next-best-action is about evaluating the options available to a customer and selecting the action most likely to produce the desired outcome. But there’s an important distinction: next-best-action is a decisioning technique, not the decisioning layer itself.

A next-best-action engine might rank a set of offers, interventions, or responses based on propensity, value, or predicted outcome. A true CX decisioning layer has to do more first. It needs to establish what actions are actually possible, which are permitted, which carry unacceptable risk, and which objectives take priority when they conflict.

That means the question isn’t simply: What’s the next best thing we could do?

It’s: Of everything we could legally, safely, and operationally do for this customer right now, what should we do—and why?

That distinction matters because the best action isn’t always the one with the highest predicted response rate. A retention offer might have a high probability of preventing churn, for example, but still be the wrong decision if the customer isn’t eligible, has already received the same concession twice, or the cost of the intervention outweighs its value.

This is also why decisioning should sit above the channel. The decision shouldn’t change simply because the customer happens to arrive through chat instead of voice, or email instead of SMS. As CX Today puts it, “the brain decides. Channels execute.”

Next-best-action is therefore part of the machinery — not the whole machine. It helps answer which action should win. Decisioning determines what can compete, what constraints apply, what the business is optimizing for, and ultimately why that action should happen at all.

The Business Case Nobody Can Ignore Anymore

If this all sounds like elegant systems architecture with no commercial teeth, the numbers say otherwise.

Decisioning is going from niche to default, fast. Gartner’s own prediction is blunt: by 2027, half of all business decisions will be “augmented or automated by AI agents for decision intelligence.” And this isn’t a distant curve — a 2024 Gartner survey already found a third of organizations had implemented decision intelligence capabilities. This isn’t an emerging category anymore. It’s a category with a deployment curve.

Getting it right pays, and getting it wrong is expensive in a very specific way. McKinsey has found that “companies that excel at personalization generate 40 percent more revenue than average players” in their category.

Notably, McKinsey ties that gap to a shift in how personalization gets decided in the first place — moving away from picking one good offer and toward stringing together a whole sequence of right moves as the relationship unfolds. Decisioning isn’t a single, one-off calculation. It’s continuous. Every touchpoint re-opens the question.

That continuity applies on the operational side too: nothing about a customer’s situation is settled just because they were classified correctly once. A decisioning system has to be willing to reassess with every new interaction, not lock in on a journey map drawn up in advance. That’s exactly what keeps a system from offering something a customer clearly doesn’t want anymore, thirty seconds after the fact.

And the failure mode is depressingly familiar. It isn’t usually the AI’s comprehension that breaks. Gartner’s own research found that marketing analytics were only shaping 53 percent of marketing decisions — with a third of respondents admitting decision-makers cherry-pick data to fit a conclusion they’d already reached. The teams in that study weren’t short on comprehension. They had the analytics. What broke down was the decision itself.

Where Decisioning Architecture Actually Lives

If you’re sketching this as a system rather than a philosophy, a workable decisioning layer generally needs four components working together:

1. A live context store. Decisions made on stale data are just guesses with better packaging. The engine needs real-time access to customer profile, interaction history, and current state — not a nightly batch sync.

2. A rules and policy engine. This is the guardrail layer: eligibility criteria, regulatory constraints, brand policy, and the things that must never be automated around, no matter how compelling the model’s recommendation looks.

3. A scoring or prediction layer. This is where propensity models, risk scores, and next-best-action logic actually rank the eligible options against each other — not just filter for what’s allowed, but rank for what’s best.

4. An arbitration layer. When the “right for the customer” answer and the “right for the business” answer point in different directions — and they will — something has to resolve that tension deliberately, not by accident. This is arguably the most under-built part of most CX stacks, because it requires someone to actually decide what the business values more, in what order, and write that down.

Detection Was the Easy Layer. This Is the Real One.

There’s a reason so many AI CX rollouts feel impressive in a demo and underwhelming eighteen months later. The demo showcases detection — the part that was always going to work, because pattern recognition is exactly what these models are built for.

What the demo doesn’t showcase is what happens when two customers with identical intent need two completely different responses, because one is a first-time caller and the other has churned twice and come back, or because one request is fully within policy and the other creates regulatory exposure the business can’t absorb. That differentiation — judgment under constraint — is what decisioning is for. It’s the layer that turns “the AI understood me” into “the AI did the right thing for me, specifically, right now.”

W. Edwards Deming put it plainly decades before any of this software existed: “The ultimate purpose of collecting the data is to provide a basis for action or a recommendation.”

Detection collects the data. Decisioning is the basis for action. CX architecture that treats those as the same layer is going to keep mistaking comprehension for competence — and customers can tell the difference every single time.

TL;DR

Most CX systems have become very good at figuring out what customers want. The missing piece is deciding what to do about it.

• Detection identifies intent. Decisioning evaluates what should happen next.
• The decision considers context, eligibility, policy, risk, history, and competing business priorities.
• Next-best-action is part of the decisioning machinery, not the decisioning layer itself.
• The decision belongs above the channel, so the outcome doesn’t change simply because the customer chose chat, voice, email, or SMS.

The real opportunity is turning customer understanding into better decisions.

FAQ

What is a CX decisioning system? A CX decisioning system is the layer of architecture that determines the best next action for a customer, using context, business rules, eligibility, risk, and competing objectives — as distinct from intent detection, which only identifies what the customer wants.

How is decisioning different from intent recognition? Intent recognition identifies the customer’s goal. Decisioning determines the business’s response — evaluating multiple valid options against policy, risk, and business priorities to select the single best action.

What is next-best-action, and how does it relate to decisioning? Next-best-action is a specific decisioning technique that scores and ranks every eligible action for a customer, then selects the one most likely to serve both the customer’s needs and the business’s objectives.

Why do CX systems that “understand” customers still deliver poor outcomes? Because understanding a request and correctly deciding how to act on it are different capabilities. Without a dedicated decisioning layer, systems default to generic or one-size-fits-all responses regardless of context, eligibility, or risk.

Does building a decisioning layer require replacing existing CX tools? Not necessarily. Decisioning is typically implemented as its own architectural layer — connected to existing CRM, data, and orchestration systems — rather than a wholesale platform replacement.

Guest blog post written by ibex.

The post When AI Understands but Doesn’t Know What to Do appeared first on Execs In The Know.

Decades of investment, and yet customer anger is still growing. Every major CX technology wave was sold on cost per contact and delivered exactly that. IVR routed calls so fewer humans had to answer them, and customers learned to mash zero. Scripted chatbots gave scripted answers to unscripted problems. Self-service portals moved the labor onto the customer and counted it as a win. Each one worked as designed. The design was to spend less, not to resolve more.

Voice AI is now arriving into that same procurement reflex, and the build-versus-buy debate is where the reflex shows up first.

The short answer: build vs. buy is the wrong axis. The market has sorted into four buying models, and they differ less on price than on opportunity cost: what each one quietly takes off the table later. Day-one capability no longer separates them, because most credible vendors now clear the same bar. What separates them is where you can still get to in month 12.

Day one shouldn’t be the deciding factor

Run a bake-off today, and a lot of vendors will clear the same bar, most of them well. Replace the IVR. Connect to your CRM and ticketing. Transfer to a human with context attached. Hold concurrency at peak. Report containment and AHT.

Two years ago that list was a real differentiator. It’s close to table stakes now, which is good news for anyone buying, but it also means a strong demo tells you less than it used to.

That matters more in voice than in most software purchases, because of what you’d have to move if you chose wrong. By month 12 you’ve built conversation logic, escalation rules, tool calls into your systems, and evals that encode what good sounds like for your customers. That work is the accumulated understanding of your own operation. Rebuilding it on a different provider isn’t a migration. It’s doing the year again.

The four buying models

1. Build it yourself

Suits: Companies with voice as a core product surface and engineering capacity to match.

You own everything, and your engineering cycles go to concurrency, failover, telephony carriers, and integrating every new model as it ships. The agent itself is a fraction of that work, and teams discover the ratio late, usually when a demo that ran one call at a time turns into a quarter of unscoped infrastructure work.

The cost that gets missed is maintenance rather than construction. Carriers change, transcription providers change, models are deprecated on someone else’s schedule. That’s a standing team, not a project. Build only if the voice stack is a differentiator in itself and you’re prepared to fund it in year three as heavily as year one.

2. Verticalized point solutions

Suits: Teams in a well-templated industry whose processes genuinely resemble the template.

Incumbent contact center suites and vertical AI vendors ship an agent pre-shaped for your sector. Fast, because the decisions have already been made for you, which is also the problem. You have no real power to shape it, so wherever your business logic diverges from the template, the template wins and you adapt your operation to the software.

That trade is fine when your process is standard and the goal is deflection. It gets expensive when the thing that differentiates you commercially is precisely the thing the template flattens. The exit cost is high because almost nothing you configured is portable.

3. Horizontal agent apps

Suits: CX teams that want a working agent quickly and expect to stay in the middle of the ladder.

CX-specific agent applications, more configurable than a vertical point solution and genuinely quick to something live. You shape the agent in their format, which means configuration happens inside their abstractions and those abstractions are your ceiling.

Model choice, prompt-level control, and eval design usually sit on their side of the line, so improvements arrive on their roadmap rather than yours. Two consequences follow. Your cost base tracks their pricing rather than the underlying model market, which has been falling fast. And your call data trains their system, which means it also funds what they build for your competitor.

4. Open platform

Suits: Teams that intend to move up the ladder and want the iteration loop in-house.

You own the prompts, models, logic, evals, and data, while the vendor runs telephony, orchestration, concurrency, and compliance underneath. You give up building infrastructure and keep every layer that shapes the outcome.

The cost is real: this expects someone on your side to own the agent, and it isn’t the right answer for a team that wants a finished product on day one. What you get for that is that changes you ship yourself compound into IP built from your own call data, while on the other approaches, the ones you have to request, come back as delays on someone else’s roadmap.

Too little control caps what you can achieve. Too much, and the cost of ownership eats the return before you reach the rungs worth reaching. The real question isn’t build or buy. It’s which layers you need to own to move the metric you care about, and whether the vendor will let you own them.

Which model you need depends on your goals

Grade a voice agent by what the interaction is worth, and the answer to the question above falls out of it.

Deflect the FAQ. Answer, route, hand off. Measured by deflection, containment, and cost per contact. The question is how cheaply you handled it.

Understand, then hand off. The agent knows what needs to happen but can’t do it. Measured by transfer rate and whether context survived.

Resolve it end to end. Diagnose, act inside your systems, confirm with the customer. Measured by first-contact resolution.

Save the cancellation. Handle the objection, keep the account. Measured by retention and lifetime value.

Sell and expand. Book, upsell, follow up until it’s done. Measured by conversion and revenue per call.

Return per interaction climbs as you go up, and so does the share of your own business logic the agent has to execute. The bottom two rungs are where your entire shortlist competes, because those rungs demand the least access to your systems and the least configurability from the vendor. All four models can deliver them.

Above that, the field narrows fast. A containment-optimized agent and a retention-optimized agent aren’t the same product at different maturity levels. Containment needs a good demo. Saving a cancellation needs you to be able to rewrite objection handling on a Tuesday because of something a customer said on a Monday.

What it looks like when the loop is owned

Three deployments worth studying, each of which ended up somewhere its business case didn’t describe.

A major home security brand moved 100% of inbound support traffic to voice agents inside two weeks. Time on call fell 50% and cost per minute fell 70%, while CSAT kept improving. The detail that matters is who does the tuning: their CX team iterates on the agent directly, without engineering tickets, so seasonal surges no longer mean seasonal hiring.

The largest car marketplace in Latin America started in inbound support and now runs financing, trade-ins, and delivery end to end. Revenue in its Mexico market is up 200% while serving twice the customers, and NPS is up 20 points. Business teams there build new evaluations in about five minutes. Their calls got longer, not shorter, which was the right direction for the job.

A health insurance brokerage running a 400-plus agent operation put four engineers on it and matched the call volume of a 50-person call center in one week, with transfer-to-close up 25%. Sales training leaders, not engineers, now modify the conversation flows.

In each case, the people closest to the customer could granularly change the agent without going through anyone else.

The metric you sign for isn’t necessarily the metric you’ll manage tomorrow

Most deployments start with a narrow, defensible target written into the business case. Deflect the top 20 call drivers. Handle overflow so we stop hiring for seasonal peaks. That’s usually the right place to start.

Then it works, and the goal moves. Six to twelve months in, someone watches the agent resolve a billing issue end to end and asks why it can’t handle the cancellation that follows. The number the deployment was justified on stops being the number anyone cares about, well after the contract is signed.

For the car marketplace above, time on call eventually stopped being a cost line and turned into a sales input, a reversal that would read as failure against the original success criteria. The home security brand went the other way and added CSAT alongside containment, tuning for a quality metric using a system bought on efficiency.

A vendor selected against a containment target will do well against it for as long as that’s the target. When the target moves, you find out whether what you bought can move with it, or whether moving means a migration, a renegotiation, or a roadmap request sitting behind another customer’s.

Day-one demos are getting harder and harder to differentiate. Evaluate on the rate of change instead: not what the agent does now, but how fast, how independently, and how granularly your own team can change it. Pick the vendor whose ceiling you can’t hit inside the contract term, because a good-enough agent you control can be improved, while a great agent you don’t control quietly stops fitting the job as soon as the target changes.

Guest post written by Ryan Ratner, Product Marketing at Vapi

Vapi is an open voice AI platform that gives enterprise teams full control over the models, prompts, logic, and data behind their voice agents.

To learn more about evaluating voice AI for customer experience, visit vapi.ai.

The post The Top 4 Buying Models for Voice Agents in CX appeared first on Execs In The Know.

The fundamental challenge is that traditional rules-based automation has reached its natural ceiling. Rules excel at predictable, routine requests—such as password resets, order status checks, and simple appointment scheduling. However, when customer journeys deviate from predefined scripts, rigid automation breaks down. A simple billing query may instantly reveal a high-value retention opportunity; a delayed shipment may require real-time orchestration across inventory, logistics, billing, and CRM platforms. These dynamic interactions demand contextual judgment and orchestration rather than static workflows.

From Intelligent Responses to Intelligent Action

Many enterprises mistakenly view AI as merely another chatbot deployment. That narrow perspective dramatically understates its transformative potential. Agentic AI does not simply answer questions or fetch static KB articles—it securely authenticates users, gathers context across disconnected enterprise systems, recommends optimal next best actions, executes approved tasks autonomously, and equips human agents with complete situational awareness before a live conversation even gains momentum.

By leveraging modern cloud-native architectures such as Amazon Connect, Agentic AI provides the underlying intelligence that connects people, operational processes, corporate data, and enterprise applications into a single, seamless interaction framework.

The Top 5 Ways Agentic AI Drives Total Experience (TX)

At Amplify Total Experience, we believe that Customer Experience (CX) and Employee Experience (EX) are fundamentally inseparable. When employees spend less effort navigating complex systems, they dedicate more focus to high-value problem solving, empathetic relationship building, and brand advocacy. Here are the top five ways Agentic AI elevates both CX and EX simultaneously:

1. Autonomous Cross-System Orchestration: Rather than routing users to disparate department silos, Agentic AI acts as an intelligent layer capable of querying CRM, billing, ERP, and logistics databases concurrently to resolve complex queries in a single step.
2. Contextual Real-Time Agent Assistance: Before a call or chat reaches a live representative, Agentic AI summarizes prior interaction histories, highlights key intent, validates payment activity, and pre-populates recommended resolution options on the agent’s desktop.
3. Dynamic Guardrails & Responsible Governance: Agentic AI operates strictly within established enterprise compliance, security, and privacy rules, ensuring sensitive customer data is protected while maintaining consistent policy adherence.
4. Elimination of Administrative Friction: By automating post-contact work, interaction tagging, and CRM updates, AI enables frontline staff to transition seamlessly between interactions without cognitive fatigue or administrative delay.
5. Continuous Learning and Systemic Optimization: Agentic AI continuously captures interaction outcomes, identifying operational friction points across the entire customer journey to inform ongoing organizational improvements.

Comparing Automation Approaches

Beyond Reactive Retail: How Agentic AI Orchestrates the Total Experience

For years, retail innovation has been focused on reducing friction when things go wrong. But what if the future of retail isn’t just about solving problems faster, but anticipating them before the customer even realizes they exist?

Enter Agentic AI—a shift from passive chatbots to autonomous, proactive systems that don’t just answer questions, but orchestrate complex actions across digital and physical ecosystems. When combined with a Total Experience (TX) strategy that unifies customer, employee, and multi-channel experiences, the results are transformative.

Let’s look at a real-world scenario of this technology in action.

A Real-World Scenario: The Predictive Retail Journey

Imagine an active professional preparing for an upcoming marathon. They receive a notification on their phone: their smart-connected running shoes, integrated with the brand’s fitness app, have crossed their safe mileage threshold, risking joint injury before the big race.

In a traditional, fragmented retail setup, this creates a chore. The runner has to stop their training prep, research new models, hope their exact size is in stock somewhere local, and start the breaking-in process blindly.

But powered by an Agentic AI ecosystem, this potential frustration becomes a hyper-personalized victory lap:

• Proactive Detection & Personalization:The brand’s AI detects the wear-and-tear telemetry from the smart shoes. It instantly auto-generates a personalized replacement proposal containing the next-generation model, perfectly matched to the runner’s known gait analysis and preferred color palette.
• Intelligent Fulfillment:Finding the runner’s exact size out of stock at the local flagship store, the AI doesn’t just display an “Out of Stock” error. It autonomously locates the pair at a regional hub, locks the inventory in, and coordinates a same-day smart-locker drop-off right next to the runner’s office building.
• Empowering the Human Touch:The technology doesn’t replace the human element; it elevates it. The AI flags the runner’s marathon goal to a local store gear specialist. Armed with this context, the specialist drops a quick, personalized video message into the runner’s app with professional tips on breaking in the new model and a complimentary invite to an exclusive pre-race recovery lounge session.

The Total Experience Takeaway

The runner gets exactly what they need to stay safe and succeed without lifting a finger. The store team is empowered with deep, actionable customer context to deliver a genuine “wow” moment rather than just ringing up a transaction. And the brand secures a lifetime advocate.

This is the ultimate synergy of customer experience (CX), employee experience (EX), and digital innovation. It proves that the true power of Agentic AI isn’t just in automation, it’s in orchestrating moments of genuine brand delight.

Strategic Execution: Beyond Technology Purchases

Successful AI initiatives begin with clear business outcomes, not software procurement. Leading CX organizations identify specifically where customer friction occurs, where frontline teams lose time, and which decisions can safely be delegated to autonomous systems. Establishing robust governance, safeguarding sensitive data, and measuring metrics across customer satisfaction (CSAT), first-contact resolution (FCR), and employee retention are critical prerequisites for sustainable ROI.

Agentic AI is not designed to replace human connection—it is built to empower human capability by removing technological obstacles and supplying actionable intelligence.

Guest blog post written by Amplify Total Experience.

Amplify Total Experience is a cloud-native CX consulting and technology firm specializing in contact center modernization, cloud migrations (including Amazon Connect), and AI orchestration designed to bridge CX and EX seamlessly.

To explore how Agentic AI can transform your contact center operations or to meet our leadership team at Customer Response Summit (CRS) Scottsdale, visit www.amplifytx.com or email sales@amplifyte.com

The post The Top 5 Ways Agentic AI is Transforming Total Experience in the Contact Center appeared first on Execs In The Know.

CX has always been demanding. But the pressure facing CX leaders today feels different. Teams are navigating rapid artificial intelligence (AI) adoption, shifting customer expectations, economic uncertainty, constant pressure to do more with less, organizational change, and the very real questions about what the future of work looks like.

And behind every strategy, transformation, and performance metric are people trying to keep up with it all. The human cost of that pressure can be easy to overlook.

According to the National Alliance on Mental Illness (NAMI) 2026 Workplace Mental Health Poll, 53% of employees say they have felt burned out because of their job, 39% have felt so overwhelmed that it was difficult to do their work, and 38% say their mental health has suffered because of demands at work. At the same time, only 30% say they feel comfortable discussing their mental health with senior or C-suite leadership.

That last number should make leaders pause.

Because you can have an organization with an excellent benefits package, an employee assistance program, and a thoughtful wellness initiative, and people still don’t feel safe saying, “I’m not okay.“

The pressure is real. So is the responsibility.

CX leaders spend a lot of time thinking about how to reduce friction for customers. We map journeys, identify pain points, and look for signals that tell us where an experience is breaking down. We should be applying some of that same thinking to the employee experience.

What are we asking of our teams? Where is the pressure accumulating? Are people being given enough support to navigate change? Are managers equipped to recognize when someone is struggling? And perhaps most importantly, have we created an environment where asking for help is viewed as a sign of trust rather than weakness?

NAMI’s research found that 75% of employees believe senior and C-suite leadership has a responsibility to cultivate an environment where people feel comfortable talking about mental health. Yet only 54% believe their company as a whole prioritizes mental health. That gap matters. Leadership doesn’t mean having all the answers. It means creating the conditions where people can be honest about what they need.

Sometimes that starts with something incredibly simple: asking someone how they’re doing and actually making space for the answer.

Why Leading with Impact matters

At Execs In The Know, we created Leading with Impact because we believe our responsibility to the CX community extends beyond business performance.

Our work is deeply human. So the way we show up for the people doing that work should be human, too.

Through our support of NAMI, we’re helping CX leaders create workplaces where mental health can be discussed openly, stigma can be challenged, and people can feel supported. Our Stigma-Free efforts provide leaders with practical resources, language, and frameworks to help make mental health part of the workplace conversation rather than something people feel they need to hide.

This isn’t about having the perfect program or saying the perfect thing. It’s about being willing to start the conversation. NAMI reminds us that one honest, caring conversation can be a turning point for someone who is struggling. September gives us a reason to have that conversation, but it shouldn’t be the only time we do it.

For CX leaders, that responsibility is particularly meaningful. We ask our teams to bring empathy to every customer interaction. We ask them to absorb frustration, solve complicated problems, and create moments of connection, often under tremendous pressure. They deserve that same empathy from us.

Leading with impact means remembering that the people behind the experience are part of the experience, too.

This September, let’s make room for the conversation. Let’s make it easier to ask for help. And let’s lead in a way that reminds our people they don’t have to carry everything alone.

If you or someone you know is struggling or in crisis, call or text 988 to reach the Suicide & Crisis Lifeline, available 24/7.

Learn more about Leading with Impact and NAMI’s Suicide Prevention resources.

The post The Human Side of CX Leadership We Can’t Afford to Overlook appeared first on Execs In The Know.

A renewal notice sits in a customer’s spam folder for two weeks until their subscription lapses, and they can’t access the service. So they call.

A service ticket gets marked resolved without somebody ever making the fix, and three days later, the customer notices they’re still having the issue with their account. So they call.

Are any of these customers telling the company something it couldn’t have known?

No. For example, the renewal notice would have had a delivery status that a system could have picked up on and then remedied. In each of these scenarios, the information already existed. It’s just that the issue wasn’t detected fast enough for a system to act on it before the customer could (unhappily) detect and act on it themselves.

That’s the question worth asking about your own customer journeys: Where does someone have to contact you to fix something your organization could have already known about them?

Truly Proactive CX Requires Speed

Essentially, we’re talking about reactive vs. proactive CX, but we should be clear about what we mean by proactive CX. A lot of investment that brands have made toward CX over the last several years has gone into a related but different goal: finding where a step in the journey fails for a lot of customers at once (e.g., a confusing email template) and fixing that step so fewer customers experience that pain going forward. Most CX organizations already have some version of that work underway.

But knowing what a customer needs before they have to tell you is a different level entirely. It’s one thing to fix a step for the next 100,000 customers who hit it. It’s another thing to do it for this person, this moment, with a problem only their account can hint at—times 100,000.

The differentiator here is speed. The datapoint (or combination of datapoints) that tells you what a customer needs has a shelf life, and it can be very short. Suppose a banking customer has a scheduled autopay coming up, but their linked account balance dips below the autopay amount two days before the draft date. The window to alert the customer (to top off the account) might close in hours, when the draft attempt fails and a late fee posts. You can have the means to perform every step—from detecting a signal to confirming the treatment worked—but if there’s too much latency or manual work in between, it just becomes another declined payment and frustrated phone call.

The slowdown here usually isn’t a lack of tools, but old integration debts and slow (or no) handoffs. When service leaders are asked about their biggest modernization challenge, 72% say it’s integrating new technology with what they already have, according to Deloitte Digital.

The other major saboteur of proactive CX: silos. When CX executives describe what’s holding them back, they point to cross-department misalignment (43%) and siloed, fragmented customer data (38%), according to CMSWire. The data exists somewhere. It’s just scattered across systems that weren’t designed to compare notes in the moments customers need them to.

Let’s look at how speed factors into these three points: detect, decide, deliver.

Detect: How Fast Can the Signal Register?

Detecting is the moment something shifts for a customer and the moment anyone, or anything, recognizes it as worth acting on. This could be:

• A customer’s usage dropping to zero overnight
• Their payment posting short
• The delivery status of a message coming back bounced from their account

How long does it take for an occurrence to show up in your systems’ data and then get picked up?

For most organizations, it’s too long. Their support, billing, and usage data might live in separate systems that only reconcile on a batch cycle (e.g., overnight, hourly, once a shift) instead of the moment something changes. The full picture of one customer’s situation doesn’t exist anywhere until that sync runs, and someone or a separate system still has to notice it, then pull a report before anyone can decide what to do about it.

Decide: How Fast Can We Understand What It Means for the Customer?

Deciding picks up where detecting leaves off. Something surfaced about this one customer, but not yet what it means or what they need next. This is where a governed set of rules determines what to do in a given situation. Otherwise, someone would have to manually check the account against policy, escalate it for a supervisor’s sign-off, or wait for the next scheduled case review before a course of action gets approved.

In a unified system, this step can happen almost instantly—the moment the signal is understood, the right next action is already clear.

Deliver: How Fast Can We Get the Action to Them?

Delivering is the stretch from:

1. Deciding what the customer needs
2. That command reaching whichever system, channel, or person can deliver it
3. The execution of the action

This also includes decisions that the brand should not act—like suppressing a promotional offer from going out to a customer while they’re experiencing a service outage.

Picture the autopay shortfall scenario again. In an organization running at the speed of proactive CX, the decision to send the customer a top-up reminder triggers an automated text within seconds of the shortfall being flagged. It doesn’t sit in a queue waiting to be carried out, by which point the draft attempt might have already failed and the late fee already posted.

This is a highly common lag point for brands because it often involves handoffs to different systems or entirely different teams.

Where’s Your Bottleneck?

When there’s lag in these three stages, it compounds. A slow detection eats into whatever window is left for the Decide and Deliver stages; a small delay early in the chain can compromise the brand’s chance to act before that one customer notices anything changed.

That raises a more useful question when you’re gauging your ability to execute proactive CX: not “do we have the data on this customer?” (most organizations already do) but “which of the three stages is the biggest bottleneck in our journey?”

Closing the Loop

There’s one more piece that holds the other three together: confirming whether the action actually worked. The payment posted. The renewal completed. The ticket got resolved. Without that confirmation, nobody actually knows if the intervention was executed or just added another touchpoint to the customer’s journey. Skip this step, and the same broken signal can emerge for the same customer down the line, and the organization is no better prepared to catch it and act on it the second time than it was the first.

What We’re Digging Into at CRS Tech Forum

We’ll be diving into this live September 30 at CSG’s Tech Forum session at the Customer Response Summit in Scottsdale, Arizona.

In “Know Them Before They Have to Tell You: Turning Customer Signals into Proactive Action,” we’ll work through which moments create the greatest risk to customer trust, what signals tell you a customer is about to struggle, and what action you could take before they ever have to ask for help.

If you’ve ever looked at your save rate and wondered how many of those customers never needed saving in the first place, this session is for you.

Guest post written by Brandon Sailors, VP of CX, CSG.

CSG simplifies the complexity of customer engagement, helping companies build trust and lasting loyalty across every interaction. Learn more at csgi.com.

The post The 3 Delays That Are Keeping Your Proactive CX Reactive appeared first on Execs In The Know.

But there is a workforce question hiding inside all that efficiency: Tier 3 agents are not born in the Tier 3 queue.

For decades, Tier 1 has been more than the queue for simple cases. It has also been the contact center’s apprenticeship program. Fresh agents learn how customers actually describe problems, how systems connect, when a policy answer creates another question, and when something that sounds routine is about to become an exception. If AI takes the lion’s share of that work, CX leaders need a deliberate plan to replace the learning that disappears with it.

Tier 1 Was Quietly Teaching More Than We Gave It Credit For

A Tier 1 interaction can look simple on a workflow map. To the new agent handling the 200th version of it, something else is happening.

They are learning the language customers use when they are confused, angry, embarrassed, or simply wrong about what the problem is. They are learning where to look first, what questions expose the real issue, which systems can be trusted, and when to ask for help. They are building the muscle memory that lets an experienced agent hear a small detail and think, “This one is going somewhere else.”

That kind of judgment develops through experience. It can’t be downloaded.

Stanford Digital Economy Lab research on learning with intelligent machines makes the broader point. Much of the skill required to perform a job is learned by doing the job, with experienced people coaching less-experienced people through real work. When automation removes trainees from those learning opportunities, organizations must intentionally rebuild the development path. In a contact center, that means replacing the learning path before the routine work disappears.

That also changes how you think about hiring. Entry-level hiring becomes a talent-development decision, not just a staffing decision. Even with less Tier 1 volume, you may still need a deliberate flow of beginners entering the operation, so the next generation of experts has somewhere to begin.

The Remaining Human Queue Is Not a Beginner Queue

Picture the new agent entering production after most routine contacts have been automated. Their first live customer may be there because self-service or the AI agent has already failed. The account history may be unusual. The customer may be angry. The answer may live across three systems. The policy may not quite fit.

That is not Tier 1 with a nicer title. That is judgment-heavy work that requires expertise.

Now add the hiring problem. If you stop bringing in entry-level agents because you expect AI to absorb the work, you also shrink the population that would have become tomorrow’s escalation agents, subject-matter experts, trainers, quality leaders, and supervisors.

You can recruit people with that experience from the outside, of course. So can everyone else.

The contact center industry could end up competing for a smaller pool of experienced people while simultaneously reducing the number of places where people become experienced. That is a talent supply problem wearing an automation costume.

Build the New Apprenticeship Before You Remove the Old One

If Tier 1 volume declines, contact centers need to preserve the learning, not the inefficiency. That means asking a more useful question: what did people learn in Tier 1, and where will they learn it now?

• Map the hidden skills inside Tier 1. Customer-language fluency, system navigation, escalation judgment, documentation, emotional control, and pattern recognition are all being practiced even when the contact looks routine.
• Keep selected live-work rotations where they still add learning value. Not every contact needs to stay with a person, but some lower-risk interactions may still be useful as a controlled development step.
• Make simulation much more serious. McKinsey has described how GenAI-based contact center simulations can recreate realistic customer scenarios, provide real-time suggestions, and help agents practice before they enter live work. That is exactly the kind of capability the new apprenticeship model will need.
• Strengthen nesting and measure readiness, not attendance. A new agent should progress because they can demonstrate sound judgment, quality performance, appropriate escalation, and confidence using AI, not because they complete a certain number of classroom days.

AI Should Change the Career Ladder, Not Remove It

As I said at the beginning, the point is not to preserve every Tier 1 job. It is to preserve the development path that Tier 1 quietly provided.

AI can remove work that people should no longer have to do. It can help agents find answers faster, handle documentation, surface knowledge, and give customers immediate service when a human adds little value. That’s progress.

But the human role that remains will require more judgment, not less. So the development system must get better at the same time as the automation gets better.

The strongest workforce model will not be “AI instead of people.” It will be AI handling what it does well, humans handling what they do well, and a deliberate path that teaches people how to become the experts who will still be needed alongside the machine.

Technology can help rebuild parts of that apprenticeship. AI simulations can give agents realistic practice before live work, while shared virtual operating environments can keep trainers, supervisors, and experienced agents close during nesting and production.

Don’t Drain the Bench While the AI Agent Is Still Warming Up

The business case for AI is real. Nobody needs to defend keeping repetitive work just so humans have something to do. The danger is making workforce decisions on the assumption that an AI implementation will reach its planned scope, quality, and timeline exactly when the spreadsheet says it will.

Laivly’s 2026 AI Deployment Index, based on a survey of 200 CX leaders, found that 43% of recent contact center AI projects were delayed or stalled and 53% exceeded budget. CX Dive, reporting on the same research, highlighted an even more uncomfortable wrinkle: pressure to show AI savings can lead companies to reduce headcount before the technology has actually proven it can carry the workload.

Gartner has issued a similar warning. It predicts that by 2027, half of companies that cut customer service staff because of AI will rehire people to perform similar work, often under different job titles and with less experience.

This is where the implementation timeline matters. If an AI project takes six months longer than expected, the organization still has customers to serve during those six months. If leadership pauses hiring based on the original timeline, the gap lands on the people who remain. Overtime rises, experienced agents carry more of the difficult queue, and supervisors have less capacity to coach and train.

The safer approach is not to keep headcount forever “just in case.” Tie workforce reductions to proven operational performance, not the planned go-live date. A pilot that resolves a narrow set of cases is not the same thing as stable production at scale.

That does not mean leaders should slow down AI investment. It means they should be careful about cutting the runway before the plane is actually in the air.

Automate the task. Just make sure you do not accidentally automate away the career ladder that produces the people you will still need.

Guest post, written by: Jason Hiland, Chief Revenue Officer at CollaborationRoom.ai

CollaborationRoom.ai is a SaaS Virtual Floor platform that connects distributed, in-office, multi-site, and outsourced contact center teams in one live operating environment for operational visibility, training, nesting, coaching, and agent support.

To learn more about building a connected operating environment for training, nesting, and supporting agents, visit CollaborationRoom.ai.

The post When AI Takes the Easy CX Calls, Who Trains the Experts? appeared first on Execs In The Know.