Boards and CFOs are already asking for more. A recent Salesforce survey found that 61% of CFOs see AI agents as critical for competitiveness, and 74% expect them to deliver both savings and growth. That’s why concepts like cost per resolution are gaining ground.

Legacy CX metrics don’t capture all the dimensions. They miss the costs of recontact, the revenue lost through refund leakage, and the value of risk avoided. The next wave of Agentic AI ROI needs to measure all of that, alongside the Digital Labor TCO that leaders use to compare AI agents with human FTEs.

The Limitations of Traditional ROI Metrics

Most automation and agentic AI business cases still lean on familiar numbers: average handle time, cost per contact, or headcount savings. Those measures show efficiency, but they don’t show outcomes.

They leave out whether a customer’s problem was solved, whether loyalty improved, or whether churn dropped. They don’t account for Risk ROI either. An AI agent that processes a refund incorrectly or mishandles sensitive data doesn’t just create an unhappy customer -it creates reputational risk and potential compliance costs.

The habit of celebrating “deflection” adds to the problem. Deflecting a call without solving the issue only guarantees the customer will come back, often more frustrated. It’s a false saving that shows up quickly in recontact rates and refund leakage.

Analysts are warning that boards are asking for more. There’s a growing need for ROI measures that cover user satisfaction, decision quality, and organizational resilience, not just efficiency.

There are real-world examples of this gap. Vonage cut its customer response time from four days to four hours, a strong efficiency gain. But the more important measure is whether those faster responses improved resolution rates and customer loyalty. Without that lens, the ROI story is incomplete.

From Cost-Per-Contact to Cost-Per-Resolution: The Crucial Shift

Forget metrics based on the number of calls handled or agents saved. What really matters is whether a customer’s issue gets resolved, and how much it costs to do that. That’s the power of Cost per resolution (CPR): take every expense tied to resolution, agents, tools, tech, overhead – and divide it by the number of issues successfully closed. That gives a business meaningful insight into actual outcomes.

Why this matters so much now:

• Customers value having problems fixed, not just conversations logged.
• CFOs are watching metrics like re‑contact rates, refund leakage, and overall time‑to‑resolution. Those measures expose the real cost of chasing the same issue over and over.
• It unlocks smarter benchmarking and sharper ROI models, far beyond top‑line cost-per-contact comparisons.

Take the move some vendors are making toward resolution-based pricing. Companies like Ada are now charging for each resolved issue rather than every conversation, aligning incentives with the outcome businesses care about most

A public-sector proof point adds real weight. Barking & Dagenham Council saw their cost per enquiry drop from £4.60 to just 5p, reaching 533% ROI in six months, thanks to AI assistance that resolved most queries upfront. Beyond the direct savings, customer satisfaction surged, pushing the value narrative further.

It’s a simple truth: volume of conversation doesn’t equal resolution. One resolved interaction is worth far more than five half-handled ones. Cost per resolution aligns metrics with customer outcomes, and it’s the backbone of any credible Agentic AI ROI model.

Beyond Cost Per Resolution: The Scorecard for Agentic AI ROI

Cost per resolution is a crucial focus point for the new Agentic AI ROI scorecard, but it’s only one part of the puzzle. Here’s how to build a framework that CFOs and COOs would actually value, one that measures real business outcomes, manages AI risk, and proves why automation pays.

Step 1: Calculate Total Cost of Ownership (TCO)

Start by comparing digital labor TCO with human FTE costs. The cost equation includes more than you might think:

• Licensing fees, integration work, and LLM/token usage
• Tooling for orchestration, governance, and observability
• Security, compliance, and ongoing oversight

Those numbers give you a CFO-ready comparison: how automation stacks up against hiring, in both cost and control.

Step 2: Identify Tangible Benefits

Every ROI model needs clear financial impact signals:

• Fewer recontacts. That means less handling and smoother operations.
• Less refund leakage through first-time resolution.
• Faster time-to-resolution, which improves customer satisfaction.

Two examples show this in action:

• Zota used Agentforce to lift resolution rates noticeably, showing that faster doesn’t just mean cheaper, but better.
• NiCE and PSCU automated 71,000 financial transactions a year, delivering clear productivity gains.

Step 3: Include Intangible Benefits

There’s a growing case for measuring less tangible returns too:

• Avoided compliance violations. GDPR can carry fines of up to 4% of global turnover or €20 million, whichever is higher.
• Brand trust. Audit trails and observability show customers and boards that AI behavior is transparent.
• Organizational resilience. ROI should include decision-making speed and agility in a crisis.
• Employee value. Bots handling repetitive tasks mean higher job satisfaction and lower turnover.

These benefits don’t sit on the P&L, but they matter at the board level, particularly when combined with insights into metrics like cost per resolution.

Step 4: Embed Feedback Loops

Without data, there’s no optimization.

• Use observability dashboards like Salesforce Command Center and Scorebuddy to track agentic performance in real time.
• Run regular audits, using ROI calculators like Salesforce’s to validate your assumptions.
• Maintain control groups to measure true gains and risk exposure.

Also, ask for direct feedback from both employees and customers. How is agentic AI improving their experience, and where is it creating friction points?

Step 5: Remember the Governance and Observability Angle

Every conversation about automation eventually comes back to risk. When AI makes the wrong decision, the cost isn’t just operational; it can trigger compliance fines or cause serious damage to brand reputation. For finance and service leaders, that makes governance and observability non-negotiable.

The EU AI Act has raised the stakes. It requires organizations to show why AI systems made the choices they did, and to prove they are fair, safe, and explainable. That means audit trails, transparency logs, and clear escalation paths need to be part of every Agentic AI ROI calculation.

The business case is straightforward. A system that resolves tickets but cannot explain its decisions exposes the company to hidden liabilities. That exposure needs to be factored into the Digital Labor TCO. Compliance safeguards and observability tools may add cost, but they also reduce AI risk and protect against reputational fallout.

Case Studies: Proof of Resolution-Driven ROI

Real business value lives in real stories, showing that companies are moving beyond efficiency, to focus on tangible gains and growth opportunities.

Look at Oldenburgishce Landesbank, a privately owned regional bank that introduce AI-powered solutions into customer service. They achieved a 15% reduction in wait times (showing efficiency gains of 510%). However, they also achieved a 5 increase in Net Promoter Score.

Satellite leader Echostar saved team members more than 35,000 hours of work annually with AI and automation, but even more importantly, the company reduced the price of sales-related calls from $26 per hour to just $2.

With Ada, Neptune Flood reduced their ticket resolution times by 92%, minimized cost per ticket by 78%, and earned $100k in operational savings in the first year. On top of all that, they set the foundation for new workflow automation flows that are primed to add value to the business and unlock new avenues for revenue.

Cost Per Resolution and Agentic AI ROI

Customer service leaders can’t afford to measure automation in the same way they did a decade ago. Cost per resolution is quickly becoming the number that matters, because it shows whether issues are fixed properly the first time and what that costs the business. It links directly to loyalty, refund accuracy, and the true cost of keeping customers happy.

For finance leaders, the comparison between human labor and AI agents is no longer simple. Digital Labor TCO has to include not only licensing and model costs but also the spend on oversight, compliance, and monitoring. Those controls reduce AI risk and protect against the reputational damage that follows when customers lose trust.

Payback can still come quickly. Independent studies suggest most automation programs deliver returns in as little as 12 to 18 months. After that, the benefits usually grow as adoption rises and workflows mature. But the organizations seeing the strongest returns are those that build guardrails into their models from day one.

That’s the direction of travel for Agentic AI ROI. It’s not about shaving minutes off handle time. It’s about proving to boards and regulators that every resolution is reliable, efficient, and accountable. Companies that measure ROI through that lens will not only show faster payback but will also protect the trust that underpins their brand.

Christian Draeger, who runs passenger experience there, talks about it in a way that’s surprisingly down-to-earth. “We’re not just starting at the airport door,” he says. “We’re already looking at customers, how they can get prepared for their travel, even days ahead of the actual travel plans.”

That’s a different way of thinking about travel, one where the airport is part of the journey, not a pause in it. Draeger’s rule is simple: “Put the passenger in the center.”

That idea is becoming more important. Around two-thirds of travelers now use AI tools to plan their trips, and most say they want services that adjust to them, not the other way around. Berlin’s answer is to mix technology with empathy, using automation to remove hassle, not humanity, and turn the everyday airport routine into something that actually works for people.

Understanding & Designing for the Modern Traveler

Christian Draeger has spent a lifetime around airports. More than thirty years in aviation have given him a deep sense of how people move, wait, and connect. During his time with Star Alliance, he helped shape what millions of passengers now recognize as the modern travel experience. When he joined Berlin Brandenburg Airport, he came in ready to rethink that experience from the ground up.

Berlin handles around 25 million passengers a year, so it’s big enough to be busy, but small enough to still care. “We also operate our premium services: two business-class lounges and an ultra-premium lounge where you get à la carte dining and a chauffeur service to the aircraft,” Draeger said.

That same care for detail extends to the parts of the journey most people barely notice. The airport also took control of its own security operations, because, as Draeger puts it, “We felt that the mandate of the federal police didn’t provide enough attention towards the passenger experience.”

Now there are 32 security lanes, 24 fitted with advanced CT scanners, so passengers can keep laptops in their bags and carry small amounts of liquid without delay. “It’s about having a consistent experience across the whole area of the airport,” he says.

Every choice is made with the passenger in mind. “It starts really by knowing our customers,” Draeger says. “If we have a family that’s traveling once a year on holiday, their prerogatives are different from a business-class customer focused on getting through as efficiently as possible.”

That balance, efficiency for some, discovery for others, is at the heart of personalization in travel, and it’s essential. A recent study found that 93 percent of travelers now expect some form of tailored service. Berlin’s approach proves those numbers translate into real-world design decisions: better security flow, less queuing, and even duty-free areas reimagined as “specialized marketplaces.”

Dual-Terminal Strategy: Two Philosophies, One Vision

A walk through Berlin Brandenburg Airport reveals something a bit different. Its two terminals don’t just separate airlines; they reflect two completely distinct types of travelers. One is designed for comfort, the other for speed. Together, they show how personalization in travel can be built into the physical space, not only into digital systems.

“The level of automation that you will find with low-cost carriers is more in focus than with a legacy carrier,” says Draeger. Terminal 2 is the efficient, minimalist one: smaller, sharper, and geared toward travelers who value simplicity and price over perks. “Terminal 2 is geared to simplicity and generating additional revenues through add-on services,” he explains.

Think self-service kiosks, intuitive wayfinding, and a layout that helps people move quickly from curb to gate. “The utilization of busses is less, you have more walk boardings,” he adds.

Terminal 1, meanwhile, is a different rhythm altogether. “It’s about efficiency and comfort, both guided by digital tools.” Business and frequent flyers pass through airport automation that’s designed to make the process seamless. Over a hundred self-service kiosks are spread across the terminal, complemented by digital signage and premium lounges.

It’s the physical version of a digital truth: no two passengers want the same thing. Some want to breeze through with a coffee and a boarding pass on their phone. Others want time, space, and a glass of something cold before they fly. Both deserve an experience that feels intentional.

That’s what Berlin is building, a new kind of airport customer experience where infrastructure itself becomes a form of personalization. Different terminals, different tools, same philosophy: know who’s traveling, and design accordingly.

AI and Automation Enhancing Personalization in Travel

Like most airports, Berlin once relied on a traditional call center. It worked, but just barely. “We were looking at our call center and we weren’t completely happy,” says Draeger. “It was limited, inconsistent, and expensive.”

That frustration turned into an opportunity. Berlin decided to replace its call center entirely with a generative AI-powered system. The result was “Berry”, Berlin’s always-on virtual assistant.

“Customers can call the AI hotline and have a conversation just like we’re having right now,” Draeger says. It took just six weeks to build and launch, and within a few months, the results were striking: satisfaction above 85 percent, costs down 65 percent, and service available 24/7.

The human element didn’t vanish; it just found a new home. Instead of waiting in phone queues, travelers get answers right away. Lost something? Need flight details? Berry, the airport’s AI agent, takes care of it and loops in a person if the question needs a human touch. It’s simple to use too: one phone number on Berlin Airport’s website connects straight to Berry.

Building the AI Layer with Berry

Behind the scenes, Berry learns fast. “After six to eight weeks we reached an acceptable level… then you could see week-to-week improvements as GenAI learned,” Draeger explains. His team fed the system with real passenger questions and prioritized the most urgent topics first, like the classic “I left my laptop on the aircraft.” “We prioritized major customer concerns to ensure correct routing from day one,” he says.

Now the airport is preparing for the next step, chat. “We want to also offer the ability to get in touch with our AI agent through chat functionality,” says Draeger. QR codes will soon appear throughout the terminals, linking passengers directly to Berry via chat, integrated into the website and app. “If you’re standing in the arrivals hall, we’ll know based on the QR code where you are, and tailor the information accordingly.”

The idea is to build truly contextual assistance: a passenger in departures might ask about gate directions or restaurants, while someone at baggage reclaim could get help locating transport or lost luggage. “Customers can switch between voice and chat depending on environment or age. My children would prefer to talk; someone in a crowded terminal might prefer to chat,” Draeger says.

Operational AI and the Quest for Seamlessness

A lot of what makes Berlin Brandenburg Airport work isn’t something you can see. It happens on the tarmac between the terminal and the runway, where planes turn around for their next flight, and timing is everything.

“We also have others more on the ramp side,” says Draeger, referring to a system the airport now uses to track ground operations in real time. Cameras watch every stage of the turnaround, feeding data to an AI that predicts how long the process will take and where it might go wrong. “They can predict turnaround durations and steer additional resources if required,” he explains. “If a baggage belt is missing upon arrival, they can autonomously act on that and resolve bottlenecks.”

This is the kind of work that truly shapes the airport customer experience. When flights leave as scheduled, lines move faster, and connections fall into place without drama. Most travelers never think about the coordination behind it all. Yet every new piece of technology adds a layer that must fit perfectly with the rest.

But every new layer of technology brings its own challenge. “We always want to have this seamless experience for our customers,” Draeger says. “As we introduce more technology, we’ll have the challenge of combining it with legacy systems.”

Airports, after all, are built to last, and that means old baggage systems, decades-old software, and miles of wiring that can’t just be swapped overnight. “Traffic is increasing significantly, and we have limited infrastructure,” he adds. “We need simpler processes and better technology to absorb growth.”

Behind the polished front end of any airport automation project lies a balancing act: new tools talking to old systems, innovation working around concrete and cables.

The Future for Personalization In Travel: Digital Handholding

When asked what he thinks the future of travel looks like, Christian Draeger doesn’t mention drones or driverless terminals. He talks about something far simpler: help that is steady, thoughtful, and personal. “We always like to call it digital handholding,” he says. “A digital entity that’s completely informed, taking the customer by the hand and guiding them through the journey.”

Many agree that this is exactly where AI in the travel industry is heading. Gartner predicts that more than 80% of all customer interactions will be AI-assisted by 2029. The difference now is how personal that assistance can become.

“In the future, we see customers having their own personalized digital agents,” Draeger says, “on mobile, VR glasses, or other interfaces.” Those agents will be able to do a lot. “They’ll be able to rebook flights, change hotels, handle issues,” he explains. “We’ll need to provide them with the knowledge base and interconnectivity so they can act.”

He describes a world where these personal assistants talk to each other. “We’ll see a marketplace developing for agent-to-agent interaction,” he says, a network where your digital travel companion can speak directly to an airline, a hotel, or even the airport itself to smooth out the details before you notice them.

Some of that is already visible in small ways. Berlin is already imagining using augmented reality to help people find their way through the terminal. “If you come to Berlin Airport, sometimes you’ll find too many information boards,” Draeger admits. “Imagine augmented reality guiding you through the airport.”

It’s easy to see where this leads: toward an airport customer experience that blends technology with intuition. The idea isn’t to overwhelm passengers with data, but to take away the stress of travel entirely.

Personalization in Travel and Airports as Experience Ecosystems

Christian Draeger talks about air travel the way some people talk about music, not as noise and movement, but as rhythm. Airports, he says, are meant to keep that rhythm steady. When they do, everything else feels effortless.

“It’s all about making travel easier,” he says. “Like when you take a train, you just arrive and go, that’s the overarching ambition.”

Mostly, Berlin Brandenburg Airport is just pushing for a calmer travel experience. From the moment a traveler checks in to the moment they leave the gate, the goal is to take away the small frictions that make airports stressful. Berry, the AI voice agent, is part of that. So are the self-service kiosks, the CT-scan security lanes, and the quiet bits of software that keep aircraft turning on time.

“It’s not about one technology: Gen AI, robotics, biometrics, or AR,” Draeger says. “It’s about combining them to make travel much simpler.”

That line sums up Berlin’s whole approach to personalization in travel. It isn’t about showing off what technology can do; it’s about how little the traveler has to notice it.

That’s the real future of airport customer experience: an ecosystem that looks complicated underneath but feels beautifully ordinary on the surface, the kind of simplicity only achieved when someone’s been obsessing over every detail on your behalf.

The agency laid off about 1,800 call center employees in May and June. But it has since brought back roughly 160 and extended others’ contracts, as the issue was escalated to Canada’s Auditor General.

The CRA said it was answering 77 percent of calls by late September, having set a target of 70 percent by mid-October. The agency handles massive call volumes — more than 32 million per year, peaking at 300,000 daily during tax season — making it one of the most heavily used government contact centers in the country.

Since implementing the plan in early September, the CRA has rolled out several self-service improvements aimed at making it easier for Canadians and businesses to access help without having to call its contact centers.

These include extended online chat hours from 08:00 to 20:00 ET, simplifying its website and launching new self-serve options in its digital accounts. The CRA has also reported growing use of its chat tool, with more than 5,400 users between October 6 and 10. An expanded AI-powered chatbot is slated for release in early November to answer a wider range of questions and further reduce call volumes.

The agency also highlighted a national Quality Monitoring Program launched in 2024 that reviews more than 100,000 call recordings annually to improve accuracy and training.

The CRA received a government investment of $400MN in 2022, to support its call center operations in anticipation of call volumes remaining above pre-pandemic levels. This was intended to enable CRA to maintain a service standard of answering 65 percent of calls within 15 minutes of a caller opting to speak with an agent. That standard has been lowered from 80 percent in 2017.

But many Canadians report seeing little improvement in the service, spending hours on hold when trying to call the agency.

As part of its efforts, the CRA added a feature to its website telling taxpayers how long they can expect to wait before they get through to a service agent when they call. But test calls by CBC News at different times and days found its estimates to be wildly inaccurate, a gap that’s eroding public trust. Unlike many private-sector call centers, the CRA still doesn’t consistently offer a callback feature, leaving callers to wait on hold for as long as it takes.

The Minister of Finance and National Revenue and the Secretary of State directed the CRA to implement the plan in early September to strengthen services, improve access and reduce delays. The focus is on four key areas: “increasing call centre capacity, expanding online self-service options, tackling the root cause of service issues, and accelerating service modernization.”

Taxpayers’ Ombudsperson François Boileau said that while the CRA has made progress, he’s concerned about what happens next.

“With some processing delays far exceeding the CRA’s usual service standards, it is unlikely that the CRA will reduce the backlog to a sustainable level by the end of the 100-day period.”

“A longer-term commitment and adequate resources will be necessary. By reducing its processing delays, the CRA could reduce the number of calls it receives and reduce wait times for taxpayers.”

Boileau noted that a report released by the Auditor General this week, found a direct link between staffing levels and service performance, a familiar story for any large contact center. He also questioned what’s driving such high call volumes in the first place, suggesting that better digital design and faster case resolution could reduce the need for people to call at all.

Unsurprisingly, the Union of Taxation Employees (UTE) echoed those comments, calling for the CRA to immediately increase its contact center staffing, as well as the training and support it provides its agents. In August, the union launched a national “Canada on Hold” campaign denouncing the job cuts.

The Auditor General’s report shows that staff shortages, a lack of training and pressure on contact center employees are compromising the quality of the service the agency provides to the public, the union stated. Marc Brière, UTE National President, said:

“Our members, who work tirelessly, are often exhausted, overworked, and under pressure to respond to as many calls as possible as quickly as possible, to the detriment of quality service.”

Why Automation Alone Can’t Fix Contact Center Challenges

The CRA’s experience is a textbook case in what happens when contact center staffing, digital experience, and operational design fall out of sync. Even with new automation tools and AI chatbots, poor workforce planning and outdated processes can quickly undermine customer satisfaction.

The agency’s troubles may sound familiar to the UK’s HM Revenue & Customs (HMRC), which has wrestled with a similar dilemma — how to shift customers toward digital self-service without alienating them in the process. HMRC has been pushing taxpayers to use digital channels such as chatbots and online forms to reduce call volumes. But when the tax authority announced plans to close its self-assessment tax helpline for half the year, and drastically cut back others, it sparked an immediate public outcry. Within 24 hours, HMRC reversed its decision, conceding that it had misjudged how far it could push customers toward digital-only interactions.

For enterprises in any sector, self-service and digital tools can only succeed when supported by the right human infrastructure. Contact centers need skilled, adequately staffed teams who can handle complex interactions that automation cannot. And, as the CRA’s situation shows, service improvement needs to be treated as an ongoing commitment, not a one-time campaign.

As the 100-day plan wraps up, the CRA faces growing pressure to tackle the systemic issues that continue to generate high call volumes.

Boileau noted, “the CRA is measuring its performance before the peak volumes it normally experiences during tax season.”

“What will happen to the progress it has made after the 100-day plan ends and tax season begins? What will happen when the calls increase? Will the contact centers still have the resources to answer the calls?”

Many of its service users will be watching to see if its attempt to improve service and reduce wait times leads to lasting change.

The cloud computing service suite and flexible payment provider have entered an AI-first partnership, aiming to improve Klarna’s customer experience with Google Cloud’s customer-centric products. 

Klarna is already reporting dramatic growth with early AI pilot testing, with a 50% increase in customer orders and a 15% increase in average time on the app. 

David Sandström, Chief Marketing Officer at Klarna, claimed that combining Google’s AI models with Klarna’s consumer insights was allowing the buy-now pay-later specialist to “craft experiences that feel smarter and more personal.  

Early pilots already show the potential: AI-driven creative concepts, from dynamic digital ‘lookbooks’ to hyper-personalized product campaigns, boosted time spent in our app by 15% and increased orders by 50%. 

Interestingly, the collaboration follows Klarna’s recent decision to bolster its customer service headcount by redeploying staff from other areas of the business.

At the time, Sebastian Siemiatkowski, CEO of Klarna, seemingly backtracked on the staunch pro-AI in customer service views he had previously espoused, admitting that the human touch was more critical in customer service than he had first thought.

While this new strategy appears to be flip-flopping back to championing AI, the company has confirmed that the partnership will initially focus on empowering Klarna’s current teams to transform the consumer experience.  

Via the Klarna App, the partnership will be concentrating on two specific areas for its customers (outlined below), which are designed to deliver a richer, more personalized shopping experiences and engaging content. 

Creative Velocity

As part of the partnership, Klarna is leveraging Google’s latest generative media models, including Veo 2 and Gemini 2.5 Flash Image (Nano Banana). 

These models are being used to create dynamic digital “lookbooks”, an automatically generated shopping gallery personalized for consumers on the Klarna app.

These “lookbooks” tailor shopping interests to each customer, designed to adapt to trends and individuals’ clothing interests.

Klarna is also employing Google’s AI models to assist in the creation of their hyper-personalized marketing campaigns for its users. 

Personalization and Beautification

Klarna is aiming to enhance its current customer service offerings by targeting its extensive library, home to more than 200 million images. 

This will be achieved by regenerating and refining its visuals, ensuring that every shopper encounters a higher quality of engagement and visual content on the app. 

Marianne Janik, Vice-President, EMEA North at Google Cloud, highlights how Google Cloud will help transform Klarna’s customer service.

“To lead in this new AI era, businesses require more than tools – they need strategic capabilities. Our partnership with Klarna is about providing just that,” she said.

Our integrated, AI-optimized platform and cutting-edge models are enabling Klarna to unlock significant creative velocity and drive innovation. 

“We’re proud to help Klarna not just adopt AI, but also use it to fundamentally redefine the customer experience.” 

Security is Key

Outside of the lookbooks and revamped images, the partnership will also allow Klarna to strengthen its security by installing Google Cloud’s AI hardware and expertise. 

This will be used to train and deploy artificial Graph Neural Networks (GNN) to tackle instances of fraud or money laundering on its platform. 

The learning models have been designed to analyze complex relationships between users, transactions, and devices, in order to detect any anomalies or suspicious patterns at a higher accuracy level. 

This security extension ensures secure protection as Klarna continues to improve its creativity on next-generation products.

The pair explore how the free, 10-minute assessment helps organizations understand their readiness for contact center AI and create a clear roadmap.

In doing so, they break down:

What Is the AI Maturity Assessment, and What Problems Does It Solve?

Kazmi explains how the tool removes uncertainty around AI adoption, helping businesses identify their current stage in a five-step maturity model.

How Does the Assessment Work?

The framework is based around those five steps Kazmi articulated:

1. Interaction channels and technologies
2. Data utilization and intelligence
3. Agent capabilities and support
4. Response and resolution effectiveness
5. Strategic approach and innovation

In the interview, he breaks each down further.

What Do Businesses Receive?

Participants instantly gain access to:

A personalized results dashboard with scores across all five dimensions
A clear AI maturity stage and tailored, non-promotional recommendations
An eBook with insights from nearly a year of research

How to Access the Assessment

Ready to try the assessment? Find it here: https://www.ringcentral.com/ai-cx-maturity-assessment.html

In an era where smartphones can stream HD video, scan barcodes, and zoom in on the tiniest detail, it’s strange to still expect people to read out serial numbers or try to describe exactly where their coffee machine is leaking.

That’s where a multimodal approach can help bridge the gap between explanation and understanding.

In a nutshell, multimodal combines voice, visuals, and messaging to make it easier for customers and agents to actually get to a fix.

“When you’re supporting physical products, voice alone is like standing next to someone but with your eyes closed,” says Gintautas Miliauskas, CEO and Co-Founder of Mavenoid.

“You can’t see what they’re doing, you can’t point to anything, and they can’t show you what they’re talking about.”

Why Multimodal Matters Now

The aim of multimodal isn’t to ditch voice, far from it. The aim is to make voice part of a richer customer experience toolkit.

Customers should be able to talk, type, snap a quick photo, or open a diagram – all in the same conversation, without feeling like they’ve been shunted into a completely different system.

This is especially critical for physical products, where the stakes can be higher, as Miliauskas  explains:

“When a customer is working on their device, there’s more risk.

“If we give the wrong instructions, we could cause harm to the user or damage the device. For us, multimodal isn’t optional; it’s essential.”

A real-life example of multimodal in action involves US home ventilation brand Broan-NuTone.

When they switched to Mavenoid’s multimodal platform, service levels jumped from 43% to 80%.

Indeed, the platform helped the company achieve higher first-contact resolution, fewer mistakes, and far less back-and-forth between customers and agents.

Beyond the Obvious Metrics

While the improvements discussed above, such as first-contact resolution, average handle time, and escalation rates, are impressive, multimodal can have an impact beyond these traditional metrics.

The technology delivers on less tangible, but equally important, fronts, including better agent accuracy and higher customer engagement.

“With multimodal, across our customer base, we see engagement above 85%,” Miliauskas says.

“When you send someone a diagram or a warranty form instead of spelling things out, compliance is naturally higher.”

Moreover, Miliauskas details how multimodal can be very effective at removing tiny frictions:

“If someone says ‘serial number SN800Z’ over the phone and it’s misheard, you end up repeating yourself.

“When it’s entered on-screen, that friction disappears.”

When it comes to Broan-NuTone, there were several of these ‘non-metric wins’.

Most notably, the improvements to spare parts identification.

With hundreds of SKUs, customers often bypassed online tools and called straight in, which didn’t always speed things up.

“By automating this process in the voice channel, we could send them a link to the exact spare part,” explains Miliauskas

“They could buy it instantly without waiting for an email or typing a long URL.”

From there, the company expanded into warranty claims, installation guides, and onboarding.

Matching that resolution capacity with human agents alone would have meant hiring an extra eight full-time employees.

Instead, the business was able to improve service while keeping costs down: the CX professional’s dream.

Keeping Implementation Simple

Like anything in life, taking the first step is hard. For many people, their brain immediately races towards the worst possible scenario.

Yet, more often than not, in reality, there was no need to panic.

This is especially true in the CX and customer service space, where businesses and professionals frequently assume that new implementations are painful, drawn-out, and expensive processes.

But multimodal doesn’t have to mean a sprawling IT project.

Miliauskas says Mavenoid’s philosophy is to “use the simplest tech that solves your problem.

“Not everything needs the fanciest AI model. We have ready-made solutions that can be configured in 15 minutes.”

The vendor also offers low-code and no-code options that help non-technical teams make updates as needed.

“We reuse existing content from the digital channel, so you don’t have to create everything twice,” Miliauskas explains.

In addition, connected devices benefit from the ability to tap into the device’s data stream, which sometimes allows users to understand the problem before the customer even calls.

Looking Ahead

You may have already picked up on this, but Miliauskas is a strong advocate for the power of multimodal.

The CEO believes the technology will quickly become the default for physical-product support – and sees Mavenoid’s specialization here as a key strength.

“Most automation platforms are generalists,” he explains.

“We’ve always focused on physical products, and that’s where multimodal really shines.”

“Think of it like a trip to the doctor: you come in with a symptom, they ask the right questions, and then prescribe the right treatment. Mavenoid does the same by deciding whether the answer is voice, multimodal, or a human agent.”

For CX leaders, the takeaway is simple: in a world where speed, accuracy, and low effort define the experience, sticking to a single channel is like trying to fix a modern problem with Victorian tools.

Multimodal isn’t just an upgrade; it could very well become the baseline.

You can find out more about Mavenoid and its full suite of solutions and services by visiting the website today.

You can also read about some of the other ways that Mavenoid is deploying voice automation by reading this article.

That frenzy ensued after Benioff’s appearance on The Logan Bartlett Show.

During the conversation, the CEO explained how Salesforce’s AI agents now handle 1.5MN of the company’s incoming customer queries.

At the same time, human support agents handled approximately the same number of conversations, with “nearly identical” customer satisfaction scores. Benioff continued:

I was able to rebalance my headcount on my support. I’ve reduced it from 9,000 heads to about 5,000, because I need less heads.

Since the podcast aired, Salesforce shared a statement with journalists, claiming to have successfully redeployed hundreds of employees into other areas like professional services, sales, and customer success.

However, some industry observers have been quick to question not only the ethics but the details of Benioff’s account.

Indeed, one former Salesforce employee accused Benioff of sharing an outright lie, claiming Salesforce’s AI is “not ready for primetime” in a viral LinkedIn post.

Others have referenced how its market rival ServiceNow had kept its headcount steady, despite implementing AI agents across customer support and deflecting 75 percent of queries.

Meanwhile, some have predicted that Salesforce may suffer the same fate as Klarna. It leaned too heavily on AI customer service, only to then reportedly draft in employees from marketing, engineering, and legal to answer the phone.

Against this backdrop, analysts pressed Benioff on the claim during his company’s latest earnings call. The CEO responded by stating:

It’s hard for everybody to get their head around what’s possible.

“Salesforce has the opportunity to… reduce everybody’s support cost, to make everyone’s sales organization a lot more productive, to make everyone’s marketing have a much higher ROI, to make every field service technician a superman or superwoman, and to make every Slack user far more empowered in their organization than ever before, and I could go on and on and on.”

Benioff also isolated three reasons market rivals may struggle to emulate Salesforce’s results.

The first is timing. Indeed, the CEO claimed that the industry hasn’t matched Salesforce’s rapid innovation curve, with the CRM leader releasing Agentforce 3 little more than nine months after the platform’s debut.

Second is noise. Here, Benioff stated: “There are very smart people in our industry and other executives who are saying absolute nonsense.” In other words, don’t always trust what’s on LinkedIn.

Finally, the Salesforce man tagged “fear”. That’s understandable, given how intimidating significant change can be and the task vendors face around AI education.

Nevertheless, despite facing that fear, Salesforce is seemingly starting to overcome it, as Agentforce adoption ramps up.

More Big News from Salesforce

12,500+ companies are now building on Agentforce, up from 8,000 as announced during Salesforce’s previous earnings call three months ago.

Salesforce execs credited the company’s strategy of delivering industry- and department-specific Agentforce offerings as a critical driver.

These equip businesses with ready-made components to build AI agents that align with their needs, lowering the barrier to entry and building confidence in the technology.

Indeed, that strategy is seemingly working with Salesforce reporting a 60 percent increase quarter-over-quarter in customers moving from pilot to production.

Alongside all the Agentforce Chatter, Salesforce looked ahead to October’s Dreamforce event, where it will officially enter the ITSM (IT service management) space.

Reaffirming the move on the earnings call, Benioff stated:

A lot of our existing customers have been asking for this, [and] we’re bringing a whole new level of capability.

Finally, in regards to Salesforce’s actual earnings, it enjoyed a ten percent year-over-year (YoY) increase in revenue growth, reaching $10.25BN for the quarter.

If it passes, the bill would serve two primary purposes:

1. Ensure customers always have the option to speak with a human support agent.
2. Keep US contact center jobs in the country.

Initially, many hailed the bill as it gives consumers more transparency and choice.

However, industry analysts are now warning that the bill, while well-intentioned, could create more problems than it solves.

A Closer Look at the Keep Call Centers in America Act of 2025

Under the new bill, customer service agents must disclose their location and use of AI to customers at the beginning of an interaction.

If their location is outside the US, the regulation would force the company to transfer the contact to a US-based rep upon request.

The bill would also compel businesses with 50+ employees to notify the Department of Labor (DOL) at least 120 days before moving contact center work overseas.

The DOL will then maintain a public list of employers that offshore contact center work, keeping them listed for five years unless jobs return or contracts change to keep work in the US.

All businesses on the list will become ineligible for new federal grants and loans, with potential penalties and cancellations applying to existing awards if companies remain listed.

Additionally, it will require federal agencies to prioritize US-based employers not on the list when awarding contracts, and mandate that all federal contract contact center work happens in the US.

Lastly, the bill will demand the DOL to report on federal call center work, including job losses tied to the use of AI.

Everything That’s Wrong with the Bill

Customer experience analysts have highlighted several concerns swirling the Keep Call Centers in America Act of 2025.

After distilling these, here are three core problems that the bill could introduce for regulators, businesses, and consumers.

Problem #1 – It Doesn’t Seem Enforceable

Here’s a story to set the scene. “Yesterday, I kept getting calls from a number with a 747 area code, which is local here in the San Fernando Valley,” said Liz Miller, VP & Principal Analyst at Constellation Research. “They called me four times.

“So, I finally looked up the number. Turns out it was a Twilio number. With a reverse lookup, I discovered it was routed through Sudan, Kansas, then relayed from the Philippines, before hitting me in Los Angeles with a 747 area code,” continued Miller.

Think about that: the contact center was clearly in the Philippines, but because the company had US headquarters in Sudan, Kansas, the call showed as local. Under this legislation, that company could still claim: “We’re in America.”

As a result, the legislation may not accomplish what it’s supposed to and perhaps shows that lawmakers don’t fully understand what they’re regulating from a technical perspective.

Problem #2 – It May Hurt Businesses Trying to Improve Service Experiences

Consider Miller’s example. The bad actors? Those who spam customers incessantly, they won’t care. Instead, they’ll just keep spoofing calls through Kansas.

Instead, the Act will make well-intentioned companies with solid practices and customer experience strategies nervous.

Of course, some businesses have tried to hide their AI use or over-automate, making it almost impossible to reach a human.

However, as Justin Robbins, Founder & Principal Analyst at Metric Sherpa, told CX Today: “Most businesses I talk to don’t see AI as full automation, they see it as augmenting human capacity.

The lesson here is simple: people want transparency, trust, and fast resolution. They don’t want complexity and cost, which will be the outcome if this goes too far.

Problem #3 – It Does Little to Address Real Consumer Issues

The bill could challenge “doom loops”, where customers interact with AI or IVR mazes without exit routes.

However, with more consumers being pushed into a US call queue, without AI to help lessen the load, wait times may surge. That could frustrate customers, put more pressure on human agents, and add to a business’s cost sheet.

The result? Those extra costs may pass on to the consumer, too.

Additionally, while the bill may help escalate to a human, it does little else to tackle the rising issue of “customer sludge”, which typically frustrates customers more than the location of who they’re talking to.

Customer sludge includes all the blockers some brands put in front of customers to prevent them from making moves they deem unfavorable, like cancelling their subscription.

Do We Really Need to Mandate Human Support?

While some may view the bill as an extension of pro-American sentiment, the notion of safeguarding human contact center agents’ jobs is heartwarming.

Indeed, with prominent AI thought leaders, like Sam Altman, CEO of OpenAI, recently predicting the demise of human customer service, the move seems positive for industry onlookers.

However, many others have challenged Altman’s prediction. For example, Gartner recently suggested that half of businesses will abandon plans to lower their contact center headcount by 2027.

Meanwhile, Zeus Kerravala, Principal Analyst at ZK Research, warned against over-regulation.

“Historically, things usually work themselves out,” he said. “For example, with the shift to online banking, people initially said, “I’ll never use a machine to deposit checks.” But they eventually did, while still having the option to go into a bank.

“Some banks like Juniper or Venmo initially had no live agents, assuming their demographic wouldn’t need them… But consumer demand eventually brought those roles back.

So yes, there’s a lot of pro-American sentiment right now. But forcing agents to disclose they’re offshore could create hostility toward brands, even if offshoring is the right business decision.

Given this, while the bill may be nice in spirit, it could result in more complexity and cost than it’s potentially worth.

Miller, Robbins, and Kerravala discussed the Act as part of CX Today’s latest Big News Update video. Joined by two other expert analysts, they also gave their takes on the recent Verint takeover and Salesforce’s acquisition streak. Make sure you don’t miss the video’s release. Subscribe to the CX Today Newsletter. 

Like similar market studies, it features a vendor evaluation, taking 16 prominent conversational AI providers and dissecting their solutions, strategy, and services.

Ultimately, that informs the title graphic, which typically attracts most attention.

Yet, the evaluation is just one part of what’s included. The report also unpacks key market trends and considerations for selecting the best-placed vendor.

Ultimately, that sets the stage for the top takeaways below. Yet, let’s first reflect on the evaluation and how Opus placed the vendors on the matrix as it did.

The Opus Research Intelliview Methodology

The Opus Research Intelliview includes 16 global conversational AI players. Each had the opportunity to share details on their products and strategies. From there, the analysts conducted research and observed demos.

As they did so, the analysts focused on each vendor’s product offering, integrations & analytics, safety & trust posture, deployment & maintenance options, and strategy & execution.

That helped Opus formulate scores, while it gave extra credit for unique capabilities and vision.

From there, it placed providers on the following matrix, considering two fundamental questions: How complete is the product? And how strong is the strategy?

However, the research firm stressed that this captures only a snapshot in time. Ultimately, the Intelliview underscores which vendors are doing interesting, differentiated work, without pretending there’s a single “right” way to win in this fast-moving space.

Given that background, here’s how each vendor performed.

Top Takeaways from the Opus Research Intelliview 2025

As noted above, the Opus Research Intelliview not only interrogates prominent conversational AI vendors on their self-service solutions but also offers market commentary and unpacks trends.

The six top takeaways below offer an overview of the study’s major talking points, including extra insight from one of its authors, Ian Jacobs, VP and Lead Analyst at Opus Research.

1. Conversational AI Vendors Are Building Solutions for Two Distinct Buyer Groups

Consider the current customer service landscape. On one end of the spectrum, there is human support, which is relatively high cost but can handle nuance, emotion, empathy, and complexity. On the other, there are entirely scripted, inexpensive chatbots that can’t manage nuance or complexity.

Large language models (LLMs) and generative AI are emerging as the bridge between those extremes, and all three models are still present in the market.

In that dynamic, Jacobs suggests that conversational AI vendors face one massive strategy question:

Should they cater to those ready to go all-in with LLMs and agentic AI, or to those who need incremental steps toward that future?

That divide between all-in and incremental adoption is perhaps the biggest split in the vendor landscape today.

Both approaches are valid, targeting different audiences. Yet, those taking the slower route are often depicted as behind in technology, when that’s not necessarily the case. They simply focus on helping customers start small.

For instance, they might take a five-year-old scripted chatbot and add generative capabilities to the interaction layer, while keeping back-end processes locked down. That’s not massively transformative, but it can deliver significant efficiency gains and build confidence in the tech.

2. Tech Leaders Are Accelerating the Transformation of No-Code Builders

Traditionally, conversational AI vendors emphasized “no-code” flow builders, where businesses plot a step-by-step process for the virtual agent to follow.

These builders were predictable but inflexible, and often time-consuming to master.

Leaders today are offering more flexibility, allowing businesses to mix and match. For example, they may design a locked-down process with a free-flowing dialogue, or vice versa. This allows businesses to tailor the self-service solution to specific intents.

After making this point, Jacobs added:

Some are moving away from traditional flow builders altogether, toward models where you define requirements through prompting, and the system builds the flow in the background.

3. Some Vendors Go Further on Trust & Safety Than Others

Damaging hallucinations, data leaks, and regulatory violations are major concerns when it comes to safeguarding self-service applications.

Some vendors go further here than others. Jacobs suggests leaders excel at:

• Simulation testing with generative AI and automated test case generation.
• Built-in evaluation metrics.
• Inbound protection (e.g., prompt injection detection).
• Outbound monitoring (e.g., preventing harmful responses, hallucinations, or brand guideline violations).
• Continuous monitoring after launch, with alerts for failures, latency spikes, or missed capabilities.

4. Usability and Pricing Models Rise In Self-Service Buying Conversations

Across the conversational AI space, there’s a shift from tools built for technical experts to those designed for business users.

That doesn’t remove the need for IT, governance, and legal teams. Still, it empowers non-technical leaders — for example, contact center managers — to test changes directly in a sandbox without long IT turnaround times.

As those managers become more savvy to the finer details of self-service, the capabilities available to less technical users are a rising consideration for many brands.

That’s alongside pricing models. Currently, there’s no single “best” approach here. However, as Jacobs stated: “Leading vendors can clearly explain why their model benefits target customers and aligns with their needs.”

5. Conversational AI & Conversational Intelligence Go Hand-In-Hand

Jacobs envisions a future where the conversational AI world – understanding what a customer wants, responding, and taking action – isn’t separate from the conversation intelligence world, which analyzes customer conversations individually and in aggregate.

Think of conversational intelligence wrapping around and dissecting insights from all conversations, including those led by humans, AI, and both.

As this happens, a bot interaction that escalates to a human is analyzed holistically, with the contact center able to unpack what happened at that breaking point, the change in sentiment, and other critical intelligence.

In this sense, it’s similar to when the contact center space moved from multichannel to cross-channel to omnichannel service: it’s breaking down silos to create a complete view of the customer journey.

However, some conversational AI vendors Opus Research evaluated are not yet building for this future and unifying analysis across human and AI-driven conversations. Buyers should consider this when assessing possible self-service providers.

6. “Agentic” Is Here Today, Autonomy & Self-Reflection Are for Tomorrow

While not all conversational AI vendors use the term “agentic”, many are building capabilities that fit under the banner.

Today, “agentic” usually means prompt-based bot creation, but many other forms will emerge.

That’s according to Jacobs, who shared two examples:

1. Self-Reflection — The virtual agent evaluates its own performance and adjusts behavior without human approval once trust is established.
2. Autonomy and Optimization — The bot prioritizes based on brand goals, whether that’s cost, speed, accuracy, or even energy usage, and optimizes its actions accordingly.

“We’re not there yet, but the building blocks exist,” added Jacobs. “These aren’t far-future concepts; they’re logical next steps.”

Critically, they also align with the broader definition of “agentic”, which is about granting agency to automated systems.

Interested in learning more from Jacobs and the Opus Research team? If so, check out the article: What Should You Look for in a Contact Center Virtual Agent?

The research firm surveyed almost 6000 customers, who reported that when agents do mention self-service, 25 percent of their comments are neutral.

More worrying still, the research found that 12 percent of agents actually make “explicitly negative remarks.”

This is especially concerning for contact centers looking to invest in and grow their self-service offerings, as Gartner reports that customers are twice as likely to use self-service in the future if an agent promotes it during an interaction.

When discussing the findings, Keith McIntosh, Senior Principal of Research in the Gartner Customer Service and Support Practice, highlighted some of the benefits of self-service for businesses and customers.

“Promoting self-service is not just about reducing costs; it’s about empowering customers to use the easiest and most efficient solution,” he said.

Agents play a crucial role in this process, and their ability to positively endorse self-service options can really matter.

The findings and McIntosh’s comments underscore the importance of training agents to clearly explain the benefits of self-service tools.

But why exactly are agents currently failing to promote self-service?

Is Self-Service Still Struggling to Deliver?

There are many reasons that agents might refrain from recommending self-service channels.

The high workloads and rise in agent burnout have been well documented, as have agent concerns about AI-powered solutions replacing them.

However, the main reason might be more practical: current self-service channels simply aren’t up to scratch.

A separate Gartner survey from last summer found that only 14 percent of customer service issues are fully resolved through self-service channels.

Even for “very simple” problems, self-service success rates only reach 36 percent, highlighting a significant gap in effectiveness.

These findings suggest that current self-service tools are inefficient at resolving queries, a far cry from the time-saving, agent-workload-reducing solutions they proport to be.

While Gartner does not provide any information on this, it is not difficult to imagine scenarios where agents are having to deal with customers following a frustrating experience with a self-service channel, making the interaction more difficult and the agent less likely to recommend self-service to future customers.

The shortcomings of self-service come further into view when compared to phone channels.

Gartner made this point, as the research firm suggested that positive phone experiences may disincentivize customers from using self-service tools.

Indeed, only 35 percent of customers who last used the phone for support say they’d be willing to adopt self-service solutions.

With the phone still the leading service channel, its reliability and comfort may be hindering GenAI and self-service adoption.

In order to encourage customers and agents to better embrace self-service, McIntosh believes that service leaders “should focus on integrating GenAI solutions that complement existing phone interactions rather than replacing them.

“By positioning GenAI as an enhancement to the phone experience, organizations can reassure customers that the digital assistant is designed to streamline their journey, offering both self-service solutions and seamless transitions to human agents when needed.

Meeting customers where they are, while offering innovative solutions, is key to driving adoption and satisfaction.

Vendors Continue to Champion Self-Service

Despite the clear disconnect between customers, agents, and self-service, vendors continue investing heavily in the tech.

Earlier this month at Customer Contact Week 2025, Salesforce unveiled its new agentic self-service solution.

Integrated into Service Cloud, the tool leverages Agentforce – Salesforce’s AI agent platform – to deliver faster, more personalized support through a real-time customer portal.

On its release, Salesforce promised “a new era for self-service”, which could eradicate many of the issues isolated above, if the new era comes to fruition. However, that’s a big “if”.

For more insights from leading thinkers across the customer experience space, subscribe to the CX Today newsletter.