Threll.ai
Threll.ai whitepaper

Voice-based AI: The next major shift in customer dialogue

From Customer Service to Operational Growth Capacity

BySveinung Totland16 min read
Two hands hold a tablet showing the cover of the Threll.ai report "Stemmebasert KI: Det neste store skiftet i kundedialog"

Executive Summary

Most businesses still think too narrowly about customer dialogue. They think support, queues, response time, staffing, and costs. That understanding is no longer sufficient.

Voice-based AI is becoming a new operational layer between business and customer – not just because companies can respond faster or automate more, but because voice-based AI can now be integrated with language understanding, decision-making, workflows, and execution in the same flow. Gartner reported in December 2024 that 85 percent of customer service leaders would explore or pilot customer-facing conversational GenAI in 2025. Salesforce reported in November 2025 that AI is expected to handle 50 percent of service cases by 2027, up from 30 percent in 2025. McKinsey also points to a growing adoption gap in customer care, while Deloitte describes service as a real AI tipping point.

Four main points stand out:

1. The value exceeds the cost cut. Voice-based AI can impact availability, capacity, quality, and growth simultaneously.

2. The great opportunity is proactive customer dialogue. Voice-based AI is not just inbound service. It opens up for follow-up, loyalty work, upselling, and lifecycle dialogue.

3. Quality can be built into the model. With context, operational memory, and continuous improvement, businesses can make customer dialogue less person-dependent and more cumulative.

4. The Nordics require more than generic setups. Language precision, trust, data control, and operational execution are not additional requirements – they are prerequisites for voice-based AI to operate effectively in production.

This shift is not primarily about assigning a voice to a model. It's about building modern customer dialogues: conversations that convert.

1. Customer dialogue is about to become a new operational layer

Most businesses still think of customer dialogue in silos. Some are handled reactively in customer service. Some are handled proactively in sales, marketing, or follow-up. But too few organize this as a coherent operational model.

It is about to become a strategic mistake.

When voice connects to understanding, decision-making, workflows, and action, customer dialogue becomes more than just a contact point. It becomes an operational layer between the business and the customer. McKinsey describes a growing gap between businesses that actually derive value from AI in customer dialogue and those that remain in pilot mode.

This is more than just a new channel and more than a new feature. It is a shift in how customer dialogue can be organized as capacity – from a support function to operational capability for quality, growth, and competitiveness.

2. Why this is happening now

This shift isn't happening in a distant future. It is happening now. Three trends are finally converging: the technology is mature enough, management pressure has become real, and customer service is transitioning from a support function to a strategic value driver.

Technological Maturity. Voice-based AI has advanced far beyond traditional IVRs and rigid script engines. Conversations can now be better understood, managed, and continued without the experience collapsing at the first deviation.

Business Pressure. Gartner found in February 2026 that 91 percent of customer service and support leaders experience pressure from management to implement AI. The priorities are largely about improving customer satisfaction, operational efficiency, and success in self-service – not just cutting costs.

The New Role of the Customer Service Function. Capgemini describes customer service as an area that is being elevated from a support function to a strategic value driver, driven by generative and agentic AI. In Capgemini's survey, 61 percent of leaders say that customer service is currently primarily a support function – but only 22 percent expect that to be the case in three years.

Voice-based AI should therefore no longer be treated as an experiment. It is becoming a management issue.

3. Where the value actually occurs

For leaders, the main question is not whether voice-based AI can reduce costs. The main question is which value drivers can be impacted simultaneously.

The value typically occurs in four layers:

Availability is about being there when the customer actually needs you – not just during business hours.

Capacity is about handling multiple conversations without linear growth in staffing.

Quality is about making customer interactions more consistent, more informed, and better documented. When summarization, knowledge lookup, verification, and next steps become part of the flow, not only is the pace improved – the quality of execution is also elevated.

Growth is something many still underestimate: customer dialogue can be used to strengthen repeat purchases, reduce churn, and drive upselling and follow-up. McKinsey states that leading players in customer care are already beginning to see the effects of AI across customer experience, cost reduction, and revenue generation. Deloitte similarly points to service as an area for scalable ROI through faster, smarter, and more personalized interactions.

It is the combination that makes voice-based AI strategically interesting. This is not just an efficiency measure. It is a new way to organize customer dialogue.

4. Financial examples that leaders actually care about

For many leaders, this only becomes interesting when it can be translated into P&L. But not all numbers in a business case have the same character. Some scenarios can be directly supported by published benchmarks. Others work best as illustrative models for how the value can be calculated within their own business.

Calculation examples in voice-based AI should be presented with appropriate precision: as decision support, not as guarantees.

Scenario Character Potential Annual Impact
1. Cost Scenario Strong benchmark-supported (Deloitte: 30% efficiency) NOK 2.4 million (at NOK 8 million in operating costs)
2. Growth Scenario Benchmark-adjusted (6% effect on additional sales and renewal) NOK 10.8 million (at NOK 180 million in revenue)
3. Loyalty Scenario Illustrative, business-relevant NOK 600,000 (for 5,000 customers at NOK 6,000 each)
4. Capacity Scenario Operational and planning relevant 20,000 cases moved from manual to AI-supported

Scenario 1: Benchmark-supported cost scenario A company spends NOK 8 million per year on customer dialogue. If improved automation, faster handling, and less rework lead to a 30 percent more efficient operation: NOK 8.0 million × 30% = NOK 2.4 million in annual effect. This is a robust scenario – Deloitte states that 43 percent of organizations believe AI will enable a reduction in contact center costs by 30 percent or more within the next three years.

Scenario 2: Benchmark-Adjusted Growth Scenario A business with NOK 180 million in revenue where improved follow-up and more relevant dialogue influence renewal, upselling, or repeat purchases by 6 percent: NOK 180 million × 6% = NOK 10.8 million in potential top-line impact.

Scenario 3: Illustrative Loyalty Scenario A subscription business with 5,000 customers and an annual customer value of NOK 6,000 reduces churn by 2 percentage points through improved lifecycle dialogue: 5,000 × NOK 6,000 × 2% = NOK 600,000 in preserved annual revenue.

Scenario 4: Operational Capacity Scenario A service organization handles 100,000 cases per year. If the AI share increases from 30 percent to 50 percent, 20,000 cases are shifted from manual to AI-supported handling – in line with Salesforce's forecasts for 2027.

The calculation examples are illustrative scenarios based on published market data and common business case logic. Actual effects will vary depending on the starting point, data basis, process design, and operational execution.

Why today's model is often more expensive than it appears

The visible costs in customer dialogue are relatively easy to measure: wages, staffing, opening hours, and volume. The hidden costs are often higher.

When employees leave, it's not just capacity that disappears – experience, conversation logic, and the ability to resolve cases correctly the first time also go with them. The result is more training, longer processing times, and greater variation in the customer experience. Traditional models are not only labor-intensive. They are vulnerable.

Modern voice-based AI can change this economy – not only by reducing cost per inquiry but also by making quality less person-dependent and improvement more cumulative.

5. From Reactive Service to Proactive Customer Dialogue

Here lies some of the most underrated potential in voice-based AI.

Many still talk about voice-based AI as if it mainly involves inbound support: the customer calls, and the company responds faster or cheaper than before. It's important. But that's only half the picture.

The greater opportunity lies in treating customer dialogue as a two-way, continuous, and value-creating relationship. The company does not only respond when something occurs – it can also reach out when the timing is right, the context is relevant, and the value is clear.

A subscription business can reach out before the next delivery or renewal. The conversation can be used to clarify satisfaction, adjust preferences, handle changes, and introduce relevant add-ons. In other industries, the same logic can be used for follow-up after delivery, quality control, reactivation, or simpler customer success processes.

This is not old-fashioned telemarketing. It is value-driven, proactive customer dialogue – and that's where the real shift becomes clear: from conversations that end, to conversations that convert.

6. Context, operational memory, and continuous improvement

One of the main reasons why voice-based AI is strategically interesting is that the technology makes it possible to build more lasting quality in customer dialogue.

In many companies, quality remains vulnerable because important experience resides with individual employees rather than in the company's operational model. In environments with high turnover, this means that valuable knowledge can easily disappear.

Modern voice-based AI has a key strength here: when the knowledge base, conversation flow, routing, summarization, and escalation logic are improved, the improvement can be integrated into the model and reused consistently. The learning becomes cumulative.

But this doesn't happen on its own. KSIndeks shows that the most important factor for service satisfaction is that the issue is actually resolved. The value is created only when technology is used to increase the resolution rate, reduce friction, and make customer dialogue simpler and more precise.

For many businesses, this has a greater business impact than mere automation. They are not just building a cheaper model – they are building a more robust and improvement-capable operation.

7. What Voice-Based AI Actually Consists Of – and What Sets the Demo Apart from Operational Use

A common weakness in the discussion about voice-based AI is that the term is used as if it describes a single function. In practice, it is a composite operational capability – and the difference between an impressive demo and a solution that actually works in operation is greater than many expect.

Want to talk it through with someone who has done this before?

Fifteen minutes on a call. We will tell you honestly whether a voice agent fits your setup – and what has to be settled before it goes live.

Book Meeting (opens in a new window)+47 24 12 66 30

15 minutes · no obligation · pick your own time

From Integration to Autonomous Action

For a long time, it has been sufficient to describe voice-based AI as "connected to CRM." That is no longer adequate. The relevant question today is not just whether the agent has access to the data – it is whether it has the authority and capacity to contribute to resolving the issue.

A mature solution can increasingly carry out actions directly within the enterprise's systems: change a delivery, issue a compensation, rebook an appointment, update customer data, or route the case further with full context. In several use cases, the agent can now be an active part of the execution itself – not just an interface that gathers information for a person who then acts.

This switch – from conversation to execution – is what makes voice-based AI operationally interesting in a new way.

Memory and Personalization

A contextless agent is an agent that always starts from scratch. That is not good enough.

Mature solutions connect voice-based AI to the customer's history – previous inquiries, preferences, ongoing cases, and past solutions. This means that the conversation can more largely start where it last left off. It is not primarily a demo effect – it is a trait that determines whether the customer feels recognized or starts from scratch.

This is closely linked to the operational memory described in Chapter 6: quality is not only built at the system level but also at the individual level over time.

Infrastructure: MCP and standardized connection

A new infrastructure layer is emerging around agentic AI. The Model Context Protocol (MCP), launched by Anthropic in November 2024, is an open standard for securely connecting AI systems to data sources and tools – much like a universal adapter between the agent and the systems it needs access to. For voice-based AI, this means that the value increasingly lies not just in the conversation itself but in how well the voice can be linked to context, workflow, and action.

The four operational layers

A mature platform for voice-based AI can be understood through four layers – where the value increases the deeper down the stack the solution actually operates:

Layer Function
Understanding The system interprets speech, intent, and context
Decision The system determines what should happen next based on rules, data, and workflow
Execution The system performs actions directly in relevant systems
Improvement The system maintains and improves quality through analysis and further development of the knowledge base

What differentiates a demo from an operational solution is not the voice. It’s how many of these layers the solution actually owns—and how well they work together.

8. Why Nordic Customer Dialog Has Special Requirements

Nordic markets are not just geographically defined – they have distinctive characteristics that influence what actually works in customer dialogue.

Language precision is not optional. Norwegian, Swedish, and Danish are relatively underrepresented in the training data of global models. This means that dialects, technical terminology, and natural conversation pace can create friction not visible in English-language tests. The customer notices it immediately.

Trust is high – and fragile. Nordic customers are generally skeptical of experiences that feel generic or impersonal. A conversation that doesn't understand the context, or that handles the transition to human support awkwardly, can quickly undermine trust that took a long time to build.

Data control is a real requirement, not just a wish. GDPR and sector-specific regulations set concrete requirements for where data is stored and processed. For many Norwegian businesses, EU-based data centers are a prerequisite, not an option.

Process integration determines the outcome. Global standard setups often deliver a functioning agent in the demo. What separates the demo from operation is whether the agent is connected to the actual systems and processes the business uses in daily operations.

KSIndeks provides an important correction from the customer side: customers care little about the technology itself. They care about whether the issue is resolved, how simple the experience is, and whether they have to contact support multiple times. This should guide the design – not technology alone.

9. Trust as a Competitive Advantage: BankID and Vipps in Customer Dialogue

Customer identification is not just a security issue. It is a prerequisite for a meaningful conversation.

When a customer calls to book an appointment, modify a delivery, or cancel a subscription, most conversations start with the same ritual: providing their name, date of birth, and zip code. There's friction before the dialogue has even begun – and it's unnecessary.

In Norway, there is a digital infrastructure that addresses this. BankID is used by 4.6 million Norwegians and accounts for nearly one billion logins and signings annually. Vipps MobilePay has over 12 million users in the Nordics. Both are well-established as identification tools that customers already know and trust.

By integrating BankID or Vipps into the incoming call flow, the customer can identify themselves before or at the start of the call – for example, via a quick Vipps confirmation or BankID verification on their mobile screen while waiting. The result is that the agent knows exactly who the customer is from the first second, what they have purchased, when they last contacted us, and what issues are unresolved.

It provides two concrete benefits:

Security. The business can safely perform sensitive actions – changes, cancellations, orders – without the risk of the wrong person acting on behalf of the customer.

Conversational Quality. The agent can start the conversation with full context. No introductory verification round. No repetition of information the customer has already provided. The conversation begins where the customer actually is.

This directly relates to what is described in Chapters 6 and 7: identification is the very prerequisite for operational memory and personalization to actually be utilized.

For outbound calls – where the business initiates the call to the customer – the opposite trust challenge applies: how does the customer know it is actually you calling? Here, a BankID notification prior to the call can confirm the sender, or a Vipps message with a link to the context of the call before the customer answers the phone.

In both directions, this is a genuine Nordic advantage that global platforms cannot replicate without local adaptation.

10. What Leaders Need to Understand – and How Threll.ai Works

The most important thing leaders can do now is not ask whether they should test voice-based AI.

The most important thing is to ask where in the company customer dialogue is still treated as a cost, when in reality it has become a competitive area.

The difference in the coming years will not just be between companies that use AI and those that do not. It will be between companies that build operational value around AI and those that still treat it as a tool experiment.

This means that voice-based AI should not be owned as a peripheral experiment within the organization. It should be rooted where revenues, costs, and customer experience are actually managed.

Those who think narrowly will look for a voicebot. Those who think correctly will build a conversation model that converts.

For businesses that want to move from insight to action, the next step is not necessarily to start with a large transformation. The right next step is to identify where voice-based AI can create clear value the quickest.

A good starting point is three questions:

Where in the customer journey can conversations create the most value? Not just where the volume is high, but where better dialogue can actually impact cost, customer experience, loyalty, or growth.

Where does the real ROI potential lie? Not in abstract AI talk, but in concrete conversation types, processes, and business goals.

What does it take to go from pilot to operation? The critical factors are rarely the model alone – what matters most are workflow, integrations, data control, language accuracy, and operational follow-up.

Threll.ai is built for this very transition. We typically work through three phases:

Analysis – We map today's customer dialogue, identify use cases with clear value, and build a realistic business case.

Design and Implementation – We design conversation flows, integrate with relevant systems, and establish the solution with Nordic language support, data control, and clear success criteria.

Optimization – We follow up with conversation analysis, testing, and continuous improvement so that the solution not only gets launched but actually gets better over time.

What sets Threll.ai apart is not just a single feature, but the entire package: voice, workflow, integrations, and continuous improvement that work together within one operational model – built for Nordic requirements and Nordic customers.

Conclusion

Voice-based AI is not just a new tool. It is becoming a new operational layer in customer interaction.

The true value does not solely lie in the technology itself. It lies in whether the business uses it to solve multiple issues more effectively – with less customer effort, higher consistency, and stronger operational flow over time.

This is why this category is so exciting right now. Not only because the models have improved, but because the prerequisites are finally in place to turn customer dialogue into something more than a cost item.

For Norwegian and Nordic businesses, this shift will be particularly important in the coming years. Those who act early not only build efficiency – they build a stronger operational model for how the business meets, assists, and develops customers over time.

Conversations that convert. That is where the next competitive advantage is built.

About Threll.ai

Threll.ai develops voice AI agents – intelligent, autonomous virtual employees that can call, understand answers, perform tasks, and adapt in real time. Our goal is to let artificial intelligence become your new colleague – one that can speak, listen, and solve tasks efficiently. Infrastructure and data storage in the Nordics — not with hyperscalers outside the EEA. We give you full control and a competitive advantage.

Call Thea at +47 24 12 66 30 for more information or help booking a meeting with the founders of Threll.ai.

Sources

  • Gartner, Survey Reveals 85% of Customer Service Leaders Will Explore or Pilot Customer-Facing Conversational GenAI in 2025, December 9, 2024.
  • Gartner, Survey Finds 91% of Customer Service and Support Leaders Under Pressure to Implement AI in 2026, February 18, 2026.
  • Salesforce, State of Service Report, November 2025.
  • McKinsey, Building trust: How customer care leaders pull ahead with AI, February 2026.
  • Deloitte, The Future of Service, 2026.
  • Capgemini Research Institute, Unleashing the value of customer service, 2025.
  • Anthropic, Introducing the Model Context Protocol, November 2024.
  • KSIndeks, Insights into what drives service satisfaction in Norwegian customer dialogue.
  • BankID and Vipps MobilePay, on the prevalence and use in Norway and the Nordics.

Download the whitepaper as PDF

Related Articles

Continue reading more articles

See all articles