Call almost any company in Portugal and you will hear the same recording from 15 years ago: "Press 1 for sales, press 2 for support, press 3 for billing, press 4 to speak with an operator." After navigating three menus, you wait 8 minutes in a queue, and when someone finally answers, you have to repeat everything you already said. In 2026, this model is not merely frustrating — it is a real risk of losing customers. Conversational artificial intelligence now makes it possible to replace these archaic systems with voice assistants that understand natural language, respond in fluent speech and resolve the majority of requests without human intervention.
Why the Traditional IVR Is Dead
IVR (Interactive Voice Response) was revolutionary when it emerged in the 1990s. It allowed call filtering, customer routing and reduced the load on call centres. But the world has changed, and IVR has not changed with it. Today's consumers are accustomed to interacting with intelligent assistants on their phone, in their car and at home. When they call a company and encounter a rigid options menu, the experience is jarring by contrast.
The numbers confirm the frustration. Studies by Vonage reveal that 61% of consumers consider IVR menus a negative experience. Worse still: 13% hang up and never call back. For a clinic that depends on phone bookings, or for an estate agency where each lead is worth thousands of euros, every lost call is money walking out the door.
The fundamental problem with IVR is structural: it forces the customer to adapt to the system, rather than the system adapting to the customer. If the reason for the call does not fit neatly into one of the menu options, the user is stuck in an endless loop of button presses. And when they finally reach a human operator, the call context has been completely lost.
How an AI Voice Assistant Works
An AI voice assistant operates in a radically different way. Instead of presenting an options menu, it simply asks: "Hello, how can I help you?" The customer speaks naturally — "I would like to book an appointment for next Tuesday morning" — and the system understands, processes and responds in real time.
Behind this seemingly simple interaction, several layers of technology work together:
1. Automatic Speech Recognition (ASR). The call audio is converted to text with an accuracy that, in the best current models, exceeds 95% for European Portuguese. The technology recognises regional accents, background noise and even colloquial expressions. We are no longer in the era of constant "I did not understand, could you repeat?".
2. Natural Language Understanding (NLU). The text is analysed to identify the user's intent and extract relevant entities. When someone says "I need to reschedule my dermatology appointment from Wednesday to Friday", the system identifies the intent (reschedule), the speciality (dermatology), the original day (Wednesday) and the requested day (Friday).
3. Dialogue Management. A dialogue engine determines the most appropriate response, taking into account the accumulated conversation context. If the system needs more information — such as the preferred time — it formulates a natural follow-up question, not an options menu.
4. Text-to-Speech (TTS). The text response is converted to audio with voices that sound increasingly natural. Modern TTS voices include pauses, intonation and cadence that make the interaction nearly indistinguishable from a conversation with a real person.
5. Backend System Integration. The assistant does not exist in isolation. It connects to the scheduling system, the CRM, the customer database and any other software required to resolve the request during the call, without transferring to a human.
Natural Language Understanding in Portuguese: The Challenge and the Solution
For years, European Portuguese was the "poor cousin" of voice technology. Models were trained predominantly in English, and when Portuguese support existed, it was optimised for Brazilian Portuguese — with significant pronunciation, vocabulary and sentence structure differences that severely affected accuracy.
This reality has changed over the past two years. Large language models — and specialised speech recognition systems — now include substantial European Portuguese datasets. This means a well-configured voice assistant today understands expressions such as "I want to book for next Monday morning", "I need to speak about my policy" or "I am calling about a boiler breakdown" with accuracy rates above 93%.
More importantly: modern systems learn from each interaction. The more calls they process, the better they understand the linguistic particularities of each sector — the medical vocabulary of a clinic, the property terms of an estate agency or the legal language of a law firm.
Want to see an AI voice assistant in action?
We provide a personalised demo with your company's call-handling flow — no obligation.
Discover AI Voice Assistant →Use Cases: Who Benefits Most
Clinics and Medical Practices
Clinics are possibly the sector that benefits most from AI voice assistants. The reason is simple: the vast majority of calls follow predictable patterns — booking appointments, rescheduling, cancelling, requesting test results, confirming times. These are interactions that do not require clinical judgement, yet they consume hours of receptionist time.
A well-implemented voice assistant in a clinic can handle 70 to 85% of calls without human intervention. It books appointments directly in the clinic management system, checks availability in real time, sends confirmations by SMS and makes automatic reminder calls 24 hours in advance. Receptionists are freed for tasks that genuinely require human attention — welcoming patients in person, managing complex situations, coordinating with doctors.
And there is an additional benefit that is rarely mentioned: the assistant answers outside business hours. At 10 p.m. on a Sunday, when a patient remembers they need to book an urgent appointment, the assistant is available. No queues, no "please call during opening hours".
Estate Agencies
In the property sector, response speed is critical. A Harvard Business Review study showed that the probability of qualifying a lead drops 10 times if the initial contact takes more than 5 minutes. For an estate agency, this means every unanswered call can represent a sale worth tens or hundreds of thousands of euros lost.
The voice assistant can qualify leads in real time — asking about the type of property sought, the area, the budget and availability for viewings — and schedule viewings directly in the consultant's diary. When the consultant receives the notification, they already have all the information organised and can prepare accordingly. The lead, in turn, has had an experience of immediate and professional service, even if they called at 9 p.m. on a Friday.
Law Firms
Law firms face a particular dilemma: they need to convey professionalism and personalised attention, but lawyers' time is far too valuable to spend answering generic calls. An AI voice assistant resolves this equation — it answers professionally, gathers case information, schedules initial meetings and routes only the calls that require immediate legal intervention.
For the client calling for the first time, the experience is remarkably positive: they are answered immediately, their information is recorded accurately and they receive a meeting date before hanging up. For the firm, operational efficiency improves dramatically — lawyers focus on legal work, and reception ceases to be a bottleneck.
Implementation Guide: From Zero to Voice Assistant in 4 Weeks
Week 1: Flow Analysis and Conversational Design
The first step is not technological — it is strategic. We analyse calls received over the last 30 to 60 days to identify the most frequent reasons, the language patterns used by customers and the failure points in the current service. Based on this analysis, we design the conversational flows: the paths the assistant will follow for each type of request, including clarification questions, exception handling and escalation points to humans.
Week 2: Configuration and Integration
We configure the voice assistant, integrate it with existing systems — diary, CRM, management software — and train the model with sector-specific vocabulary. At this stage, it is critical to test natural language understanding with real examples, adjusting parameters until accuracy reaches acceptable levels for production.
Week 3: Controlled Environment Testing
Before going live, the assistant is tested internally. The team makes simulated calls with real scenarios — including difficult cases, strong accents and ambiguous requests. Every failure is logged, analysed and corrected. The goal is to ensure the assistant handles gracefully any situation it cannot resolve, transferring to a human transparently.
Week 4: Launch and Monitoring
The assistant goes into production, initially in parallel with human support. We monitor key metrics — resolution rate without escalation, average call time, customer satisfaction, abandonment rate — and make daily adjustments during the first week. Full transition typically happens between the second and third week after launch.
What to Expect in Terms of Results
Results vary depending on call volume and request complexity, but the patterns we observe are consistent:
• 60 to 80% reduction in calls requiring human handling. The majority of routine requests — bookings, information, confirmations — are resolved automatically.
• Waiting time reduced to zero. The assistant answers on the first ring, 24 hours a day, 7 days a week. No queues, no "all operators are currently busy".
• 25 to 40% increase in out-of-hours bookings. Customers who previously postponed or forgot now book when it suits them — at 11 p.m., on weekends, on public holidays.
• Significant operational savings. An AI voice assistant costs a fraction of a full-time receptionist's salary, and needs no holidays, breaks or sick days. For businesses with high call volumes, ROI is achieved in 2 to 3 months.
• Improved customer satisfaction. Paradoxically, many customers prefer interacting with an efficient assistant rather than waiting for a human. The condition: the assistant must work well. A poorly implemented voice assistant is worse than an IVR.
The Most Common Implementation Mistakes
Not all voice assistant implementations succeed. The most frequent errors we observe include:
Trying to automate everything at once. The goal is not to eliminate human support — it is to free it for where it truly makes a difference. Start with the 3 to 5 most frequent call types and expand gradually.
Ignoring conversational design. A voice assistant is not a chatbot with audio. Voice conversation has different rhythm, expectations and limitations compared to text. Investing in conversational experience design is as important as the underlying technology.
Not testing with real users. Laboratory scenarios never capture the full variability of the real world. Testing with real customers, in real conditions, is essential before any launch.
Forgetting the escalation path. When the assistant cannot resolve, the transition to a human must be instant and transparent. If the customer has to repeat everything they already said, the experience is worse than having no assistant at all.
Conclusion: The Future of Phone Support Is Conversational
The traditional IVR had its time. For two decades, it was the best available technology for managing call volumes. But conversational artificial intelligence has made it obsolete — not because it is cheaper (although it is), but because it offers a fundamentally better experience for both the customer and the business.
The question is no longer whether AI voice assistants will replace IVR menus — it is when. And the companies that adopt this technology first will capture the customers that others lose. In a market where customer experience is increasingly the differentiating factor, phone support can no longer be the weak link.
If your company still asks customers to "press 1, press 2", it is time to consider an alternative that speaks your customers' language — literally.
