We must thank João Camarate, an advisor to Visor.ai. A long time TADS member, we ran TADSummit 2022 at his GoContact office in Aveiro after Broadvoice bought them in 2021.

Since 2016, Visor have improved and accelerated the customer service experience by automating interactions through different channels such as chat, email, and voice. Beginning with chatbots, and moving to LLM (Large Language Model).

Gonçalo is a breadth of fresh air, frankly sharing the contact center situation. They began with chatbots, then expanded to other channels including voice and email. And built out the flexibility to meet their customers’ needs, being banks and insurance there were many specialized needs.

While Talkdesk, another Portuguese customer service company took a pure SaaS model targeting large global deployments, Visor focused locally and built a locally defensible business

Sales cycles can take 3 months to 3 years (worst), with an average of 6 months. The nice thing with Portugal is once you’re with a client, they remain a client for many years and you grow together. Most clients already have a contact center solution. However, the chatbot piece can be quite old, perhaps 10 years. So Visor comes in as a modern AI layer.

In 2022 they made the decision to expand their offer in virtual assistants across voice, agent assist, quality assurance, and in workflow automation across email (using a slow yet intelligent LLM to process what can be quite complex emails with multiple attachments) and associated business processes. Last year we had Surbhi from Symbl.ai present at TADSummit on such QA solutions in measuring the quality of real and virtual agents. It was great to hear Gonçalo’s appreciation for automating such measurements, given the cost and variability of human based measurements.

Visor’s aim is to become the AI partner for contact centers. Even if there are point solutions like Symbl.ai, the integration and orchestration is complex, which Visor fulfils. Interestingly, the first 20 clients were banking and insurance, today the customer base is much broader. 80% of their clients are from Portugal, the rest from Spain, Africa, and Latin America.

Visor’s business is currently self limiting because of client’s implementation, customization, and specific vertical needs. Which at this point helps them deliver excellent AI solutions. As well as a defensible business.

Karel asked about how Visor deal with conservative clients like banks. Guard rails are important. Also demonstrating the performance of the top three models with quantification of the hallucination metrics. The model can be limited to simple FAQs. Visor’s ability to build to the clients’ needs enables them to find sweet-spots, coupled with Quality assurance measurements.

We then discussed model preferences, Visor had built its own. However the AI Act in Europe, is a heavy set of regulations that make it difficult for smaller companies to build and maintain their own models. So today they evaluate the latest models and enable clients to change model if required. Additionally, the Act limits the data sources EU companies can use. This led onto a discussion about local language performance, which Karel knows well.

Karel asked a question around latency when using voice. Which has been a perennial problem. Gonçalo quoted an example from a pharmacy, generally used by older people for booking flu, COVID vaccine appointments. It was created in September last year. The results are, 99% of people who used the app made an appointment. They created a voice using an employee, as the text to speech Portuguese voices are generally poor. The constrained application, and using a real voice, led to most users thinking it was a real person, not an AI! They also use Deepgram and Google Gemini for the speech to text, which perform well.

The pharmacy numbers are unusually high, a car inspection center with 80 locations in Portugal has a booking rate of 70+%, which is more typical.

We then moved onto channel to market. Initially direct as they built the business, focusing on filling the chatbot gap for banks and insurance. As they expanded their offer the channel became more indirect, through contact centers, e.g. GoContact, where Visor fills the AI gap and its professional implementation services. The sales pipeline is good, with new clients every week. The focus has shifted to ensure consistent implementation quality. Bringing in AI is like hiring a new colleague, onboarding is key, training and context are critical.

With GoContact as a channel, broadvoice becomes a natural next step for Visor, opening the North American market. One of the learnings with GoContact as a channel is using AI Quality Assurance as an entry point. To demonstrate the value, performance, and saving AI can bring in what can be a service with mixed results. That for me is very interesting and backs up what Surbhi from Symbl.ai was sharing. Plus it gets the AI trained up, smart!

Onboarding your AI adequately is critical.

Karel asked if they were pure SaaS; with some clients they run on a virtual private cloud, because of the client’s specific needs. Gonçalo compared Visor to Talkdesk, another Portuguese company which has taken a pure SaaS approach. This enabled them to grow fast, but are less suited to parts of the Portuguese / European market. He also highlighted the importance of ISO 27001 certification, compliance with GDPR, other industry specific regulations, and the Center for Responsible AI.

Gonçalo explained the processes they use to protect customer data in erasing the call data after AI Quality Assurance. I couldn’t help but get a mention of STROLID’s work on vCons (virtual Conversation) for all the use cases around this topic.

We moved onto the topic of corporate planning. They raised 4M Euro in 2023. What’s the plan? For Portuguese companies with profits, Portugal is very generous, where 80% of taxes can be set aside for R&D and corporate planning. However, they must comply the previously mentioned regulations, there’s about 400 companies in Portugal that meet this criteria, Visor is one. Karel and Gonçalo Will have some interesting follow on conversations.