How We Created Lyro – the First Fully Conversational AI Chatbot for SMBs – in Only 3 Months

It has already been over two months since Lyro - our latest Tidio feature saw the daylight.

What is Lyro? It's the first conversational AI chatbot developed to help small and medium-sized businesses offer exceptional customer service. After launching it in the Tidio panel, it answers up to 70% of customer questions, precisely as a human support agent would.

How did we make Lyro and, therefore, change the whole unique selling proposition of Tidio in only 3 months? Well, this is a heck of a story! It involves the real startup vibe, dedication, and a cross-team collaboration on scale.

Fasten your seat belt and read on!

Phase 1: The Game-changing Trick

From the very beginning, our goal was to create a conversational AI that could engage in natural and meaningful conversations guided solely by an algorithm/model. Before Lyro, in our chatbots, we could only automate individual parts of a conversation with a website visitor (i.e., their intent and sentiment and analyze those using clustering methods). Still, it was never a fully automated and independent bot with human-level performance.

With the arrival of ChatGPT, we began experimenting with its capabilities. The generated answers showed significant improvement, but a lack of control hindered the safe implementation of the technology in Tidio. Hallucination and turbo creativity posed persistent challenges.

In early March, Wojtek Pyrak, our Senior Machine Learning Engineer, created a simple demo of a conversational AI based on ChatGPT and a knowledge base consisting of FAQs (scraped by our FAQ Wizard). While the demo showed promise, it still suffered from hallucinatory tendencies, making it unsuitable for a fully functional product.

Later in March, the team, including Wojtek Pyrak, Łukasz Woronkiewicz (also Senior ML Engineer), Marcin Gwizdała (Chief Strategy Officer), and myself (Bartek, ML Lead), met in our Szczecin office to plan the future roadmap for Tidio AI area. During these workshops, we envisioned creating our own Language Model (LLM) that could be controlled. We planned all the necessary steps towards that, but due to the required work, it still seemed like a distant dream.

However, fate had something extraordinary in store for us. One night during the team’s reunion, Wojtek delved deep into the realm of controlling language models and stumbled upon a game-changing trick. He incorporated this newfound knowledge into his demo and presented it to the team on Friday. Astonishingly, the trick worked! Although not perfect, it showcased incredible potential and allowed us to gain a measure of control over the language model. Fueled by this discovery, we embarked on a journey to develop the model further and transform it into a product for our customers as soon as possible.

Phase 2: Cross-team collaboration

Empowered greatly by our discovery, we drew our future plans on board with ease and immediately called upon Piotr Cygan (Senior Product Manager) to share the workload with his team. Piotr then invited i.a. Marcin Stoll (Chief Product Officer), Andrzej Jagielski (Head of Product Design), Łukasz Witczak (Staff Backend Developer), and many other people who significantly contributed to the process.

We were all working together to refine and expand the project plan, all in a matter of 2-3 hours, just before our departure from Szczecin. The atmosphere was electric, I felt like I was in a fast-paced, early-stage startup in a small room filled with enthusiastic individuals. Ideas overflowed, filling every inch of available space on the whiteboard! We decided to put aside all other projects and focused solely on developing the model, whose name we democratically elected to be – Lyro.

Building the first of its kind is never easy. But time and again, this team has proven we can handle any challenge. We navigated uncharted waters, figured things out as we went, and executed them at the highest level. Everyone stepped up, brought their A-game, and made sacrifices to see this vision through. The sheer talent and teamwork on display have been humbling and inspiring. Over 30 individuals from various departments collaborated in unprecedented teamwork, making this project truly one-of-a-kind.

Phase 3: The results bring real satisfaction

It turned out the process was more complex than we initially had anticipated. The Tidio team was up to the challenge, and everyone was giving their best. Changing priorities and fixing bottlenecks lightning-fast. It felt like a hackathon that lasted several weeks - an authentic startup vibe, as we all had always imagined joining Tidio (at least I had).

The journey was incredible, yet the true satisfaction came from the end result. After 3 months of hard work (I’m not counting the brainstorming in March), in early July, we proudly released Lyro to all Tidio customers, becoming the first company to offer such a solution to the small and medium businesses segment! 🥳

The fact is a real game changer - not only for our customers and Tidio users (which are over 6.2% of the global population), but also for the whole industry. Why? First of all, Lyro is easy to implement and doesn’t require any training - anyone can start using it in under 5 minutes. Second of all, we greatly minimized the risk of hallucinations -  Lyro is very unlikely to confuse users by giving unverified information. The AI uses the knowledge base you provide and gives personalized support to your customers, just as a support agent would. Lyro stays within the boundaries of the provided data, so one can rest assured that its answers are accurate. Speaking answering clients – our first research shows Lyro answers up to 70% of customer questions. Imagine the time you free up this way. And the best thing? We made it free! 😎 Lyro is available on all Tidio plans, including our free-forever one (up to 50 conversations). This means anyone can benefit from the latest technology to grow their online business. Isn’t it incredible?

Phase 4: Looking into the future

Even though Lyro is already out there, our hard work has not ceased. The entire team here at Tidio continues to work to ensure the high quality of our product and expand its capabilities. We are improving the core AI engine at the heart of Lyro, integrating new techniques to make it more powerful and efficient. Additionally, we are developing new tools to make it easier for our users to create robust knowledge bases covering a wide range of topics and subject areas. This will allow Lyro to understand and converse about more concepts more accurately. We are also expanding language support, enabling Lyro to comprehend and communicate in multiple languages (so far Lyro is available in English).

Through these ongoing efforts, we strive to make Lyro the most capable and flexible AI assistant on the market. The whole shows incredible dedication to make it happen. What drives us forward is the shared vision to keep pushing Lyro's boundaries and deliver the best possible product to our users.


So, to the whole Tidio team, thank you. Thank you for your brilliance, your passion, and your tireless efforts. Thank you for coming together to make the impossible possible. Together, we created Lyro, the first fully conversational AI chatbot for the SMBs segment that changes the game for our customers and online businesses at large.

I could not be prouder of what we've achieved. Lyro is a testament to the Tidio team's hard work, perseverance, and commitment to innovation. And it's just the beginning! I’m sure we can create anything together – stay tuned for more!

Share this article:

You may also like