"We are constantly working on improving the bot's conversational intelligence"

Pablo Palafox, CEO and one of the founders of Happyrobot, shares how he started his business venture and the latest developments of this tool that offers voice solutions through the AI application, with the aim of establishing the most human connection possible with customers.

How was Happyrobot born?

Javier and I are brothers, and Luis has been Pablo's friend since the first day of the race. The three of us share a love of technology and entrepreneurship. So, Happyrobot was born many years ago, when we created a WhatsApp group to talk about robotics and AI advances, and where the dream of creating a company together one day is always present.

Based on the learnings from reading the blog of Paul Graham (PG), founder of the Y Combinator accelerator, the three of us were clear that the best way to start was to work on a side project, to test ourselves as a team. And so we did. 

But when Happyrobot was really born is when, with no intention of staying in the academic world after getting my PhD, nor to stop working in a company where projects do not advance fast enough, I started to encourage Luis and Javier to start something more seriously. In this case, a solution to a problem I had faced while training computer vision models in my PhD. 

Everything changes when we enter Y Combinator

Happyrobot was therefore originally a platform for processing datasets, i.e. images, and training computer vision models in a more automated way. The first customers were robotics companies, hence the name.

But everything changed when we joined Y Combinator with what looked like a stable and growing business. Another key lesson from PG is that a startup has to try to grow at a breakneck pace or change direction, iterate, if that is not happening. We are talking about weeks, days. We moved a lot with the computer vision platform and saw that, while it was a good business at the time, it was not going to experience exponential growth. So we decided to pivot. Another of those terms so common in startups.

It was around this time that ChatGPT and the famous LLMs were taking the world by storm. One area that still had many unresolved problems was the application of LLMS to voice applications. We decided to get to work on that and focus it on a very specific industry. From his experience at a food distributor, Javier knew that the logistics industry in the US is still largely driven by phone calls. Therefore, in November 2023, we started contacting companies in the sector with a minimum viable product that we built in a matter of days.

What makes Happyrobot different and different from its competitors? 

Today, the biggest differentiation lies in the level of realism of the product

There are several tools for creating AI solutions for voice. Some of them are tools for programmers, while others are solutions or services that build the chatbot for you and get it up and running. In terms of tools for programmers, there are complete solutions with all the "pieces" to build the chatbot, and others that provide only a few "pieces" to assemble it. At Happyrobot, we decided to make a complete tool for developers in enterprises. That is, companies with sufficiently large teams can set up their own voice AI applications, using our tool, but where those teams are not experts in the area of voice AI. These companies are often open to either buying an existing solution or building it in-house. With Happyrobot, you have both alternatives. We can let them use the tool to benefit from the architecture we are building, or we can do an implementation where we get more involved.

Today, the biggest differentiation lies in the level of realism of the product. Ideally, our customers do not want to distinguish whether there is a bot or a person on the call. And this is how it happens on many occasions. Our focus is on solving many technological aspects of that architecture so that the conversation is fully fluid and realistic, and just as importantly, so that it can scale to thousands of calls in parallel. We ourselves are consumers of some parts already assembled by other teams, while others are internally developed algorithms.

Some of the advantages of a bot are the fact that it adapts to demand immediately

Finally, another differentiating factor is verticalisation in an industry. This has allowed us to understand a number of specific use cases very well and to 'speak the language' of our customers. As well as working on integrations with other information systems that, once you have it for one client, you can scale it up for everyone using that system.

It would be interesting to make the comparison with other non-technological solutions. Many companies have outsourced these services to other countries that offer labour at very low cost. For example, Colombia or Mexico. To compare with humans, AI still has a long way to go. In fact, we believe that the best solution is a combination. But some of the advantages of a bot are the fact that it adapts to demand immediately, i.e. it can answer hundreds of calls at a time, and when there are no calls, there is no charge; it can provide these services uninterrupted every day of the year; it requires training only once; and it is much faster at answering questions or recording information, since it is directly connected to the systems.

What's new with Happyrobot? 

The ultimate goal is to have a bot that is very good and efficient

At Happyrobot we are working on a UI that allows us to build a first viable version of the chatbot very quickly and then, if the use case requires it, to take samples of real calls and give examples to the AI of how a specific point should have been handled. The latter would be that fine-tuning or "little training" of the AI with customer data. In other words, we help customers create their own models for their use cases, and this offers several advantages, including greater efficiency.

The ultimate goal is to have a bot that is very good and efficient in the use case for which it has been created. And, generally speaking, bots should always be very natural in conversation. To this end, we are constantly working on improving the bot's conversational intelligence, allowing situations such as the interlocutor interrupting the bot and the bot being able to react in a natural way.

International presence

Our largest customer is Circle Logistics

Right now we only have a presence in the US. In the US, and residually in Mexico, as some of our customers are transporters crossing the border and have truck drivers who only speak Spanish. We are starting to have some contact with large Spanish companies, but for the moment the priority is the American market.

Our largest customer is Circle Logistics, a logistics broker with a turnover of close to one billion dollars. In addition, we have a few other large companies as clients, such as one of the largest moving companies in the USA. We helped them to capture customers during the hours when their employees are no longer working. What keeps us busiest are pilots or proofs of concept with several larger logistics brokers, some in the top 10. Sales to large corporations are slow and time-consuming, but they are certainly our main target, as some of these companies have thousands of calls a day, hundreds of them at a time. We really went for this industry, but there are other markets where there are hundreds of phone calls and where AI can bring a lot of value.

What did it mean to be selected by Y Combinator, Silicon Valley's most exclusive accelerator?

It has been a tremendous learning experience. Something that sticks with you for life if you go with an open mind, that's for sure. Y Combinator are experts in failure. They have seen it so many times, despite being the most successful accelerator in their investments, that they have made it their differentiator. So no one is going to tell you what to do, but they are constantly analysing whether what you are doing is too similar to what many others did that failed or never really got off the ground. The demand is maximum and the level of the rest of the startups is incredible. It is not so much because they are far superior, but because they are extremely hard-working and focused people. Some of them, aged 18, have been programming since they were 11, worked for companies since they were 16 and have dropped out of university to found their startup. You leave with a great network of friends and people with whom you can identify and who are working on very similar things, so there are a lot of synergies and we help each other a lot in that community. 

Finally, one thing YC brings to the table is branding. And that helps a lot to sell. The companies here know the accelerator and know that it is somehow a "seal of quality".

But none of this is enough to achieve success. There are many variables that, in theory, need to be mastered. For example, it is important not to contract prematurely. The founders are the ones who have to do all the heavy lifting to build the first versions of the product and the first customers. There are many things, and all very obvious, but it is surprising to see how we all sooner or later deviate from logic and make mistakes that jeopardise the delicate existence of a startup on its way to success.