Toqan is your product and engineering teams’ AI Assistant, created by a team of data scientists, engineers and technologists at Prosus, on a mission to empower enterprise teams. For nearly a decade, we have been building first-in-class AI solutions with the world’s leading technology companies. With a strong foundation in applied research for machine learning and artificial intelligence, our team prioritizes business impact, supporting our multi-billion dollar operations and catering to over 2 billion customers.

Toqan treats you like a team member by considering your goals, actively planning and going out to execute those plans drawing on your internal data, public information, and the best AI capabilities available. It is a smart addition to every product and development team, built for impact and collaboration.

We have a history of building GenAI solutions to complex business problems. We released FinBert in 2019, the leading language model trained for financial sentiment analysis with over 30M downloads to date. Next, we built FoodBERT, a language model to improve search results for iFood’s 40M users and drive over 10 billion AI-based recommendations. In 2023 we trained and deployed the language model behind Sololearn’s Kodie, which is used by millions of users and was featured in Apple’s Trend of the Year report alongside the ChatGPT and Canva apps. Today, we’re driving enterprise adoption of GenAI at over 30 tech companies across the globe with Toqan, attracting attention from Forbes, the Wall Street Journal and Fortune among others.

We know that identifying, building and deploying AI solutions in business contexts is not simple. As we roll out Toqan to new customers, we have been reflecting on what we have learned and what we take forward into this next chapter. Here are Toqan’s three guiding principles for AI in business.

Focus on User Needs

When we hit a problem, we stop and ask ourselves: What are our users trying to achieve? Our users are working towards some amazing goals, and Toqan is helping them get there faster and more confidently.

Toqan usage data

Sometimes Toqan helps in small ways, like solving a software engineering problem, drafting an email or summarizing dense information. Other times Toqan leans in a bit more by clarifying a tough decision using sourced internet research to outline and compare options. In some cases, Toqan can step into the driver’s by analyzing and visualizing transactions in a SQL database to help identify an unnoticed bottleneck.

Placing user needs at the core of every decision we make is the foundation for delivering value to business customers.

Align Technology with Value

For many of our users, this means building powerful but easy to use GenAI tools for professional work. Toqan lives inside of your organization which means you can safely discuss business information, create ideas, and share analyses and workflows with your colleagues. This enables Toqan to provide your teams with broad and intuitive access to the latest GenAI capabilities, increasing productivity, accelerating product innovation and fostering new ways of working.

Like any team member, trust is a cornerstone of Toqan’s value. When our customers demanded a secure environment for sensitive data we obtained SOC2 and ISO27001 certification. When our users told us that it was hard to trust AI generated content we worked to bring down the user reported hallucination rate from about 1 in 9 to about 1 in 100.

Toqan hallucinations

When customers asked to help drive productivity, we partnered with them and went to the source to ask users how they are faring. In April 2024, 81% of Toqan users self-reported a productivity increase of at least 5% - 10%, and an additional 17% of users self-reported a productivity increase of over 20%.

When our customer recognized a need for broad GenAI literacy and education for their teams and leadership, we teamed up with Udemy to produce an industry leading GenAI course for enterprise leaders using Toqan as an experiential learning platform. Today, we’re proud that 96% of users report Toqan contributes positively to gaining a better understanding of how to use AI in their role and 86% report that Toqan helps them to understand the impacts of GenAI on their organization.

Aligning technology with value is the natural extension of focusing on user needs. The collaboration between our customers, users and Toqan is at the heart of how we work.

Never stop learning.

We believe the best way to learn is to do. There is little value in a technology that stays in the research lab, as it is the appropriate application of technology to real world problems that delivers value, not the technology itself.

Our leading applied research yields both insights into the landscape of new technologies and an understanding of how and when to apply them. From building and launching the earliest experiments in agent-based AI systems to pragmatically scaling high impact GenAI applications using bespoke trained models, we are never afraid to approach a sticky problem in a novel way if there is demonstrable value on the table.

LLM Leaderboard snapshot

Consider our ongoing research into benchmarking the latest language models. When we noticed a disconnect between the LLM performance we were seeing in real business applications and the apparent results available from standard benchmarking sources, we asked the natural question: Why?

It turns out that standard evaluation methods, like those behind Hugging Face’s LLM Leaderboard or Stanford’s Helm benchmark, rely on unrealistic and awkwardly academic scenarios to rank the capabilities of language models. In other words, the tests used to evaluate LLMs are biased towards use cases that have little application to the work that Toqan users do every day.

We saw the value in matching user needs with the most relevant models and decided to build our own benchmarks. The outcome is the ProLLM leaderboard, a compilation of benchmarking results using Toqan’s proprietary evaluation sets which are based on realistic business use cases. Our leaderboard is a key resource for understanding the quality and accuracy of language model performance on tasks that actually matter to our product and engineering teams. These grounded results provide crucial input that helps Toqan to more effectively select the best tools to give you the most relevant information when you ask for it.

We’re so excited about our insights and rapid evaluation process that we are sharing the ProLLM Leaderboard with the world. Check out the upcoming post from our team for a deep dive into our process, methodology and results.

Looking forward

This blog is where we will share how Toqan is helping customers empower product and engineering teams. We will also highlight how our applied research translates into continuous improvements in our products while always remaining grounded in the realities of enterprise use. Finally, we will use this space to share reflections on AI in the workplace, how work is changing and where we are heading next.

To keep things organized, all of our publications will support one of Toqan’s three pillars of value: enabling product and engineering teams to safely use the best AI Technology, in context, as your partner for the journey ahead.

We've also published some of this content in a separate article by Euro Beinat on our Medium blog. If you are interested in trying Toqan, want to book a demo, or want to discuss our applied research, events or upcoming releases, please sign up below!