Economía para la Pipol, a Colombian digital media outlet, partnered with Datasketch to create an AI-powered chatbot that transforms how citizens access and understand economic information. Rather than relying on jargon-heavy traditional channels, the chatbot delivers insights in the accessible language that has become Economía para la Pipol’s signature style.

The problem: When economics meets the streets

The outlet emerged in 2021 during social unrest triggered by a controversial tax reform. Many Colombians feared its impact on their finances but struggled to grasp the technical details. “We wanted to bridge the gap between how the government and traditional media explain economics and what people understand from it,” says Camila González Olarte, CEO and Co-founder.

Initially, their content on social media answered basic questions and proved highly successful, but these mostly reached urban, internet-savvy audiences. To connect with wider groups, the team launched a website and eventually developed the chatbot.

Building the solution: From social media to structured intelligence

The chatbot was designed around the questions people actually ask. The team mapped recurring queries from social channels, validated them through audience research, and created a database of more than 900 answers. This was enriched with official requests to Colombian government entities, ensuring the responses reflected accurate and current information.

Tools and technical implementation

The chatbot’s architecture combines natural language processing with direct access to Colombia’s official datasets, particularly from DANE (the National Administrative Department of Statistics) and the Central Bank. This allows it to provide real-time indicators rather than static explanations. Responses are grounded in Colombian terminology and complemented with visuals, graphs, and links to related videos.

The team developed two distinct AI models to power the chatbot's response capabilities:

• RAG Model (Retrieval-Augmented Generation): This model finds and communicates the most relevant text information based on their curated question database. The RAG approach enhances response quality by retrieving specific knowledge from their extensive collection of pre-written answers before generating the final response, ensuring accuracy and relevance.

• Datasketch's Proprietary Natural Language to SQL Model: The partnership implements a specialised model that converts natural language questions into SQL queries. This capability allows the chatbot to directly query public databases in real-time, providing up-to-date economic data from official sources.

Building the right team

Around 12 professionals from Economía para la Pipol and Datasketch collaborated on the project, bringing together journalists, data scientists, developers, and designers. This mix of editorial and technical expertise ensured the tool was accurate, user-friendly, and trustworthy.

Navigating complex challenges

The project encountered several significant obstacles that tested both technical capabilities and strategic thinking. One of the most persistent challenges has been accessing official information and connecting new government databases to the chatbot. In Colombia, information access laws require formal legal procedures for many data requests, creating delays and complications. The team must navigate government communication channels and freedom of information processes to obtain the real-time data that makes the chatbot valuable to users.

Another key challenge has been understanding user language patterns. Despite Economía para la Pipol's experience with question-based content, the team discovered that people formulate queries differently in a chatbot environment compared to social media searches. This required continuous refinement of the database and question interpretation algorithms, face-to-face testing and online sessions with audience members to gain insights into how users naturally ask economic questions. These findings led to significant adjustments in their approach.

Finally, the chatbot has to position itself in a market already dominated by ChatGPT, Gemini, and other general-purpose AI tools. Testing, however, revealed its unique value proposition: users appreciated the specialised economic focus and trusted Economía para la Pipol's journalistic approach. They particularly valued the integration of visual content, including graphs and videos from the outlet’s social media, which created a multimedia learning experience not available in generic AI tools.

The opportunities: Future opportunities and strategic vision

The chatbot project has revealed several strategic opportunities that extend beyond its immediate functionality.

• Sustainable revenue models: With media grants becoming increasingly scarce, the chatbot represents a potential path to financial sustainability. The team is exploring partnerships with organisations that need economic information translated for broader audiences, as well as a WhatsApp version.

• Expanding the content ecosystem: A single piece of research can be transformed across multiple platforms and formats, maximising the value of journalistic work by creating content that serves audiences across different preferences and consumption patterns.

• Deeper audience understanding: The chatbot development process has provided unprecedented insights into Economía para la Pipol's community. Beyond social media metrics, the team now understands why people seek economic information and what specific concerns drive their questions.

Lessons for newsrooms

• Technology as a journalistic multiplier: The chatbot project demonstrates how AI can amplify existing editorial strengths rather than replace journalistic expertise.

• Partnership success requires shared values: The collaboration between Economía para la Pipol and DataSketch succeeded because both organisations shared fundamental beliefs about information democracy and accessibility. Technical capabilities alone were insufficient; the partnership needed alignment on mission and values.

• User research reveals hidden assumptions: The difference between how people search for information on social media versus how they ask questions in a chatbot requires significant adjustments to content strategy and technical implementation.