Generative artificial intelligence has established itself as a strategic tool for companies around the world. According to McKinsey, 65% of organizations already use it regularly, almost double the number just ten months ago. Additionally, more than 72% of companies have integrated some form of artificial intelligence into at least one business function, reflecting unprecedented global interest in this technology.
However, this rapid growth does not always translate into deep or broad use. The same research reveals that, although many companies have managed to implement it in one or two specific scenarios—such as marketingproduct development or customer service—very few have scaled the technology to multiple operational areas. In fact, only a small fraction of organizations report that generative AI significantly impacts their profitability, highlighting the difficulties in taking these solutions beyond initial testing.
The challenges to scaling generative AI are significant. Issues such as data quality and governance, lack of specialized talent, and risks associated with privacy and inaccuracy of results limit the reach of the technology. Additionally, many companies have not yet imagined their processes to fully leverage the capabilities of this tool, leaving them trapped in superficial implementations dependent on generic models.
Despite these challenges, the potential remains enormous. Companies that have managed to integrate it effectively report concrete benefits: reduced operating costs, increased revenue and a greater ability to respond to market demands. But the message is clear: to unlock the true value of generative AI, it is not enough to adopt it; It is necessary to scale it strategically, redesigning processes, forming teams and mitigating risks from the beginning.
Generative AI: let’s talk about a success story
At PiP Latam, a generative artificial intelligence project began with a curiosity that arose in the general direction: What can we do with AI? The question, simple but ambitious, led Carlos Solorio, development director of softwareand his team to explore the capabilities of this technology in a business context.
The team, made up of members from the technology areas and business areas, began by identifying specific operational problems that could benefit from this tool. In particular, they found a recurring challenge related to the quality of data extracted from key financial documents, such as prospectuses and contracts. These documents contained unstructured information essential to evaluating financial instruments, but capture errors, such as commas or misplaced periods, were affecting the accuracy of the processes.
“We knew that many artificial intelligence projects fail because the use cases are not selected well,” says Carlos. With this in mind, they opted for a disciplined approach, ensuring that the selected problem met three criteria: being meaningful, being well-defined, and having sufficient historical data to train and validate the models.
Identifying the use case, the team began their research by testing different language models, from ChatGPT to solutions on platforms like AWS Bedrock. The key was hands-on experimentation: They fed the models historical data to evaluate their ability to accurately extract critical information. In that sense, the participation of the business user was essential. Tonatiuh Hurtado, Deputy Director of Operations, and his team played a relevant role in the preparation of the prompts suitable for extracting information clearly and accurately from investment prospectuses. End-user involvement thus created a virtuous circle that quickly built trust and reinforced general management sponsorship.
#adoption #generative #companies
