Product descriptions:
Puma Energy is a downstream energy company that supplies fuel, lubricants and aviation products across Latin America, Africa and Asia-Pacific. With a purpose to “energize communities,” the business sought to predict fuel stockouts, generate safety audit reports and personalize loyalty offers, using the power of proprietary data and AI. Puma Energy invested in a unified data platform to overcome fragmented data, slow dashboards and manual workarounds to achieve these ambitions. These inefficiencies meant decisions lagged by hours, making it harder to respond quickly to customer needs and operational changes. After extensive evaluation, they chose the Databricks Data Intelligence Platform with Agent Bricks to power the next iteration of their business, reducing data latency by 98-99%.
Turning data into better customer and dealer interactions
Puma Energy, a downstream energy retailer, supplies fuel across its retail stations and to B2B customers. It’s portfolio of energy products includes diesel, lubricants, jet fuel LPG, bitumen as well as solar. With this broad portfolio, the company launched initiatives to improve planning, procurement, project management and customer satisfaction.
To identify optimal sites for new fuel stations, Puma Energy analyzed traffic and vehicle flow data in Africa, which is similar to how fast-food chains use footfall analytics. They also streamlined station construction and supply chains by tracking supplier performance and payments through a centralized system. Once stations were operational, the focus shifted to understanding customer behavior. By analyzing when and where customers refuel, Puma tailored loyalty offers and sent more personalized, timely app communications.
To capture sentiment and feedback, Puma leveraged generative AI to synthesize unstructured data from sources like social media, surveys and support tickets, creating monthly summaries with recommended actions. The same approach was applied to emails, invoices, and documents to eliminate back-office bottlenecks. They are also on a journey to use a multimodal GenAI model to analyze CCTV footage, detect policy violations and enhance safety across the energy sector.
Breaking down the barriers to scalable business insights
Puma Energy faced several obstacles in its data journey. While operating across regions had advantages, it also led to data silos. Business users often downloaded data from dashboards to work offline — 95% of dashboard use involved raw data exports — highlighting a failure in self-service analytics. As a result, insights were delayed, taking 4–5 hours to reach decision-makers. Even though the BI dashboards are refreshed every 30 minutes, the underlying data is updated only every four hours, creating misaligned metrics.
IT also built data platforms without enough business input, resulting in tools that lacked calibration and saw poor adoption in commercial teams. “Our teams wanted speed and control, but the setup made us choose one or the other,” said Tanay Tiwary, Global Head of Digitization at Puma Energy. “The infrastructure wasn’t built to scale how the business needed — everything required too much DevOps effort.” Even when access improved, it brought new challenges in cost, governance, and compute sprawl, with no clear path to scale self-service. Databricks ultimately provided the flexibility and fine-grained governance needed to give business users fast, secure data access for strategic initiatives.
Putting the right data infrastructure in place for GenAI
To support data democratization and governance, Puma Energy chose a centralized platform that prioritized speed, autonomy, trust and practical GenAI. Their Databricks journey began with Delta Lake as the foundation for unified data storage and real-time analytics, powering use cases like inventory tracking and loyalty program optimization by centralizing data across regions and functions. This eliminated silos and ensured consistent, accurate data for both BI and GenAI applications, including fuel stockout prediction and mobile app personalization.
Unity Catalog then governed all structured and unstructured data in Delta Lake with fine-grained access controls, allowing teams to manage permissions across regions, business units and roles. This prevented data sprawl and unauthorized access while making self-service exploration safer and easier to scale. With Databricks SQL, teams could query live data, monitor performance and visualize trends in near real time — replacing the outdated practice of offline analysis. Crucially, teams could explore data freely within budget guardrails that limited compute usage, enabling broad access without infrastructure overhead or cost surprises.
Scaling AI for operations and customer impact
Once data management and governance were in place, Puma Energy introduced Databricks Agent Bricks to tackle high-impact operational and customer challenges with GenAI. They jump-started the initiative by running retrieval-augmented generation (RAG) — powered by Databricks Agent Bricks Vector Search — to extract insights from unstructured data, turning previously disconnected feedback sources like social media, surveys and support tickets into actionable intelligence. This analysis is fed into outputs ranging from monthly franchise reports to internal audit prep and voice-of-customer analysis. As internal adoption grew, Puma Energy leaned on Databricks Agent Bricks Model Serving and automated MLOps pipelines to scale these AI-driven processes. What began as small, low-risk pilots quickly evolved into fully operational systems that automated back-office tasks, like invoice processing, and delivered monthly field reports directly to business teams.
With a strong CI/CD setup, the data team could iterate quickly, while non-technical users received expedient and structured insights. Lakeflow Jobs, MLflow, and MLOps pipelines enabled Puma Energy to automate model development and deployment, allowing them to run multiple GenAI workflows in parallel and rapidly move from experimentation to production. Because they could now utilize a mix of models — including GPT, Llama and Claude — for different tasks, they applied this flexibility to leveraging multimodal GenAI for workplace safety through Guardian Vision, a tool that analyzes CCTV footage for policy violations. Tanay concluded, “We’re using the full Databricks Agent Bricks suite, from MLflow for experiment tracking and model management to the Agent Framework for orchestrating more complex GenAI workflows. The entire platform works together, allowing us to build, manage and productionize quickly and reliably. With all the backend components running seamlessly, our customers get a better experience, often without even realizing it.”
Reducing stockouts, costs and manual work
By consolidating its global data onto the Databricks Data Intelligence Platform, Puma Energy turned hours of waiting into near-instant insight — cutting data latency by 98–99%, from 4–5 hours to under 5 minutes. This speed allowed local teams to respond to issues faster, preventing operational disruptions and reducing fuel stockouts by more than 20%+ — a critical improvement for the communities they serve.
Operational processes became just as efficient. Safety and audit preparation that once took two weeks can now be completed in just one hour, enabling more frequent compliance reviews without adding headcount. And by retiring legacy infrastructure, Puma Energy reduced data costs by 15%, turning a former technical bottleneck into a source of savings and flexibility.
Beyond these immediate wins, currently governing data across 20+ countries has unlocked a new way of working. Business users no longer rely on static dashboards or manual workarounds; instead, they can self-serve insights, act faster, and collaborate on GenAI experiments built on a single source of truth. The result is a “low-risk, high-velocity” approach to innovation that fuels both operational reliability and customer satisfaction.
Driving next-tier automation and personalization with AI agents
Puma Energy is now working to further close the gap between insight and action. While AI agents can already identify patterns and recommend actions, such as offering a discount to a specific customer, executing those decisions still requires manual intervention. To remove this friction, the team is building reverse data pipelines that allow insights to flow back into operational systems, like CRMs and ticketing tools. The goal is to encourage agents to trigger actions directly, without relying on downstream integrations to catch up. As Tanay put it, “The vision is for Databricks to become the central highway that connects tools within the business, not just for data ingestion, but also for orchestrating decisions and outcomes across the enterprise.”