Improving ad campaign performance for clients around the world
Locala leverages the Databricks Data Intelligence Platform to build a strategic planning solution
To develop a GenAI virtual assistant using Databricks Mosaic AI
Locala helps brands and agencies efficiently plan, activate and measure campaigns personalized to the local consumer. Before leveraging Databricks, Locala’s data team of five faced significant challenges in developing, training and deploying machine learning models. The Databricks Data Intelligence Platform transformed their operations, enabling just one person to create a GenAI solution that could support Locala’s planning process by providing natural language explanations of advertising data. This innovation significantly reduced the complexity and time required to interpret campaign data, enhancing operational efficiency and client interactions. Databricks Mosaic AI tools allowed Locala to automate workflows, manage data efficiently and experiment with advanced models, leading to improved performance and cost savings.
Small data teams are tasked with time-consuming model training
Locala’s omnichannel advertising platform leverages granular insights and cutting-edge AI to help marketers plan, activate and measure campaigns personalized to the local consumer. Their platform specializes in transforming complex mobility and consumer data into actionable audience insights, fueling advanced media strategies that deliver smarter business outcomes. Furthering their commitment to innovation, Locala recently developed a GenAI solution to provide advanced, lightning-speed media recommendations to their clients.
One of their key advancements is the creation of a virtual assistant, dubbed Insights AI, that’s designed to enhance the insights and planning process. This feature allows users to input natural language prompts and receive explanations of various plots, graphs and charts related to advertising campaign performance. As Thibault Camper, Senior Data Scientist at Locala, described, “The idea was to develop an assistant to help our team more easily understand and share with customers how local insights can fuel media strategies. We want anyone to be able to see inside our platform to explain the figures and curves they see.”
This GenAI tool significantly reduces the complexity and time required for users to interpret and explain insights to clients, enabling more efficient and effective client interactions. Without it, they’d need extensive expertise in Locala’s massive internal platform to derive insights from the data — a lengthy and demanding process.
To build the virtual assistant, Camper knew he needed a comprehensive data management platform that could streamline the machine learning (ML) lifecycle and securely access Locala-related data in real time. “As you can imagine, being able to develop, train and serve models is a complex and time-consuming task for us,” Camper noted.
Camper and his small team of engineers and data scientists turned to Mosaic AI tools powered by the Databricks Data Intelligence Platform to simplify the process, allowing Locala’s small but mighty team to focus on GenAI innovation.
Leveraging Databricks Workflows to automate ML pipelines
With Databricks robust AI/ML lifecycle management capabilities, Camper could swiftly develop a virtual assistant. “We used the Llama 2 model for pretraining and fine-tuning due to its cost-effectiveness. At the time, we also thought it was the best model for our use case,” explained Camper. “Now we’re experimenting with other larger models. This is easy to do within the Databricks Platform because you don’t have to move the data.”
One of the significant benefits Locala experienced with Databricks was the ease of managing the entire ML lifecycle. They utilized MLflow for version control, experimentation tracking and model storage, greatly simplifying their workflows. Databricks Workflows proved instrumental in orchestrating and automating the scheduling of data pipelines for their ML workloads, including extract, transform and load (ETL) processes and ML model training. “We use Apache Airflow for our daily jobs, but it has limitations and doesn’t work well for large language model (LLM) training and hosting,” explained Camper. “We rely on Databricks Workflows to sequence these LLM tasks. The advantage of Workflows is that we can select machines we want to work with based on the task so the workload is optimized and precise. Selecting the right machines at the right time allows our varied workloads to be performant at a lower cost.”
This transition allowed Locala’s five-person data team to streamline their processes and improve collaboration with engineering. Camper finds the improved monitoring and observability to be helpful. He particularly likes being able to tag projects and easily see historical data runs and process duration. “Originally, I wasn’t using Workflows, and it was time-consuming and costly to manage our clusters. Workflows simplified everything, lowering operational costs while freeing up my time to focus on model training,” added Camper.
Databricks Model Serving enabled Camper to experiment with different versions of models to compare and benchmark performance effectively. The data they’re sourcing for model serving is all internal data. When they fine-tuned the model, they used a mix of datasets: ORCA, a general Q&A database with different prompts and contexts that provided a large variety of examples; a custom dataset with 30,000 rows containing definitions around media planning and how to leverage their platform; and use case data provided by internal experts. All of this is stored and managed in one location.
Throughout this build, Databricks Unity Catalog played a crucial role in data management and governance. “Unity Catalog helps manage the rights and access controls for the users globally — for Locala’s entire platform, not just for GenAI use cases,” noted Camper.
Using Databricks Mosaic AI to build a GenAI tool that generates more sales
“Databricks simplifies it all,” Camper stated, highlighting the ease with which Locala went from concept to production. “Databricks enables us to go from idea to realization in record time. The Databricks Platform has adapted quickly to the GenAI world and makes it easy for me to do my job.”
Locala has seen significant improvements in operational efficiency and how they service their clients. The adoption of their GenAI tool has allowed Locala teams to generate briefs faster and more accurately.
The virtual assistant has also played a crucial role in their promotional activities, serving as a key feature presented by their CEO at major industry conferences. Although precise metrics on time and cost savings are not yet available, the positive reception from clients underscores the tool’s impact. According to Camper, “It’s mostly a tool for our team to sell more. It’s a great feature that has really hooked clients.”
To make this even more impressive, Camper didn’t need to pull in an entire team of engineers and data scientists to build this solution. With the help of his business counterpart and a fellow data scientist, he was able to develop Insights AI in no time. Databricks not only enhanced Locala’s operational efficiency but also empowered their teams to deliver more impactful insights to clients — solidifying their position as a leader in location-based advertising solutions. Looking ahead, Locala plans to expand the capabilities of Insights AI to query more parameters and provide more accurate answers. To further enhance client value, “the goal is to move this model to RAG so that we can query information across more parameters and more available data to ask more pertinent questions with more accurate answers,” explained Camper.
To learn more about how Locala helps leading consumer companies plan, buy and measure local ad campaigns at scale, visit: https://asklocala.com/
