mrge completes | Data platform migration to support growth and operational efficiency. July 1, 2026 Sergio Sulca Executive summary. mrge, the most trusted performance marketing platform, needed to modernize, its data infrastructure to support massive data volumes, complex data pipelines, and growing operational demands. As legacy systems struggled with large-scale updates, data consistency, and operational stability, mrge initiated a migration from Clickhouse + BigQuery and to Databricks. Muttdata partnered with mrge's Data organization to accelerate this transition, improve data ingestion and transformation processes, and build a scalable lakehouse architecture capable of supporting both analytics and operational use cases. The result: a more reliable, cost-efficient, and scalable data platform - delivered on time and ready to serve as a foundation for future growth and adoption across the organization. About the company. mrge is the most trusted performance marketing platform, operating globally across the affiliate ecosystem. The company connects more than 5,500 publishers, 80,000 advertisers, and 200 affiliate networks, enabling e-commerce businesses to scale their sales through performance-driven strategies. Through brands such as MaxBounty, Shopping24, DigiDip, and Yieldkit, mrge has built a complex and distributed ecosystem that generates billions of tracked links per year, translating into more than €2 billion in additional e-commerce revenue for its clients. At this scale, data is a core business asset. The ability to ingest, process, and activate large volumes of performance marketing data is essential to delivering accurate reporting, optimizing campaigns, and supporting downstream business systems. The challenge. mrge operates in a high-volume, multi-source data environment, where data is continuously ingested from databases, APIs, files, and event-based systems. As the platform scaled, its existing architecture - based on a cloud data warehouse - began to show critical limitations. Data update complexity at scale. One of the most significant challenges was handling updates on large historical datasets. Updating transactions associated with older events (e.g., clicks from months ago) required heavy merge and upsert operations, which introduced: * Unpredictable processing delays * Pipeline instability and failures * Frequent need for manual intervention Additionally, ClickHouse did not guarantee strong consistency, meaning updates were not immediately visible. This directly impacted the reliability of reporting and business metrics. Costly workarounds and operational burden. To maintain consistency, mrge relied on full dataset refreshes across tables with 100M+ records, resulting in: * Hours of daily compute time * Increasing infrastructure costs * Hidden operational overhead (retries, monitoring, manual fixes) These inefficiencies made the platform increasingly difficult to scale. Fragmentation and growing complexity. The challenge was further compounded by: * Multiple data sources (databases, APIs, files, streaming events) * Integration of multiple acquired companies * Increasing demand for reliable data across analytics and operational systems Migration under time pressure. mrge had already begun migrating to Databricks, and by this stage most of the core pipelines, dbt models, and main tables had been moved from the legacy platform. However, the internal data team was balancing migration work with daily operational responsibilities. With a critical final stretch still ahead, internal teams identified a clear risk: they would not meet their migration timeline without external support, particularly to ensure business teams and stakeholders could properly adopt the new platform. The solution. Muttdata partnered with mrge as a strategic execution partner, working closely with internal Data Engineering and Infrastructure teams to accelerate the migration while improving the reliability, scalability, and efficiency of the data platform. Muttdata joined the project as a strategic execution partner in the critical final stage. By that point, the bulk of the technical migration was already advanced, and the priority shifted to closing the project on time and ensuring business adoption. Muttdata's role was to add technical capacity alongside the internal team, accelerate the final push, and focus specifically on the reporting layers: validating, correcting, and optimizing the dbt models that powered business reports, revenue metrics, and other critical downstream outputs. Building a scalable lakehouse architecture. The new architecture was designed following a Lakehouse approach, centralizing storage and processing in Databricks and implementing an ELT workflow with a Medallion Architecture data model. This enabled mrge to: * Ingest data from multiple heterogeneous sources * Process and transform data at scale * Serve data for both analytics and operational use cases (Reverse ETL) Data ingestion optimization. mrge's ingestion layer was redesigned to handle different data types efficiently: * Airbyte for: * Relational databases (MySQL, PostgreSQL) * APIs (e.g.,Salesforce) * Databricks Spark Jobs for: * Event-based data arriving directly in S3 (JSON Lines format) This hybrid approach allowed the team to: * Optimize ingestion strategies (full refresh vs incremental) * Reduce unnecessary processing * Improve scalability and cost efficiency Orchestration and data transformation. The platform uses a modern ELT stack: * Airflow for orchestration * dbt for transformations * Delta Lake as the storage format Data is processed through a Medallion architecture: * Landing | raw data * Staging | cleaned, normalized, deduplicated * Reporting | business-ready metrics and datasets This structure ensures: * Data ownership is central * Improved data quality * Scalable transformations Improving data consistency and update efficiency. One of the most critical improvements was enabling efficient updates on large datasets. With Databricks, mrge was able to: * Avoid heavy full table refreshes * Execute incremental updates more reliably This eliminated one of the main operational bottlenecks and significantly reduced pipeline instability. Enabling analytics and operational use cases. The platform was designed to support both analytics and operational workflows: * BI consumption * Tableau directly connected to Databricks * Reverse ETL * Data activation in systems such as Salesforce and DynamoDB This ensured that data could be used for reporting. Infrastructure and cost control. The platform is deployed using Infrastructure as Code: * Terraform for AWS and Databricks resources * ArgoCD for service deployment (e.g., Airbyte on EKS) Additionally, clear operational practices were established: * Managing external tables in S3 * Cleaning unused data * Controlling storage costs Closing the final stretch: reporting layer focus. Once the foundational pipelines had been migrated, the final stage focused on the layers closest to the business. Muttdata worked alongside mrge's data team to: * Validate and reconcile reporting models against the legacy platform * Correct inconsistencies surfaced during migration * Optimize dbt models powering revenue and business-critical dashboards * Ensure smooth adoption by analytics and business stakeholders This focus on the final mile was decisive in turning a technically advanced migration into one that was production-ready and trusted by the business. The impact. The collaboration between mrge and muttdata delivered measurable improvements across performance, cost, and operational efficiency. On-Time Migration Delivery mrge completed its migration within the expected timeline. With a small internal team balancing migration work and daily operations, the final stretch was the highest-risk phase of the project. Muttdata's involvement provided the added capacity needed to close the project on time and with business confidence in the new platform. Wrap-Up. mrge's data platform transformation was driven by the need to handle scale, improve reliability, and reduce operational complexity in a high-demand environment. By partnering with muttdata, mrge was able to not only complete a critical migration on time, but also address fundamental challenges in data consistency, pipeline stability, and cost efficiency. Today, mrge operates on a modern lakehouse architecture that supports both analytics and operational use cases - positioning the company to continue scaling its data capabilities with confidence.

Kickstarting AI-Powered Promotion Discovery at Modo. April 22, 2026 Tomas Pastore Executive summary. Mutt Data Group partnered with MODO to prototype an AI-powered promotion discovery experience and help their leadership team envision how generative AI could transform the way users find offers inside the app. At the time, MODO was exploring how emerging AI technologies could improve the visibility and usability of its growing catalog of promotions. Through the development of PromoBot, an experimental AI-powered assistant, Muttdata helped demonstrate how natural language understanding could power a new generation of promotion discovery tools. The prototype validated a powerful idea: users could search for deals using everyday language rather than rigid keywords. This early MVP played a key role in shaping the internal product vision that later evolved into SearchAI, Modo's AI-powered promotion search engine. About the company. MODO is a rapidly growing digital payments platform that connects banks, users, and retailers within a unified ecosystem. Focused on making it easier to pay, send, and receive money, MODO has become a core financial tool for millions of users across Argentina. As their commercial team expanded its network of promotional partnerships, one persistent challenge emerged: users struggled to find the offers that mattered most to them. The challenge. A Problem of Visibility, Relevance, and Engagement Making promotions easier to discover. Modo offers hundreds of promotions across banks, retailers, and product categories. However, the existing search experience inside the app was designed around traditional keyword-based logic. This meant that users often needed to know exactly what to search for in order to find relevant offers. Slight variations in spelling or phrasing could lead to empty results, making discovery difficult. For example, a user typing "supermark" instead of "supermarket" could receive no results at all. As the number of promotions continued to grow, MODO's team began exploring how artificial intelligence could improve the experience and make promotions easier to discover. The opportunity was not only to improve search, but to explore how generative AI could act as a recommendation assistant inside the app. MODO wanted to test whether AI could: - Help users discover promotions they might not otherwise find - Increase engagement with related promotional suggestions and AI-generated comments. - Demonstrate internally how GenAI could power future product capabilities To explore this vision, MODO partnered with Muttdata to develop an MVP. Its solution. Prototyping AI-Powered Promotion Discovery with PromoBot To explore how generative AI could improve promotion discovery, Muttdata and MODO collaborated on the development of PromoBot, a chatbot prototype powered by Large Language Models. PromoBot allowed users to search for promotions using natural language instead of rigid keywords. Users could ask questions such as: "I'm going out with friends on Friday - are there any bar promos? or "Where can I get discounts with my bank card this week?" Instead of requiring precise queries, the system interpreted the intent behind the request and returned relevant promotions. This prototype allowed MODO's product and leadership teams to visualize how AI could fundamentally improve the promotion discovery experience. This proof of concept marked MODO's first step into generative AI. PromoBot included several experimental capabilities designed to test how AI could enhance the promotion search experience: Natural Language Search Users could search promotions using conversational queries instead of exact keywords. Contextual Recommendations The assistant interpreted the user's intent and suggested relevant promotions based on categories, merchants, and dates. Promotion Comparison PromoBot could compare two promotions and explain which one offered better value. Moderation Layer A safety layer ensured that responses remained appropriate and relevant. MVP search use cases. During the MVP phase, Muttdata and MODO focused on several key scenarios where traditional search struggled: Search by Bank Users could ask for promotions associated with a specific bank. Search by Date Queries like "tomorrow", "this weekend", or "next Friday" were interpreted and matched to valid promotions. Search by Merchant Mentions of specific stores or brands triggered relevant promotion results. Search by Product Type (SKU) Search by Events or Holidays Queries like "Christmas" or "Mother's Day" returned seasonal promotions. Architecture. Under the hood, PromoBot leveraged a cloud-native architecture powered by: - Large Language Models via Amazon Bedrock - REST APIs built with FastAPI - Containerized deployment using Docker and AWS ECS - Observability through OpenTelemetry and Datadog The system interpreted user queries, extracted structured filters (such as category, date, or merchant), and used them to retrieve and rank promotions from MODO's existing promotion services. From PromoBot to SearchAI. The insights gained from the PromoBot MVP helped Modo refine the product vision for AI-powered promotion discovery. As the concept matured, the team decided that AI should not live inside a chatbot interface, but instead power the core promotion search experience directly inside the app. After its collaboration, MODO continued evolving SearchAI and enhanced the search engine by incorporating semantic search. This optimization reduced LLM calls, improving latency, cost efficiency, and overall scalability, ultimately enabling the product to scale in production. The result of this evolution is SearchAI, a new generation promotion search engine now entering production. While the final product differs significantly from the initial prototype, the early MVP played an important role in demonstrating the potential of AI-driven promotion discovery and helping align leadership around the vision. The results. While PromoBot was developed as a prototype, the ideas tested during the MVP phase helped inform the design of the production SearchAI engine. These results validated the core hypothesis behind the project: that natural language search and AI-powered recommendations can significantly improve promotion discovery. Ready to unlock the full potential of your company? Do a discovery call with Mutt Data Group to discover how its expertise can elevate your capabilities. The Impact While PromoBot was developed as a prototype, the ideas tested during the MVP phase helped inform the design of the production SearchAI engine.