Data turnaround dropped from 40 minutes to under 60 seconds
Shifted 100% of data engineering team focus to AI modeling and ML use cases
Scaled data access to 150 non-technical staff members
Eliminated manual maintenance overhead through data as code
Enabled real-time pricing choices by blending market data
Kogan.com integrated BigQuery and Looker with conversational AI to unlock instant, secure self-service analytics organization-wide.
Kogan.com integrated BigQuery and Looker with conversational AI to unlock instant, secure self-service analytics organization-wide.
Migrating from legacy BI to an
engineering first foundation
Kogan.com, a pioneer in Australian ecommerce, scaled rapidly from a garage operation into a major digital marketplace. In their earliest days, the company kept storage costs low by dumping raw data into BigQuery as serverless CSV files, querying data only as needed. To analyze this data, they built out a business intelligence team that relied on multiple tools to explore data and construct fast, complex dashboards. However, this setup depended on source systems delivering pre-aggregated, specifically formatted report exports to keep the setup running quickly. If a single upstream export script failed, a whole day of data was completely missed, creating critical pipeline vulnerabilities and severe data reliability issues.
To eliminate this brittle architecture, Kogan.com restructured into a data-engineering-first organization migrating the core logic into the data layer itself. For dashboards, Kogan.com executed a strategic department-by-department migration, prioritizing heavy user bases, starting with their business-critical purchasing profitability dashboards. They funneled all raw staging data directly into BigQuery and introduced dbt to transform the raw tables into a structured analytics architecture using Kimball facts and dimensions. Because most of the underlying data was already cleansed and modeled during the dbt migration phase, transitioning the actual visuals into Looker was incredibly fast.
Our legacy setup relied on fragile upstream exports. If one script failed, we lost entire days of business intelligence. Transitioning to a Semantic Layer and Looker allowed us to model data as code with automated testing, turning a reactive ticket queue into a rock-solid, highly reliable data engine.
Goran Stefkovski
CTO, Kogan.com
Achieving true self-service through
conversational market intelligence
While centralized modeling ensured data integrity, standard and inflexible interfaces presented a hurdle for Kogan.com’s business teams. Users had to explicitly understand the interface, locate specific metrics, and drag and drop dimensions to simply build a graph. The true self-service breakthrough occurred when Kogan.com connected their curated BigQuery datasets and Looker’s semantic layer directly to internal AI agents through an enterprise Model Context Protocol (MCP) fleet server, giving users a simple text chat box interface.
This conversational layer completely revolutionized data access for marketing, purchasing, and marketplace departments. Instead of spending 40 minutes navigating fields, filters, and metrics, employees can ask a natural language question, such as asking for top trending products over the last three weeks by revenue or quantity sold. The AI immediately constructs a custom dashboard in a single minute.
Beyond standard numbers, users can seamlessly blend internal margin records with external market metrics. The AI agent can look up live competitor pricing across the web, cross-reference Kogan.com's direct margins, and instantly deliver next-best-action recommendations for agile pricing adjustments. Dashboards are also enriched with automatic AI commentary, advising sales or account managers on localized category drops and detailing steps to improve these metrics for their customers and business partners.
To protect sensitive records and avoid wild-west deployments, Kogan.com wraps the open MCP architecture with an OAuth security layer. This zero-trust approach mandates user authentication before granting server access, allowing Looker to execute its native row-level permissions. Employees can only see and query the precise data subsets they are authorized to view within the core BI platform. The engineering team is now working to build additional skills that help these AI agents seamlessly push user-generated dashboards back into the main Looker interface for permanent, shared corporate usage.
Beautiful UIs mean nothing if users lack the time to navigate them. By exposing our LookML layer to AI agents via MCP, we turned complex analytics into a simple text chat box.
Goran Stefkovski
CTO, Kogan.com
Creating a highly optimized data pipeline
By unifying dbt and Looker, Kogan.com created a highly optimized, cost-effective data pipeline. The dbt repository extracts data from source systems, selects the valuable fields, and builds scheduled fact and dimension tables directly within BigQuery.
For massive data volumes, Kogan.com launches batch processing jobs inside Google Cloud to run transformations directly under BigQuery, keeping cloud processing costs strictly optimized. Metadata and column definitions managed in dbt flow directly into BigQuery and surface automatically inside Looker as field descriptions.
Kogan.com stores their dbt models, LookML files, and dashboard configurations together within code repositories. This completely unified data-as-code approach means the entire infrastructure runs on continuous integration jobs with automated testing and observability alerts. If a data stream drops in volume or encounters anomalies, the system instantly logs the issue and generates precise alerts, entirely eradicating the manual maintenance overhead and day-to-day data gaps that plagued their old legacy environment. This high-performance architectural stability has successfully freed the data team to pivot away from dealing with broken pipelines and manual report queues. Instead, they focus 100% on high-value predictive data science projects, such as building an advanced machine learning model for price elasticity that directly helps Kogan.com offer the most competitive product pricing to their customers.
Unifying dbt and Looker on Google Cloud completely removed our maintenance burden. It freed our data team to build advanced price-elasticity machine learning models, while effortlessly transforming 150 non-technical team members into independent, data-driven vibe coders who generate their own operational insights daily.
Goran Stefkovski
CTO, Kogan.com
Removing the friction of complex query tools has sparked an exponential explosion of data engagement among Kogan.com's roughly 160 local employees. Within three days of launching the chat interface, staff across multiple operational teams were autonomously discovering unique, hyper-localized insights that previously would have taken weeks to clear the central IT backlog.
The magic of offering clean, reliable data with the power of generative AI means Kogan.com team members who know their customer best are the ones able to access insights at their fingertips.
