A data platform from zero, as the first technical hire.

Starting from nothing

Cedar Planters is a Canadian e-commerce company that makes modular planters. When I joined in January 2024 there was no technology team. My first job was to set up the basics any engineering department needs: CI/CD, security practices and Agile ways of working. Then I turned to data.

The company's data lived in many places: the e-commerce platform, analytics and marketing tools, advertising platforms and more. Reports were pulled together with third-party connectors and spreadsheets, each one built by hand, and a new report could take months.

The decision: one platform instead of many tools

I proposed moving away from the fragmented tools to one central data platform we owned. It would give us a single source of truth, standard reports, proper data governance, and lower running costs.

Don't wait for the perfect warehouse

Choosing a data warehouse is a big decision, and it deserves a real evaluation. But a long proof of concept can stall a project for months while the business still has no reports.

So we split the two problems. I built the ingestion pipelines to load data into a relational staging database straight away. That let us start collecting history and producing reports from day one, and decide on the long-term storage later, based on performance data instead of pressure.

In the end we didn't need one big warehouse. The platform settled on a mix, with each store doing what it is best at: a relational database for structured data, a document database for semi-structured data, and cloud object storage for raw and historical files.

The data platform

E-commerce platformOrders and sales

Analytics and marketing toolsTraffic and campaigns

Advertising platforms and moreAd performance and other sources

Ingestion layerOpen-source connectors for each source

OrchestrationLightweight, low-latency workflows

Relational databaseStructured data

Document databaseSemi-structured data

Object storageRaw and historical files

Central reportingFirst in a commercial BI tool, now in our own in-house reporting tool

Choosing the pipeline tools

We ran a proper research phase for the ingestion and orchestration layers, comparing established open-source pipeline and orchestration tools. They are powerful, but we chose a more modern, modular pair: a connector-based tool for reliable ingestion and a lightweight orchestrator for low-latency workflows. The combination balanced developer productivity and scalability in a way the older tools couldn't.

The change of ingestion tool came from inside the team. Our data analyst questioned whether our first tool was still the right choice, researched alternatives and made the case for a better one. I wrote about how that happened in From data analyst to data engineer.

Reporting people actually use

Once the data was in one place, we automated extraction and transformation and replaced the third-party connectors and spreadsheet reports with standard, central reporting. We started with a commercial BI tool, then moved to a reporting tool we developed in-house. I also redesigned the data models behind analytical and operational workloads, which made retrieval and analysis 20% faster.

The result is the number I care about most: a new report that used to take months now takes days. With AI-assisted development and analysis, many now take hours.

Leading a new team

• A team of three. A small team meant everyone owned real parts of the platform from the start.
• From tool operators to system builders. Moving away from off-the-shelf tools worried some people at first. I framed it as a chance to grow, and gave the team the training and time to feel confident with the new stack.
• Data visibility in every sprint review. Every two weeks we showed stakeholders progress on their reports. Their feedback decided what we built next, so the most useful reports came first.
• Hands-on first, then governance. I started 100% hands-on. As the team grew, I moved to setting the architecture and standards, and let the team own the implementation.

Results

• One source of truth, replacing third-party reporting tools and spreadsheets with standard, central reporting.
• Time to a new report down from months to days, and to hours with AI.
• 20% faster data retrieval and analysis after the data model redesign.
• Lower running costs from dropping paid third-party tools.