Count AI Beta is here!
Otrium is an Amsterdam-based fashion marketplace where brands can sell leftover inventory at attractive discounts.
There are 4 data teams at Otrium - data engineering, analytics, recommendations, and pricing. There are a total of 14 people across all 4 disciplines.
The data stack is Fivetran for ingestion, Databricks on AWS, dbt Core, Looker for core reports, and Count across the stack for engineering, analytics, and reporting.
“Count is an umbrella above everything we do from the beginning to the end of the pipeline.” - Veronika
Veronika Cucorova has been at Otrium for over two years in the pricing team, and more recently the data engineering team. She recently sat down with Taylor Brownlow to discuss how she and the Otrium team use Count to solve problems faster, communicate more effectively, and collaborate with the wider business.
See their full conversation here.
Many industries, like e-commerce and fashion, are very visual. To make the best pricing models and recommendation systems, you need more than data. Visually inspecting products and how they look on the page, or near other products makes for better pricing and recommendation models.
The Otrium data team regularly pulled up product images while doing an analysis, but it was confined to only those using Jupyter notebooks.
“Stakeholders needed to see images as well, so they would have to download the data, and open an image URL one at a time in a new tab. It was so slow and painful.” - Veronika
Count is one of the only tools designed for both qualitative and quantitative information. The canvas has both data cells - allowing teams to query in SQL and Python, but also an abundance of other objects - like images, embeds from other tools, and videos.
This is one of the first ways Otrium began using Count:
“We were able to display images of products based on a bit of Python code in the context of a wider piece of analysis. And it was all dynamic, so if someone wanted to look at a different subset of images it was so simple. We went from 70 tabs to one, which has made so much of a difference.” - Veronika
Otrium, like many data teams, uses dbt to run and maintain its data models. These models are often complex, and therefore, difficult to make sense of.
“This was a common scenario for us: You need to alter a model someone else has written before you, or maybe you wrote it a year ago or something. And it’s a massive piece of code, but that’s the simplest way someone could have done it. It might take you a day and a half to make sense of that model, and to do that you’re switching between 100 tabs in VS Code. It’s really frustrating but a really common issue.”
With Count, Veronika and the team can import any dbt model, and its upstream and downstream dependencies split that model into its component CTEs, and see the live results of those queries on the database.
This kind of transparency means they’re able to understand models in minutes instead of days, and they have a new appreciation for complexity:
“In the last year, everyone has gotten used to using Count for any kind of dbt model debugging, work-in-progress, or development of new models that we aren’t afraid of complexity any more. I know that someone else on the team can import it into a canvas, explode the CTEs and make sense of it quickly. Understanding our models isn’t an issue anymore.” - Veronika
It’s nearly impossible to do a piece of analysis, develop a model, or build a report without input and context from the business. Usually, that information is only gathered when a request is submitted to the data team, or when the dashboard/report/model is completed and live.
With Count, Veronika and the Otrium team were able to change that and make sure the data team was working with stakeholders, gathering feedback through the entire analytical workflow.
One project in particular demonstrates this change:
Veronika was working on a pricing model that predicted the inventory levels of certain products throughout the year. She created an initial prototype in Count and got feedback before productionalizing her code.
The stakeholder was able to quickly give context as to why some of the forecasts needed to be adjusted. Veronika made a new iteration of the model right next to the original, comparing the old model to the new one so she and the stakeholder could get to the right model, in a fraction of the time it would have taken with the old feedback cycles.
When scaled across all the work of the data team, this collaboration has a profound impact:
“People are way more involved now. Initially your stakeholders might feel like ‘we define what we want then it’s out of our hands,’ but now we have so many checkpoints before something get into production so they can still influence the final product to make sure it’s right. Count enables you to speak to each other in a way that works for both us in the data team, but also for non-technicial stakeholders.”
Start solving your organization's biggest problems with Count today.
