Founded in 2010 by sisters-in-law Veronica Miele Beard and Veronica Swanson Beard, Veronica Beard is a uniform of cool classics that speak to today’s multifaceted, multitasking woman. The collection is built for her layered life – so she can look good, feel good and do good. As the company exists today, there are 3 ways that they distribute their products - through their website (i.e. eCommerce), self-owned retail stores, or 3rd party retail stores (i.e. wholesale).

Going back to the CSV munging - Localization of data isn’t a problem for small teams who can quickly share their screens to isolate any discrepancies, but at this point, Veronica Beard had been so successful  that the team had grown to ~200 employees and a desire to maintain that growth, using data from their online presence to back decisions.

Enter Max Lagresle - Veronica Beard’s first data hire and currently the Director of Data. Max was originally brought in to implement Segment as a standalone Customer Data Platform to help the eCommerce and Marketing teams better understand how to drive more sales. In the process of setup, he found that there was more opportunity for Veronica Beard to grow through the use of data. One notable example that stands out is the need for leadership to understand company wide sales. They were willing to put up with one entire week of latency required of financial reconciliation across those CSVs.

Setting up a net-new data stack

It’s hard to get budget for new initiatives and tooling. When Max first joined Veronica Beard at the end of 2019 and pitched a data stack implementation, he had to answer “How do we justify an investment in data?” Max was kind enough to break this question down into: 

• What’s the scope of this project and who are the owner(s)?
• What human resources and technical skillsets do we need for this project?
• Which tools are you recommending, and what’s the order of implementation given their use case(s)?

Data Stack Scope + Ownership

Everyone at the company had questions related to their work, as evidenced by the heavy utilization of Excel, so it’d be natural to try to serve every potential internal customer. Max recognized that path, however, would’ve required many more resources than was available at the time. Instead, he decided to start with the initial use case that he was hired for, which was to support the e-commerce and marketing teams to help them better understand customer behavior.

❝ It’s not realistic to pitch a data strategy to every department at the start. We took a steady approach to rolling out additional data analytics project and recommend starting with 1-2 departments that have analytics needs.”

Over the course of a year, Max was able to successfully implement a warehouse to answer recurring questions and address questions that were previously too cumbersome to answer through Excel. Fast forward 4 years, and in addition to the eCommerce and Marketing teams, the data team now also supports 6 additional departments (CRM, Planning & Buying, Retail, Merchandising, Finance, Wholesale).

❝ By starting with the ecommerce and marketing departments, we were able to prove that we could answer their questions and use cases. In the process, we were also able to demonstrate how much easier it was to answer questions that couldn’t even be answered before setting up a warehouse, such as the difference in lifetime value between single-channel customers (e.g. ecommerce or retail) vs multi-channel customers (i.e. omnichannel).”

Resources 

If you recall, when Max first joined, there wasn’t a data team. He started off by reporting directly to the Chief Marketing & Digital Officer. There were understandable concerns about whether he’d be able to deliver on such a broad project. Through proactively proposing work, executing successfully, utilizing their agency when needed, and steadily rolling out more requirements, Max grew his team from just himself to onboarding 3 additional team members.

Implementation timeline

Max also implemented several other tools early in the process of standing up Veronica Beard’s data stack:

• Fivetran - This was needed to offload ingestion pipeline management so that they could accurately capture raw data from sources such as ecommerce, POS, or site systems.
• dbt - After ingesting data, Max wanted a more repeatable way to model, with the understanding that they’d want to scale this process out to the rest of the company over time.

After handling storage, ingestion, and modeling, Max quickly implemented Sigma Computing for their spreadsheet-like interface used to explore data, knowing that this would be familiar to the company’s existing Excel power users.

❝The beauty of implementing a modern data-stack is that it helps building a highly scalable data infrastructure with limited resources."

As the team continued to see success, they continued learning about new ways to improve their work - including learning about Data Observability.

Metaplane to guarantee data quality

At this point, Max and the growing data team are trusted by multiple departments to handle questions such as:

• How do we think about planning development? What should store inventory levels be at for maximum opportunity capture?
• How can we improve company operational efficiencies and spend money wisely?
• What sort of product lines have done well and how should we plan for the future?
• Where are the biggest opportunities for store opening?

As you can imagine, these questions had direct impacts on the company’s bottom line, which was why they began to evaluate data observability solutions to ensure that questions were answered accurately.

Like other teams, they had already been using dbt tests alongside their model builds, but ran into problems with:

1. Scaling the number of tests that they had, both in terms of types of metrics and coverage for all of their new models to support additional departments.
2. Interpreting the nuance in results so that they wouldn’t just be alerted to whether there was a null value, but seeing what % of null values would be considered normal, given past history.
3. Data quality issues being created throughout the day, but only being tested for at the beginning of each day during a dbt build.

❝We still use dbt tests, but often with the raw data, Metaplane is always catching it earlier. Because our dbt builds run roughly once a day, that limits how often we can catch errors.”

Despite evaluating other vendor solutions, Max and his team ultimately chose Metaplane due to being able to solve for the issues above, commercial agreement flexibility that could accommodate Veronica Beard’s growth, their interactions with the Metaplane team, and the ease of use in the product.

❝We wanted a solution that we could grow with, and the Metaplane team and product has been amazing to work with so far. Metaplane's team are true experts in their domain and data industry in general and their white-glove onboarding has been truly amazing.”

Metaplane Benefits

Now, more than a full year after their initial Metaplane implementation, the data team at Veronica Beard has taken advantage of all of the new offerings that have come out since they’ve started. In addition to our core feature, data quality monitors, Max specifically mentioned:

• Data CI/CD - This is the feature that broadened his initial scope of data observability evaluation beyond just data quality monitoring. With a native Github app, Metaplane is able to forecast how downstream tables and BI dashboards would change given an update to a dbt model.

❝There’s some benefit in ‘peace of mind’, which can’t be measured, when merging a dbt PR (pull request) and knowing that you won’t be disrupting anything downstream. It can affect how much sooner you’re able to merge with that increased confidence.”

• Lineage  - Metaplane uses metadata to generate column level lineage so that users can understand where their ingestion tool is loading data down to which business intelligence dashboards are impacted by a data quality incident.

❝With lineage, we can see which Sigma dashboards are potentially affected and ‘reverse engineer’ it to figure out which table was responsible.”

When those features were used in conjunction with the machine-learning based monitors, the biggest benefit that Max spoke to was trust.

❝For me, my main measure of success for business stakeholders is ‘trust’. Data is all about trust. We’ve invested so much time and budget in building the single source of truth. Every time the business users identify an issue that the team hasn’t proactively found, we naturally lose a little bit. It’s not a problem with making a mistake; everyone makes mistakes. But being able to tell others (about an issue) first can actually flip the situation and increase trust, which Metaplane’s helped us with.”