HomeCase Studies
Slice Brings Machine Learning to Pizza
Fivetran
HomeCase Studies
Slice Brings Machine Learning to Pizza

Slice Brings Machine Learning to Pizza

With Fivetran and Databricks, Slice reallocates the efforts of three data engineers to mission-critical projects and adds a data science team.

By Ciara Rafferty, December 7, 2020

Key Takeaways

  • Three data engineers move their focus from ETL to projects that will grow the business

  • With Databricks, Slice focuses on machine learning and hires a Director of Data Science to build a team

  • Current/upcoming projects: personalization with a trait service, CRM messaging capabilities, building the restaurant operating system and more

  • Data team commits to a 3-5 minute SLA, but achieves just milliseconds between production and reporting with a modern data stack

Modern Data Stack

Slice gives pizzerias the specialized technology, data insights, loyalty marketing and shared services they need to serve customers. The business makes it easy for customers to order pizza via the Slice app or website while meeting its mission of serving small businesses and communities. Headquartered in NYC, the business has expanded operations globally, with offices in Belfast and Macedonia and 700+ global employees.

Behind every slice of pizza ordered, delivered and enjoyed, lives a host of technology and data. In the past two years, Slice has been laser-focused on data strategy; its team has built a data warehouse and a delta lake with Databricks. Achieving this required moving significant amounts of data between systems.

Scaling Data Pipelines With Fivetran

At first, the business was writing and maintaining code for its data pipelines internally. If something failed, it required manual intervention. Big plans for 2021 – including going deeper into restaurant operations and owning point of sales – will dramatically increase Slice’s data. To meet these growth goals, Slice needed a scalable solution so data engineers can work on solving business challenges, not moving data.

Jason Ordway, CTO at Slice, and a few other employees had used Fivetran at a previous company and knew the solution would help the business scale:

I re-engaged with Fivetran, did the math and it just made sense. We’d rather use humans to solve challenges unique to our business. Our director of data engineering in Belfast did a POC and it was a no-brainer. It saves the time of up to three full-time engineers.

Slice plans on adding the Fivetran connector for MySQL, its master data store. The business is rolling out new services that each have their own data store, including a menu service, shop service and order service, which will need to scale quickly as Ordway explains:

Each of our data stores are going to scale exponentially. It would take far too much time to write code to move data from different databases every time we launched a new service. The point-and-click with Fivetran fixes a multitude of issues and keeps our AWS bill reasonable by eliminating coding mistakes.

Building a Data Lake

When Ordway joined Slice over three years ago, the “data warehouse” was really an unstructured database of multiple data sources – “a dumpster fire of data” as he puts it. Once the data warehouse was structured and in a good place, the team was able to begin building a data lake. Fivetran brings the data into S3 buckets, which the data lake runs off of. Ordway explains the functionality:

Let’s say I want to compare last month’s order data to last week’s order data in Fairfield, CT between 7:00 pm and 10:00 pm. Databricks spins up the servers to run the queries and creates different, dynamic views of the data. You can do your reporting and then, when you’re done, all the servers that were spun up shut down. More importantly, it gives you the ability to go back and forward in time and make comparisons, without beating up your warehouse.

Machine Learning and Predictive Analytics

With the stack in place, Slice hired a Director of Data Science who has hired two machine-learning scientists and to focus on machine learning. The first step is personalization with a trait service for its apps. Slice will know what customers’ favorite shops are, what slices they like, and make recommendations based on their behaviors. The business is also looking into CRM capabilities around messaging. For instance, if an individual typically orders on Friday around 4:00 pm, this may prompt a message to place an order at their favorite shop leading up to that time, which Ordway is very excited about:

The trait service is our first foray into personalization as a company. We’re definitely just scratching the surface with all of this data. Our data team is building and training these models and pushing the output into the consumer applications. It’s amazing.

A More Reliable Infrastructure for Business Intelligence

Having accurate data impacts everything, from reporting on top-line metrics to employee compensation to board expectations. When one piece of the data puzzle is missing, it impacts the entire organization.

We’re very sensitive about our data and its freshness. Luckily, we’ve built an infrastructure where, at worst, we have a delay rather than a failure. When we built the data lake, our goal was a five-minute delay between production and reporting, and our SLA is 3-5 minutes. But what we’re seeing today is milliseconds between production and reporting.

With Fivetran and the rest of the modern data stack in place, reporting is more up-to-date and there is no longer a downstream effect when a job or code fails. Previously, a failure would have required human intervention and led to delays in reporting. Engineers are building out the restaurant operating system, the consumer experience, and predictive modeling – not managing data pipelines. This is a huge win for Ordway:

Most importantly, our humans are focused on building stuff. We’re excited about what 2021 looks like because we can build an amazing product based on data. Knowing what Fivetran does, why would we ever try to do that internally?

Learn how Fivetran and Databricks are solving some of the biggest challenges in data in this blog post, learn how to set up the Databricks connector in this quick demo or experience the power of the modern data stack for youself with our free trial.

About Fivetran: Shaped by the real-world needs of data analysts, Fivetran technology is the smartest, fastest way to replicate your applications, databases, events and files into a high-performance cloud warehouse. Fivetran connectors deploy in minutes, require zero maintenance, and automatically adjust to source changes — so your data team can stop worrying about engineering and focus on driving insights.

About Databricks: Databricks is the data and AI company. Thousands of organizations worldwide — including Comcast, Condé Nast, Nationwide and H&M — rely on Databricks’ open and unified platform for data engineering, machine learning and analytics. Databricks is venture-backed and headquartered in San Francisco, with offices around the globe. Founded by the original creators of Apache Spark™, Delta Lake and MLflow, Databricks is on a mission to help data teams solve the world’s toughest problems. To learn more, follow Databricks on Twitter, LinkedIn and Facebook.

About Looker: Looker is the business intelligence (BI) and analytics platform part of the Google Cloud data and analytics suite. Transcending traditional BI, Looker powers data experiences that deliver actionable business insights at the point of decision and infuses data into products and workflows to allow organizations to extract value from data at web-scale.

Start analyzing your data in minutes, not months

Launch any Fivetran connector instantly.