FinOps tips to ensure valuable data spend

Presented at Snowflake Data Cloud Summit 2024 by Patrick Barch, Sr. Director, Product Management, Capital One Software and Farial Shahnaz, Sr. Director, Software Engineering, Capital One.

From leveraging AI to making impactful business decisions, data plays a central role in the success of businesses today. Companies relying on data to enhance customer experiences or fuel other aspects of the business require a thoughtful and well-managed approach to cloud and data spend. 

Financial operations (FinOps) is a concept that emerged to help companies manage their cloud spend with the goal of maximizing the business value of the cloud. At Capital One, we found it can also apply to data infrastructure to ensure data spend is valuable to businesses. The combination of data growth and greater interest in AI increasingly means companies need to prove the value of their data spend. We presented at Snowflake Data Cloud Summit on how a FinOps strategy can be applied to data infrastructure, bringing teams together in a shared accountability model that leads to greater visibility and control around your data spend. 

At Capital One, we set out on a tech transformation journey more than a decade ago. Among many big decisions, we went all in on the public cloud, modernized our data ecosystem and adopted Snowflake. Our move to the cloud brought the benefits of instantly scalable capacity, easy-to-access infrastructure, and built-in security and resiliency. But operating at scale can also present challenges managing cloud spend. That’s where FinOps can come in as a solution for many companies. 

The FinOps Framework is defined by the FinOps Foundation as “an operational framework and cultural practice which maximizes the business value of cloud, enables timely data-driven decision making, and creates financial accountability through collaboration between engineering, finance, and business teams.” 

There is no direct mention of cost savings in the definition. Rather, the key words that stand out are business value, accountability and collaboration. FinOps at its core is about maximizing the business value of the cloud.

While FinOps is traditionally a practice for cloud infrastructure, FinOps principles can also be applied to ensure valuable data spend. In talking about data spend, we are referring to the subset of cloud spending that comes from storing and processing data.

Over the years, Capital One has leveraged Snowflake at scale to now support more than 50 petabytes of data and thousands of analysts running more than 4 million queries per day. We have a large cloud budget, 40% of which is allocated to shared data platforms. When our lines of business review bills for their usage at the end of the month, they want to know: Is that cost good? Is it bad? What did I get for that money? This scale requires us to have a strategy that helps us measure the value of our data spend. Because the cloud enables teams to leverage nearly unlimited amounts of data and data continues to grow, stakeholders will need to bring value-based, thoughtful approaches to the conversation on data spend as they scale, especially when making advances in AI.

At Capital One, we’ve seen how applying FinOps principles can provide more visibility and control within our data environment. So what does it look like to apply FinOps to data spend? A FinOps strategy can be applied to managing data to bring together technology, finance and business teams in a shared accountability model that ensures data spend is valuable. 

At Capital One, we organize our cloud FinOps strategy around four pillars that drive engineering excellence through shared ownership and accountability:

1. People across teams must work together

2. Governance and standards ensure controls are in place

3. Monitoring and analytics measure efficiency

4. Tooling supports shared accountability

Let’s review each of these pillars and how you could apply them in the context of managing a Snowflake environment.

A FinOps strategy requires people across teams to work together to achieve a shared outcome. To make this approach successful, we must understand what motivates each team persona.

A finance team member is responsible for managing the infrastructure and data budget along with tracking ROI. This includes allocating spend to the relevant business units for reporting. This persona must also identify and understand cost spikes, and reallocate the budget as necessary. An example is an enterprise FP&A (financial planning and analysis) leader looking to provide guidance on tech investments.

The technology team securely enables the business at the right cost, such as a technology leader responsible for the appropriate use of data platforms. This means understanding cost drivers and highlighting inefficiencies such as improperly sized infrastructure and poorly designed workloads. This persona also drives the remediation of those inefficiencies. 

Lastly, the business team is responsible for customer experience and optimal performance. For example, the team may include an application engineering leader responsible for new customer experiences. This persona brings an understanding of the impact of suggested changes on business processes and implements them when necessary quickly and easily.

The cloud has created new challenges across teams for managing data spend. Each team has unique, but related needs in service of their shared outcome. A successful FinOps approach requires consideration of what each team needs to do their job well. For example, business teams need a way to predict how much a new use case is going to cost them before they move forward with onboarding. Finance needs tools to forecast the monthly and annual data spend, along with setting a budget based on those predictions. Technologists must oversee the overall spend on their platforms, understanding the inefficiencies across the data stack and taking action to remediate them.

Often, this means technology teams are caught in the middle. As they work to run platforms efficiently and determine if spend is valuable, they hear from the finance team that the company is spending too much. At the same time, business teams are telling them that the spend is justified.

Technology, finance and business teams can come together to measure the value of data spend in a shared accountability model. The cost variability that comes with managing data in the cloud requires teams to work together frequently in real time for greater efficiency and making strides in business value. A shared accountability model means everyone takes ownership of data spend and shares responsibility in managing data effectively. But of course, this is not easy as the motivations of individual teams can conflict. Focusing on the four pillars of a FinOps strategy can help teams come together and ensure data spend is valuable to the business.

To be successful in this model, different teams need to work together continuously for daily monitoring of usage. Working across teams requires the right cadence of check-ins based on each team’s motivations and incentives. This may include weekly check-ins with the performing teams along with daily monitoring of usage with reach outs as needed. On a monthly basis, finance, technology and LOB (line of business) leads may engage in spend reviews with the right granularity applied to the reviews. For example, finance should refrain from sending each LOB a single line item bill for the platform.

Businesses need to ensure controls are in place in their data infrastructure to operate with speed and scale. The first step is to determine the governance and standards, which will then be enforced through tooling. This can look like the following:

• Governance ensures each warehouse has an owner accountable for usage and costs. Each warehouse also has a budget in partnership with the warehouse owner. Any usage increase or new user onboarding must go through the proper approvals and optimization review to check that the business is receiving the best value for the increased cost.

• Standards maintain the right configuration applied to each warehouse. This is made possible with tools such as auto-suspend idle warehouses that prevent businesses from being charged for dormant compute power. optimal warehouse size selection tailors the data user’s choice based on the use case. Retention policies and data ingestion patterns help support regulations and consistent data management practices across data sets and sources. 

The next pillar helps companies understand their efficiency, which means striking the right balance between performance and cost. By monitoring usage and providing analysis using tooling, teams can measure efficiency and determine what matters most for the organization. 

Checking elsewhere for answers on how to be more efficient most likely will not help – ultimately this is a decision unique to each business. For example, how important is it that this job runs fast? Organizations need to balance how much to pay with how fast they want their queries to run. 

Effective conversations can happen with business partners when you measure what matters most. To determine what matters most, we ask three questions to measure efficiency:

• Are our users operating efficiently? For example, users should be avoiding common SQL mistakes and large scanning jobs. Data may be published once a day, but refreshes could be happening multiple times in a day. 

• Is the infrastructure being used efficiently? Each warehouse should be operating at its optimal size while using the least amount of compute resources necessary. Consider efficiency signals that provide insights into whether bin packing is accounting for concurrency, queuing and spillage.

• Where can we innovate? Organizations can regularly benchmark against emerging technologies such as SnowPark and Iceberg Tables. Businesses can also consider if the execution for existing jobs is still efficient based on new features.

Lastly, tooling is necessary to support the shared accountability model by establishing the visibility and insights needed to evaluate performance and make strategic decisions. 

Good tooling should provide the following capabilities:

• Create and manage a cross-platform budget

• Forecast and track spend for new use cases

• Calculate the cost of data by business area

• Identify opportunities for improvement

• Enable self-service issue remediation

• Alerting as conditions change

At Capital One, we provided all of these capabilities through Capital One Slingshot, a tool that helps companies understand their cost, performance and usage of Snowflake, gives actionable recommendations to optimize resources, and streamlines capabilities to easily manage and scale. Slingshot also automates governance and enables proactive monitoring across different teams through a shared accountability model.

Putting all of these FinOps pillars together, let’s walk through a day in the life example of how teams can use Slingshot to implement FinOps principles for managing data spend. In our example, finance users are proactively notified of contract consumption, specifically that the organization has spent 95% of its contracts this month. A monthly breakdown of costs by business organizations helps to identify the costliest areas.

Tech users identify opportunities to increase efficiency after receiving an alert that warehouse spending increased by 200% this week along with recommendations on what to do about it. 

Tech users can also use tooling to identify the costliest costly queries and analyze the queries for how they can process more efficiently.

Business users provision new compute through tech-defined managed workflows, such as a request for approval of a data warehouse modification.

We believe there are three key points to remember in applying FinOps principles to managing your data spend:

• Understand the needs of the users and the business. Before solving a problem, you need to understand the personas for which you’re offering solutions.

• Develop a governance strategy and use tooling to implement. 

• Measure what matters most. In other words, help your partner teams prioritize. Creating KPIs that help the teams understand where to invest will deepen the partnership and lead to successful outcomes.

The need to account for data spend will only increase for organizations as data volumes grow and interest in AI-driven innovations and efficiencies rise. Applying FinOps to data infrastructure provides a way for companies to implement a shared accountability model to ensure business value from data spend while meeting the unique needs of various teams.