The problem is evident: Most engineers don’t understand the burden that cloud spend presents to organizations, while most finance teams lack the technical savvy to forecast and control cloud spending.
Indeed, the cloud poses a particular problem with forecasting because of how it traditionally shifts the oversight of the procurement of IT services and infrastructure from finance and procurement teams to individual developers and engineers. In principle, this problem space is very similar to that which faced software development organizations leading to the proliferation of DevOps: the chasm between software developers writing code and the IT operations folks responsible for implementing and maintaining it led to massive complications in forecasting cost, schedule, workload, and quality.
DevOps integrated these disparate practices with outstanding results. With developers and IT operations acting as one team, organizations push products and services into customers’ hands quicker and in closer alignment to the forecasted cost and schedule goals. FinOps is transforming the finance of cloud services the same way that DevOps transformed IT development and operations by providing greater traceability between cloud cost drivers and business outcomes.
According to the State of FinOps Report 2021, the average FinOps team has grown by 75 percent in the last 12 months. With 50 percent more growth anticipated in the next year, it’s clear that organizations recognize how the convergence of people, culture, technology, and processes drive value in financial operations the same they do in software development and IT. These organizations ramping up their FinOps competency understand a direct correlation between cloud usage and their organization’s need for mature FinOps practices. Despite the apparent progress in the field’s ubiquity, there remains a significant opportunity for growth and maturation. Out of surveyed participants, only 15% of organizations report a mature and evolving FinOps practice, while over 43% report that they are only beginning to grasp the basics. The organizations that get this right will gain the capability to model, forecast, and control cloud spend proactively rather than just reacting to developer decisions and customer needs.
The Role of Forecasting in FinOps
When aligning business outcomes to cloud spend, accurate financial forecasting is the difference between projects that drive value and those that harm the business. Unfortunately, forecasting for the cloud is even more complex than traditional financial management for IT. In the past, financial managers had the luxury of clear procurement plans focused on acquiring infrastructure, licenses, and even staffing from vendors with transparent pricing models and relationships built over many years and historical data to provide guard rails for future cost estimates. To take full advantage of the cloud, organizations need to empower engineers, developers, and operators to leverage the elasticity of service offerings. But with that comes a loss of control and visibility. It’s not even always possible to rely on historical data to mitigate these concerns because capacity and cloud service needs vary significantly based on project requirements.
In some cases, as with companies that use consumption-based revenue models to make critical decisions on pricing and architecture, inaccurate forecasting can lead to missed growth opportunities or revenue shortfalls. A straightforward way that this can happen is through the use of serverless architecture.
One AWS customer migrated his website to a serverless AWS stack. He cited the ability to autoscale “for that off chance something goes viral” as the primary reason. 2 weeks later, he received a notification stating that his actual costs were 41x higher than forecasted. AWS now projected his charges to be 12x higher than predicted for the remainder of the month. Thankfully, this was a small project, and the damage was in the hundreds rather than the thousands or tens of thousands it could be for a medium-sized business.
This example shows how poor forecasting can lead to an overreliance on expensive elastic solutions in the cloud, but there is another side to this pendulum. Failure to forecast high demand for services can also lead to server overloads and outages. A study from the Ponemon Institute suggests that the cost of an outage is an average of $9,000 per minute for large companies.
Accurate forecasting empowers organizations to right-size project budgets and builds trust with the key stakeholders that hold the purse strings. But what constitutes precise forecasting? That depends on who you ask. The State of FinOps Report categorizes organizations by their self-identified FinOps maturity level: Crawl, Walk, Run. ‘Crawl’ organizations generally stated they expect to see forecasting accuracy within 20 percent of actual expenditures. ‘Walk’ organizations look to see 15%, while ‘Run’ organizations consider 12% a relatively accurate forecasting model. These acceptable variance levels demonstrate how much volatility is taken as a matter of course in the cloud spend universe. It makes sense then that 26 percent of respondents to the survey highlighted forecasting as the utmost challenge facing FinOps teams regardless of organizational maturity level.
One of the most profound benefits of better forecasting in a FinOps context is its impact in empowering product teams to stay within acceptable budget ranges while still taking advantage of the cloud’s on-demand elasticity.
The importance of machine learning and AI for demand forecasting
One of the most difficult challenges facing FinOps practitioners is in demand forecasting. As engineers and developers iterate through the development lifecycle, designs and system architectures mature over time until it is relatively easy to predict the required services and technology needed to build and maintain the system. However, overall demand (in the form of concurrent users or service consumption) can spike in unanticipated ways. It can be challenging to predict how unexpected usage patterns will impact which services and infrastructure components the hardest. On the opposite end, some companies try to build this uncertainty into their infrastructure design, leveraging native elastic services offered within cloud service providers like AWS. These solutions are helpful but are far from a silver bullet, often incurring a high overhead cost for their usage.
Many companies are turning to AI and machine learning as a complement to sound FinOps forecasting methodologies and the native offerings of the cloud. In truth, it’s more than a compliment: It’s a requirement. What AI brings to the table is continuous forecasting throughout every phase of the project lifecycle. These forecasts impact architecture and consumption in very tangible ways. Take, for example, in AWS, where an accurate forecast can facilitate reserving an instance for a year rather than relying on an on-demand elastic service which is significantly more expensive. In addition, some resources are so granular and some patterns so complex that it is challenging for traditional manual methods to identify the impacts in time to implement a meaningful decision. AI-enhanced forecasting methodologies can account for actual usage and small changes in patterns that fundamentally alter how a cloud-based system should be architected to maximize performance and minimize cost.
Recently, a company used a combination of tagging individual AWS services to monitor their specific usage and cost and a machine learning process capable of analyzing data processing within their S3 buckets. These actions helped identify and eliminate a constraint while also significantly improving the accuracy of their cost forecasting. The result was cost savings of over $360k annually in cloud-spend. Outcomes like these can scale dramatically. The more an organization spends in the cloud, the greater the benefits solutions like these can provide.
If there is one clear takeaway from the State of FinOps Report 2021, the field is snowballing as more organizations expand their infrastructure (and cost) footprint in the cloud. It’s encouraging how many organizations are in the ‘crawl’ stage of FinOps maturity. It will only enhance the ecosystem of technology, processes, and professional competency to enhance critically important financial management practices like forecasting to improve business outcomes related to the cloud. That time is now. There are already AI and machine learning tools capable of empowering forecasting to help businesses achieve growth and meet demand with the same elasticity of the cloud.
Shay Lang is VP R&D and a founder of Anodot. He has spent his career directing R&D and product engineering teams in the software and cybersecurity space, for companies such as Trustwave, M86 Security, Finjan Software, and Voltaire. He holds a bachelor’s degree from Technion – Israel Institute of Technology and an Executive MBA from Tel Aviv University.