Running a fintech company is a tough nut to crack if you can’t predict your Opex and are in a constant battle for your profit margins. You’re busy innovating your payment products, creating omnichannel experiences, or automating insurance processes.
As fintech companies rely on the cloud and containerization, cloud costs contribute a lot to that struggle.
I know what that feels like. At the end of each month, at my last company, we sat with our CTO and the executive team to try to determine whether we were spending the right amount on cloud resources.
Improving on fintech’s critical ability to scale
Back in December 2019, the trading app Robinhood received 100k requests per second at peak time. By June 2020, that number increased to 750k requests per second.
Can you scale up like that and manage a 7x load increase (or more) when Elon Musk tweets about your app? Or when a global pandemic suddenly inspires everyone to invest in ETFs?
You can prepare your fintech product for events like that and stay cost-effective at the same time.
5 problem areas for cloud cost optimization in fintech
- Planning and managing cloud costs – fintechs often experience high barriers to planning, managing, and reporting on cloud costs throughout a project’s lifespan and across several projects/teams within a time period.
- Allocating cloud costs to teams – this area is challenging as well, if you’re not sure which projects or teams have used which resources. How else can you tell which resources were required for the job and which ones were wasted?
- Forecasting cloud expenses – the way cloud providers bill your project might not fit nicely with your company’s quarterly or annual financial plan. Cloud costs rise and fall with demand, which is something you can’t predict with full confidence.
- Complex cloud service billing – the sales and billing procedures used by cloud infrastructure providers are hard to understand, even if you spend days analyzing your cloud costs. Just take a look at a typical cloud bill, and you’ll see that it’s complicated and hard to unpack.
- HR effort – For cloud optimization to work, a few engineers need to dedicate their time to it, which for you, means attracting and finding the right talent.
Best practices to address these optimization issues
1. Build a culture of cloud cost ownership
You probably run a modern, automated CI/CD pipeline where developers can easily test and deploy code to improve your product faster.
But are they aware of the costs when spinning up new virtual machines?
Cost management has never been part of their job.
Empower your engineering team to take ownership of cloud budgets. Give them a solution that provides full visibility into these expenses. Ideally, it should also take as many cost-related tasks off their plate as possible.
How? By automating them.
If you ask your engineers to consider all the 700+ instance types AWS offers before making a decision, they’re unlikely to act on this absurd request.
By equipping your engineers with a solution that picks and matches virtual machines to their application requirements, you’ll begin solving the cost problem without adding any extra work to their already busy schedules.
2. Prevent cloud waste with automated rightsizing and autoscaling
Many aspects of the financial world change dynamically. Remember that Robinhood story I shared in the intro?
Your product needs to translate a sudden influx of requests for buying stocks into executable operations, even if it’s a 7x increase over six months.
What you need is compute capacity that can smoothly adjust to dynamically changing circumstances.
This is where automated rightsizing and autoscaling can help.
When people are storming into your app, you don’t have time to pick the most cost-effective virtual machines. You need them NOW.
Software can take care of that. AI cost optimization can assess the most suitable VM type and size, then scale back down once the rush is over.
3. Prevent cloud sprawl with automated resource scheduling
But 9 times out of 10, companies end up spending more than they should. It can be shadow IT projects launched by your marketing or sales team. Or a bunch of virtual machines used for testing a new feature the team forgot to remove.
Some cloud optimization solutions offer resource scheduling. This means that they constantly monitor your infrastructure for resource use, provision new resources when necessary, and then scale them down when the rush is over.
4. Don’t fall into the trap of reservations or Savings Plans
Reserving capacity upfront makes a lot of sense to fintech companies. After all, it’s not like your team is suddenly going to move your application on-prem. And you get to enjoy great savings.
But consider this: you’re entering into a deal lasting one or three years. Both are eternity in the tech world – and especially in the cloud computing market.
A lot can change during that time, for example:
- The cloud provider might sunset a service or change their offer,
- They might add new instance types that match your use cases perfectly,
- The fintech market might go through dynamic changes,
- You might want to refactor your application,
- Or pivot your business to a new direction,
- What if your app gets mentioned by a celebrity and your popularity skyrockets?
This illustrates the primary pitfall of reservations and savings plans. Knowing how much capacity you need three months from now is already hard. What about one or three years ahead?
Instead of locking yourself into a deal you can’t escape later, use automated solutions that select the most cost-effective resources and constantly adjust them to your changing needs.
5. Use Spot Instance automation to cut costs
Spot Instances are a great choice for training and retraining the ML models powering your fintech product. You can use them for anything – from security and fraud detection tools to asset management or Business Intelligence solutions that use AI to support decision-making.
This type of virtual machine might get interrupted when the cloud provider reclaims the capacity. If you’re prepared for that and have a plan for managing interruptions gracefully, you can use Spot Instances across more workloads – even the customer-facing things in production.
Automation is essential for taking advantage of the discounts Spot Instances offers. Just to remind you, we’re talking about up to 90% off the on-demand rates!
A cloud automation platform picks the best Spot Instance candidates for your application, taking into account their interruption rate and cost-performance balance.
If that instance gets interrupted while your workload is running, the automation mechanism replaces it with another one to keep it going. If no alternative Spot Instances are available due to high demand, the Spot Fallback feature will temporarily move your workload to an on-demand instance.
Give cloud automation a try
One of our financial services customers – Wio Bank – can now run 90% of its non-production environments on Spot VMs, unlocking massive cost savings.
“By not having to spend time managing Kubernetes clusters, we can focus on other priorities and bring more capabilities into our environment, which is our main focus right now. The time saved thanks to Cast AI is being reinvested into other areas of our business.”
Yannick Janssen,
Deputy Chief Technology Officer at Wio Bank
