Company
InCred is a new-age financial services company that uses technology and data science to make financing quick and simple. The company’s portfolio is diverse, encompassing personal loans, student loans, school finance, and loans against property, with co-lending arrangements. InCred also boasts a large workforce, with over 2,700 employees and a network of over 140 branches throughout India.
Challenge
Reliability and scalability have always been critical at InCred, a cloud-native financial services organization. As the company scaled, the complexity and cost of the infrastructure increased. The company used a mix of open-source solutions to optimize costs, but both came with operational trade-offs, particularly around upgrades, support, and the manual effort required to maintain efficiency.
This prompted InCred to search for a solution that would unlock deeper savings and streamline operations. The seamless onboarding, hands-off maintenance model, and powerful out-of-the-box workload autoscaling feature of Cast AI were exactly what the company was missing.
Solution
Cast helps InCred scale smarter, cut costs, and reduce operational overhead—without sacrificing reliability. Two standout features are rebalancing, which offers a holistic view of savings and automates node optimization, and the Workload Autoscaler, which intelligently adjusts Kubernetes resources with no manual effort. Cast is a key enabler of hands-off, intelligent optimization across the infrastructure.
Results
- 30% cloud cost savings on compute
- Dramatic increase in productivity: a task that took seven days now gets done in three to four days
- Extra cost savings unlocked from Spot Instances thanks to automated instance selection
Massive increase in resource utilization
By integrating Cast’s automation features into an already well-optimized cloud environment, InCred Finance dramatically reduced its rate of overprovisioning.
The example below shows a cluster running at only 2.3% overprovisioning:
Workload Autoscaler
The ability to automatically set workload requests and limits was a game-changer for InCred Finance.
The graph below illustrates a dramatic reduction in the volume of workload requests in one of the clusters:
We track the aggregate savings we’re achieving. As I mentioned earlier, we were already fully using Spot Instances, so when we started with Cast, we didn’t expect huge gains.
We’re currently seeing about 30% in monthly savings on compute, which is a significant number for us. We’re super happy with that. Honestly, this wasn’t even on our radar. We had always assumed we were already pretty well optimized. So this felt like a real win for us.
Dheeraj Arani, Head – Cloud Infra, DevOps & SRE at InCred Group
Managing infrastructure costs with reliability in mind
How do you tackle cloud cost management while maintaining reliability at scale in such a high-pressure industry like financial services?
We at InCred—being a financial services organization—are born out of the cloud. All of our services run in the cloud, so their reliability is absolutely paramount for us.
With that in mind, and considering our recent growth, we’ve fully embraced Kubernetes. Scalability and reliability have been key priorities for us since day one. We started with Kubernetes and have been running it for about four years now.
As we scaled, we encountered some cost challenges, and we had to address those growing concerns. That’s when we came across Cast, which has been tremendously helpful in that journey. So yeah, we’re super happy with the results we’re seeing now.
Let’s start at the beginning: what were the first steps you took toward optimizing cloud costs?
We have been using Spot by NetApp’s for three years. When we were considering the switch to Cast, we actually ran a POC with Amazon’s open-source tool, Karpenter. We spent quite a bit of time evaluating it too. At that point, we were managing around seven Kubernetes clusters, running a mix of Spot by NetApp and Karpenter.
Having used both, Karpenter definitely came with its set of challenges—particularly around upgrades and ongoing maintenance. It required a team with the right skills to manage it continuously, which added to the operational overhead.
We were looking to unlock new levels of cost savings and Cast’s out-of-the-box workload autoscaling feature really stood out—it was something we felt was missing in our previous tooling. The switch to Cast was pretty seamless, and that’s a big reason why we decided to move forward with it.
Could you describe the Cast onboarding process in more detail?
When we first started having conversations with Cast, we were really skeptical at the time—I mean, we run seven clusters, everything is optimized, we know what we’re doing. We didn’t think Cast could really help change or improve much of what we were already doing.
Honestly, we just went through the motions—got on the call, saw what Cast does—and really liked the UI from the start. It looked clean, and it was easy to work with.
Starting with the Cast AI Agent
The onboarding began with Cast’s agent, which passively monitors the clusters in a completely nonintrusive way. It gathers data and then generates a report showing the potential savings. It doesn’t take any actions on its own—just observes.
So I thought, okay, fine, let’s give it a try. Since it’s hands-off, there’s no harm in seeing what it finds. Worst case, I could just use the data to show my team how well-optimized we already were.
But once we onboarded the agent in monitoring mode, it showed around $3-4k in potential savings—and that genuinely surprised me. We were confident our clusters were already running at peak efficiency, so seeing that was eye-opening.
Support throughout the integration
From there, the Cast team really supported us—they walked us through everything. The actual steps we had to take were super simple. In the end, we just had to install a couple of Helm charts. It was completely seamless—I think onboarding took us maybe 10 minutes.
Since then, we haven’t really had to touch the configuration at all. It’s been about two years now, and it’s still as seamless as it was on day one.
Gradual rollout
It took us around three months to roll Cast out across all our clusters, and we ramped up our non-production usage gradually each month. The Cast team stayed in regular contact — I think we were syncing every couple of weeks to discuss the next steps and where we could further optimize.
So, we started seeing real ROI after about three months. All in all, I’m really happy with the experience and how quickly and smoothly we were able to roll it out across all of our clusters.
How did you proceed with safely rolling out Cast across your production environment?
We initially had this concern—are we offloading too much control to a tool we’re not completely sure of? That’s why we decided to run it in our non-production clusters for a month first.
But honestly, there were absolutely no complaints from the team. We did run into a few minor hiccups, but the Cast team addressed them immediately, which gave us the confidence to move forward.
After that, we extended the same controls from non-production to production. Now, all of our clusters run fully on Cast. We’ve been quite aggressive in terms of what we enable—we’ve turned on all the features Cast offers. So far, things have been going really well.
Unlocking 30% cost savings
Which Cast features turned out to be game-changers for your team?
Rebalancing
Our first game changer is the rebalancing feature—it shows us the current state, the desired end state, and the delta savings we can achieve. We actually use it quite frequently—we run it every couple of days.
That’s something new that Cast brings to the table. All the other tools we used before handled it on demand only. They didn’t do a comprehensive aggregation of all nodes to perform a true rebalance—they just looked for isolated opportunities for node reshuffling. Cast takes a much more holistic approach, which makes a big difference.
Rebalancing is definitely one of the standout features for me.
Workload Autoscaler
The second feature we can’t live without now is the Workload Autoscaler. Cast was the first tool we saw that handled this automatically. Sure, there are other tools, including open-source ones, that provide recommendations but those still require manual review and implementation.
What Cast does is track workloads continuously, with a model that learns over time, and then applies those scaling decisions automatically when we’re ready. We’ve rolled this out broadly and have been using it everywhere.
Network monitoring with egressd
Another feature I’ve started using recently is cost monitoring with the egressd daemon. It shows the data transfer split between services and that was eye-opening. I got so many insights from this.
This capability gave us visibility into how much data was being transferred between two of our services. That insight alone led us to optimize our application in ways that weren’t even Kubernetes-related. So yeah, that’s a feature I’ve really started to rely on lately.
Automated K8s upgrades and node updates with full visibility
There’s also that sense of comfort we now have—knowing that if we get blocked with something Kubernetes-related, especially around upgrades, Cast makes things much easier. Kubernetes upgrades are generally hard. In the past, when we had to do an upgrade, we had to think about creating node groups, node templates, the latest AMIs available in the market — all of that.
With Cast, that’s no longer a concern.
There’s also something related to node objects that I hadn’t mentioned earlier: periodic node updates happen automatically as part of the rebalance cycles. That’s now something I rely on heavily. Being in the financial sector, patching and updating our servers is a top priority.
Cast has made that process significantly easier. We don’t have to worry about how I’m going to handle patching anymore, it’s all handled automatically in the background.
And what’s really helpful is that it’s all demonstrable. We can go into the Cast console and, at any point, show an external auditor exactly what’s been set up and what actions have been taken. There’s a clear audit trail, and that gives me a lot more peace of mind.
This was a real pain point for us. Even when we used Karpenter—which does offer a similar feature—the experience wasn’t nearly as smooth. There was a lot of manual work involved. We had to double and triple-check everything to make sure we were doing it right because even a small mistake could have serious consequences. That’s something we simply don’t have to worry about anymore.
What level of cost savings did these features help you to achieve?
I track the aggregate savings we’re achieving. As I mentioned earlier, we were already fully using Spot Instances, so when we started with Cast, we didn’t expect huge gains.
We’re currently seeing about 30% in monthly savings on compute, which is a significant number for us. I’m super happy with that. Honestly, this wasn’t even on my radar. We had always assumed we were already pretty well optimized. So this felt like a real win for us.
Did Cast help you to unlock further cost savings from Spot Instances themselves?
Oh yeah, I think one of the key advantages Cast brings is its intelligent approach to selecting spot instances — it’s not just about grabbing the spot instance size you think you need.
Typically, any engineer managing infrastructure might say, “Okay, we need 2 CPUs and 4GB of RAM,” and then go with the available spot instance that matches that spec. The assumption is it should cost the same as any other machine with those specs. But what Cast does is far more intelligent.
It evaluates the broader picture. Sometimes, going with a larger instance type turns out to be significantly cheaper than running multiple smaller ones—and that was a completely new insight for me when we first onboarded with Cast.
So this feature has been tremendously helpful—not just in the context of Kubernetes, but for cloud cost management in general.
Did Cast have an impact on the workload and productivity of your team?
Yeah, this is definitely something worth highlighting. I have a team of 10 engineers, and earlier, we were managing everything manually: shell scripts, digging through Karpenter documentation, and deciding when and how to scale using HPA and VPA. It was all very hands-on.
Now, with Cast, both Horizontal and Vertical Pod Autoscaling are handled automatically. That’s been a huge productivity boost for us. We no longer worry about setting min/max thresholds for each workload—Cast takes care of it. It’s been incredibly helpful.
We run close to 100 microservices today, and as we grow, that number keeps increasing. Thanks to Cast, even our developers can move faster—they don’t have to think about resource provisioning or scaling as they build and deploy. Across the SDLC, it’s just smoother now.
We’ve always had a lean team, and the backlog was always infinite. But with the time savings Cast gives us, things that might’ve taken seven days now get done in three or four. So yeah, Cast has definitely helped us deliver faster — and while we can’t put an exact number on the productivity gains, I’m very happy with what we’re seeing.
To what kind of company would you recommend Cast?
I talk to a lot of friends in the industry, and most of them seem to be using Cast already. So I wouldn’t say this applies to just one specific industry—really, anyone who’s fully leveraging Kubernetes today should give Cast a shot. Chances are, they’ll be surprised to find there’s still room for improvement, even if they think they’re already optimized. So yeah, if you’re running a cloud-native tech stack built on Kubernetes, Cast is definitely worth trying.
