Company
Moonshot Marketing is a Tel Aviv-based performance marketing company that operates an affiliate network and online comparison platform, connecting advertisers and publishers across various industries, including financial services, lifestyle, iGaming, and technology. By leveraging AI-driven campaign automation, real-time bidding, and predictive analytics, the company matches high-value users with the right brands to scale profitable customer acquisition.
Challenge
Moonshot Marketing products rely on real-time data, automation workflows, and heavy traffic across comparison platforms, requiring the cloud infrastructure to be both agile and cost-efficient. The company faced high operational costs and limited visibility into how its Kubernetes clusters scaled.
Manual autoscaling required constant tuning, and workloads often ran on overprovisioned resources, resulting in wasted capacity. This prompted the search for an intelligent, fully automated solution that would rightsize workloads and clusters while maintaining peak performance and reliability.
Solution
Cast AI provided Moonshot Marketing with a fully automated solution that continuously rightsizes and rebalances cloud capacity in use, intelligently manages the use of Spot and On-Demand compute instances, and provides clear cost controls. Cast enables the company to reduce both its cloud spend and operational overhead – without compromising uptime or performance.
Results
- 40% of cloud costs savings achieved continuously, every month
- Reduction of manual autoscaler tuning to near-zero
Workload Autoscaler
Following the adoption of Workload Autoscaler and the enablement of scheduled rebalancer across all staging, integration, and production clusters, Moonshot Marketing reduced provisioned CPU by 70% and provisioned memory by 65%.
Cast automatically sets workload requests and limits to dramatically improve resource utilization and generate cost savings.
In this cluster, Cast reduced the number of CPUs by 77% and memory GBs by 76%, resulting in an overall 77% decrease in the cluster’s running cost.
Cluster rebalancing
Rebalancing is designed to keep a cluster in its most efficient and up-to-date state. It works by automatically replacing underperforming nodes with new ones that are more cost-effective and run the latest configuration settings.
In this example, Cast reduced memory overprovisioning by 79% from 205 to 42.5 GB:
What was especially impressive was that the savings Cast AI generated weren’t just a one-time drop. After two or three months, the cost reductions remained consistent. It wasn’t a temporary improvement – the optimizations held over time.
That continuous, sustained reduction proved that Cast AI wasn’t just a short-term fix but a lasting solution for keeping our infrastructure efficient and cost-effective.
Adi Tal, VP R&D at Moonshot Marketing
Extending FinOps with automation
What is your approach to managing the costs that your Kubernetes clusters generate?
FinOps is really part of our everyday work; it’s something we focus on constantly. We’re always looking for new ways to automate and strengthen that area, using systems that can actively support our cost management goals. We also work closely with one of our AWS partners, who provides a FinOps platform that monitors our data multiple times a day and delivers detailed daily cost reports.
I always encourage the team to stay on top of our FinOps status – to understand where we stand, what’s driving changes, and how we can optimize further. We’re continuously adding new tools and features to make sure our costs stay predictable and under control, without unexpected spikes or inefficiencies.
When I first got access to Cast AI, I saw a system that gave me real visibility into our cloud costs, it was a game changer. Since then, it’s become part of my daily routine.
Every morning, I check the graphs and review the cost data – it’s almost like a hobby now. Having that level of insight helps me understand exactly what’s happening in our environment and keeps me focused on making sure everything stays optimized and efficient.
Adi Tal, VP R&D at Moonshot Marketing
What is your approach to automating cloud cost management?
Back then, we were already managing many advertising offers of the top websites across several industries. To deliver on our commitments to customers, everything needed to be automated – for example, collecting data and processing it quickly to ensure users always saw real, up-to-date information.
When I first joined Moonshot, a lot of tasks were still manual. We started automating wherever we could, and my team eventually implemented Cast AI to let developers focus on application development instead of worrying about how the application scaled up or down.
That’s where Cast really made a difference. After implementing it, we stopped having issues with scaling altogether. Our applications and microservices became highly scalable, and that’s been one of the biggest benefits we’ve gained – we’re very happy with the results.
Peleg Bublil, DevOps Engineer at Moonshot Marketing
A phased onboarding process
What was onboarding your clusters to Cast like?
We discovered Cast during our search for a Kubernetes cost optimization solution. Onboarding was fast and straightforward: we connected AWS/EKS, installed the agent, and applied our scaling and cost policies. We first activated Cast in “recommendations” mode to validate everything against our guardrails before enabling full automation. Once connected, Cast immediately began optimizing our clusters by rightsizing and rebalancing workloads to deliver instant value.
The impact was immediate: by the end of the first month, our AWS costs were down by roughly 40%.
In the pilot phase, we enabled the autoscaler and optimizer in non-production, canaried a few workloads, and eventually replaced the native Cluster Autoscaler. For observability, we added alerts and dashboards, tagging costs by team and service to verify savings. Finally, we rolled out to production gradually, enabling automation by namespace and workload, and scheduling rebalances during maintenance windows.
A few integration details made the transition seamless: Cast is EKS-native, respecting taints, tolerations, priority classes, and PDBs with safe drain and cordon flows. It’s also IaC-friendly – our policies and configurations are versioned in Terraform and Helm with per-environment overrides. And thanks to staged cutovers, we minimized risk for stateful and latency-sensitive services while maintaining full control.
Having full visibility into all our clusters from a single SaaS platform – with clear dashboards showing cost trends and optimization impact – has made it much easier to manage and monitor our environment.
Peleg Bublil, DevOps Engineer at Moonshot Marketing
The value of observability
What role did Cast’s reports and dashboards play in the integration process?
When we first started using Cast, we immediately noticed how powerful the dashboard was – it provided a ton of actionable recommendations right from the start. That made things much easier for us.
Most FinOps tools can monitor costs and give general recommendations based on your AWS account, but as DevOps engineers, we know that an EKS cluster is like its own standalone system – it’s complex and not easy to analyze or optimize effectively.
With Cast, we finally got clear, data-driven insights specific to our Kubernetes environment. It identified over-provisioned workloads and showed exactly where we were allocating too much CPU or memory.
Peleg Bublil, DevOps Engineer at Moonshot Marketing
Once we had all the information in one place, it really transformed the way our DevOps engineers work. Everything became more connected and transparent. We could easily see spikes in usage directly on the Cast graphs and understand how those spikes correlated with increases in cost.
Before, investigating cost anomalies meant jumping between multiple tools and dashboards. Now, when someone notices higher spend in our FinOps platform, they can immediately go into Cast, see the related alerts, and pinpoint what happened. The visibility and context Cast provides make it much faster and easier to understand and resolve issues in real time.
Adi Tal, VP R&D at Moonshot Marketing
Saving 40% on the cloud every month
What level of cost savings have you achieved?
Overall, we achieved roughly 40% cost savings, with a full payback in the first month.
The savings have proven durable thanks to several key factors:
- continuous rightsizing and autoscaling that eliminated overprovisioning;
- policy-driven use of spot and on-demand instances to balance savings with reliability;
- smarter bin-packing and periodic rebalancing that minimized idle capacity;
- and off-hours scale-downs in non-production environments that made the savings repeatable.
We continuously validate these results through cluster and namespace cost tagging, weekly pre- and post-change comparisons, and ongoing SLO monitoring – all confirming the sustained ~40% cost reduction and stable, high-performing infrastructure.
Peleg Bublil, DevOps Engineer at Moonshot Marketing
Cast provided great optimization opportunities, and once we enabled features like rebalancing and adjusted the rightsizing of our Kubernetes nodes, the impact was immediate across all our accounts.
What was especially impressive was that the savings Cast AI generated weren’t just a one-time drop. After two or three months, the cost reductions remained consistent. It wasn’t a temporary improvement – the optimizations held over time.
Adi Tal, VP R&D at Moonshot Marketing
How soon have you seen positive ROI from implementing Cast?
Optimization began the moment automation was enabled – within the first month, we reduced our AWS spend by about 40% while also cutting day-to-day operational toil. Since then, we’ve consistently maintained a lower cost baseline without any impact on performance or SLOs.
Workload rightsizing and node autoscaling made a difference
Which features were game-changers for you?
Cast’s key automation features have become central to how we manage and optimize our Kubernetes environments:
Workload Autoscaler provides policy-driven scaling for each workload type, reacting instantly to changes in demand while honoring Pod Disruption Budgets (PDBs) and priority classes.
Evictor continuously drains underutilized nodes, keeping clusters right-sized without the need for manual cleanup.
Workload Rebalancing ensures ongoing bin-packing across instance types and availability zones, reducing fragmentation and improving overall utilization.
These capabilities have been game-changing for us. We’ve nearly eliminated manual autoscaler tuning, allowing engineers to focus on infrastructure and CI/CD instead of capacity management.
Smarter placement and rightsizing have locked in roughly 40% cost savings with no impact on SLOs, and our clusters now stay consistently optimized – clean, efficient, and free of drift over time.
Did Cast impact your engineering workload and developer happiness?
Cast has eliminated most of our day-to-day capacity management toil, allowing engineers to focus on higher-value work like automation, CI/CD, and reliability improvements.
Here’s what changed for the team:
- No more autoscaler babysitting – Policy-driven scaling replaced the constant manual tuning of HPAs, VPAs, and node groups.
- Hands-off housekeeping – Evictor keeps clusters right-sized automatically; no more manual draining, cordoning, or cleanup of idle nodes.
- Fewer firefights – Smarter bin-packing and continuous rebalancing reduced paging incidents during traffic spikes and deployments.
- Simpler operations – Clear guardrails and IaC–managed policies replaced ad hoc scripts and one-off fixes, making operations cleaner and more predictable.
Overall, Cast has streamlined our workflows, reduced cognitive load, and freed the team to focus on innovation instead of cluster babysitting.
We definitely saw a boost in team happiness after implementing Cast, largely because of how easy the system is to use.
Before automation, we had to monitor everything very closely and perform a lot of manual work – rebalancing, resizing, and managing clusters every single day. It easily took around two hours of maintenance daily just to keep things running smoothly. Now, all of that happens automatically.
The team no longer needs to spend time on repetitive operational tasks, which means we can focus on improving our infrastructure and building tools that support our engineering teams. It’s been a big shift – not just in efficiency and cost savings, but in morale and how much more productive the team feels.
Adi Tal, VP R&D at Moonshot Marketing
Where have you reinvested the time your team reclaimed thanks to Cast’s automation?
We’re reinvesting both the time and savings from Cast into advancing our platform’s maturity – expanding Spot automation, tightening scaling policies, and improving reliability and developer velocity.
After automating much of our capacity management, we’ve been able to focus on smarter scaling strategies: rolling out Spot instances across non-production first, then selectively in production for stateless or interruptible workloads with defined disruption budgets.
We continue to fine-tune workload-level policies like PDBs, priorities, and rebalancing windows to keep automation safe, predictable, and aligned with performance goals.
At the same time, we’re investing in reliability through capacity modeling, load testing, and chaos drills to harden resilience while maintaining low costs. On the developer side, we’ve used the freed-up time to enhance CI/CD pipelines and self-serve infrastructure, enabling faster iterations and fewer operational bottlenecks.
What are your future optimization plans with Cast?
Beyond the current rollout, our roadmap centers on safely expanding Spot usage, deepening rightsizing efforts, and refining placement and governance strategies to keep savings durable without ever compromising SLOs.
We plan to extend Spot capacity to selected production workloads with clearly defined disruption budgets, PDB- and priority-aware scaling, and automatic fallback to on-demand instances during volatility windows. We’re also scaling pod-level right-sizing by leveraging Cast recommendations to enforce percentile-based CPU and memory requests and limits, automatically generate pull requests, and gate merges with policy checks to prevent drift.
On the infrastructure side, we’re broadening our instance mix to include Graviton/ARM where compatible, consolidating node groups, and allowing Cast to continuously rebalance for the most cost-efficient, high-performance blend.
Finally, we’re implementing smarter placement rules that use taints, tolerations, and priority classes to co-locate interruptible or stateless workloads, isolate stateful or latency-sensitive ones, and further improve bin-packing efficiency across clusters.
