Fintech Loan Servicer Cuts Model Deployment Costs by 90% with Bento

“The Bento Inference Platform has completely changed how we operate. We no longer waste days firefighting deployments — our team can focus on building models and moving the business forward with confidence.”

—— Director of Data Science, Fintech Loan Servicer

About#

This fast-growing fintech loan servicer operates in the consumer lending space, where speed and accuracy in decision-making directly impact revenue. The team manages dozens of models in production to power automated underwriting and lead acquisition, processing thousands of applications daily.

Their data science team relies on Python-based frameworks and cloud infrastructure to develop and deploy tree-based models for tasks like credit scoring, risk assessment, and lead valuation. These models are critical for scaling their lending operations efficiently.

Challenge: Scaling model deployment on legacy infrastructure became unsustainable#

Operating in a highly regulated, security-sensitive environment, the fintech loan servicer needed to ensure strict compliance while still maintaining agility. The loan servicer originally ran its models on Flask and EC2 instances, but as traffic grew, that setup quickly hit its limits.

Seeking more scalability, the team migrated to a new stack that introduced fresh challenges. Deployments failed without clear logs, and versioning issues caused instability. Without visibility into production issues, engineers spent days untangling errors. Even small model updates could take days to push to production, delaying the company’s ability to ship new products or respond to business needs.

These technical setbacks had a direct financial impact. At one point, the team was forced to overprovision massive instances just to get models live, a process that drained budgets and slowed the pace of innovation.

Beyond costs and instability, demonstrating compliance at scale also required significant manual oversight from engineers and security teams. These manual compliance tasks strained team resources and pulled focus away from business priorities.

The team explored multiple paths forward, from managed ML platforms to expanding in-house tools. But each alternative either failed to satisfy regulatory demands, provided insufficient monitoring and reliability, or proved too costly at scale. That’s when the loan servicer’s Director of Data Science found the Bento Inference Platform.

“Sometimes deployments just wouldn’t work. We had to slim down, change versions, even roll back between releases. We’d spend days debugging as errors sucked up time from every data scientist and pulled people into areas outside their expertise.”

Solution: Bento Inference Platform restored control, compliance, and efficiency with BYOC#

The Bento Inference Platform’s Bring Your Own Cloud (BYOC) option proved critical for the loan servicer. Deploying securely inside its own AWS environment satisfied strict compliance requirements for handling sensitive credit bureau data while maintaining full control of infrastructure.

The BYOC onboarding process took less than a week. After a quick architecture review with the customer’s security team, the DevOps team ran a single script to create a least-privileged access role. From there, Bento’s system automated the rest. This smooth, tested process has already been verified by IT and infrastructure teams in highly regulated industries, demonstrating that BYOC deployments are both secure and straightforward.

Once the Bento Inference Platform was live, the team paired it with Comet ML for model tracking. This provided unified visibility into every model in production, covering lifecycle management, reproducibility, and performance across a rapidly expanding ML catalog.

The Bento Inference Platform also eliminated the operational inefficiencies of the company’s legacy stack. Deployments that once failed without logs or required lengthy rollbacks now run consistently. Thanks to dependency pinning and built-in monitoring, every model behaves predictably in production, with actionable logs that make it easy to diagnose issues. Deployment cycles that previously took days are now reduced by 20–40%, enabling the loan servicer to ship about 50% more models, including innovative projects that would have been out of reach with the old system.

With Bento’s responsive support, urgent issues are consistently addressed in under 30 minutes — a stark contrast to the black-box experience of previous platforms. This minimizes downtime and ensures the team can innovate without fear of disruption.

The Bento Inference Platform also delivered unexpected value, uncovering a timeout issue that had been silently cutting off traffic. By resolving the issue, the loan servicer recovered about 10% of lost leads over 30 days.

“The biggest thing for us was knowing our models would run the same every time. With Bento, pinning dependencies and having clear logs meant we could finally trust our deployments.”

Results: 90% lower compute costs and 50% more models shipped#

In just two months, the fintech loan servicer turned their biggest operational challenge into a competitive strength. With the Bento Inference Platform, they transformed an unreliable, resource-draining deployment process into a predictable, efficient system that fuels growth and innovation.

• 90% lower compute costs with BYOC deployment: By migrating to the Bento Inference Platform’s BYOC deployment inside their own AWS environment, the loan servicer reduced compute costs dramatically. These savings came without sacrificing performance or reliability, freeing up budget to be reinvested in new model development and business-driving initiatives.
• 75% overall spend reduction, including platform expenses: Total infrastructure and platform spend dropped significantly, thanks to lower compute requirements, simplified maintenance, and predictable contract costs. The team can now allocate more resources toward building and shipping new products.
• 50% more models shipped through streamlined deployment: With dependency pinning, monitoring, and smoother rollouts, deployments that once took days are now predictable and fast. The team has shipped roughly 50% more models in recent months, including several high-value projects that would have been too costly to attempt with the previous stack.

With infrastructure no longer a bottleneck, the team has shifted from maintenance to innovation. The loan servicer is currently rebuilding its decisioning flow, a project that will involve dozens of new models, and exploring advanced inference optimization and scaling features to further reduce costs. With the Bento Inference Platform in place, the team is confident they can scale and tackle more ambitious initiatives without a repeat of past infrastructure headaches.

“It feels good to add new models without worrying about reliability. With the Bento Inference Platform, we can keep scaling without slowing down.”

More resources#

• Need help with scaling enterprise AI? Schedule a call with our experts.
• Sign up for Bento Inference Platform for free. Build and scale your custom AI systems in production, with any model, on any cloud or on-premises environment.
• Join our Slack community to get the latest information about self-hosting AI models and inference optimization.