Building AI-powered debt collection solutions
FutureSoft, one of Africa’s largest debt collection software providers, is expanding its AI capabilities through Merlin AI, a platform designed to enhance transcription, summarisation, compliance monitoring and customer communications in contact centre environments.
Built on FutureSoft’s Excalibur ecosystem, Merlin AI helps transform complex audio interactions into actionable insights that improve compliance, customer service and operational decision-making. Within FutureSoft, development teams design and maintain platforms that integrate .NET APIs, Python AI services and GPU-enabled workloads to ensure systems remain performant, secure and scalable.
As AI adoption accelerated, FutureSoft required infrastructure capable of supporting AI processing workloads while keeping sensitive customer data hosted locally in South Africa.
Why GPU infrastructure was essential
FutureSoft’s AI workloads, including speech-to-text transcription, summarisation, analytics and compliance monitoring, could not run efficiently on CPU-only infrastructure. Early testing showed that larger AI models and growing recording volumes created bottlenecks that slowed processing and limited scalability.
GPU acceleration became essential to:
- Reduce processing times for AI services
- Support larger AI and language models
- Improve transcription and summarisation accuracy
- Scale workloads predictably as customer demand increased
- Enable multilingual and compliance-driven AI use cases
“Xneelo’s guidance made deployment easy and gave us the confidence to push our GPU workloads further.”
How it works:
FutureSoft implemented xneelo Self-Managed GPU Servers as the foundation for its AI platform, combining GPU acceleration with containerised application services.
Their environment includes:
- CUDA tuning and optimised Docker images for GPU workloads
- Container orchestration across .NET Aspire and Python services
- Secure cross-service authentication via IdentityServer
- Integration between .NET APIs and Python AI services for transcription, summarisation and analytics workloads
Rather than relying on a number of oversized GPU systems, FutureSoft adopted a distributed architecture using multiple smaller GPU servers. This approach allows workloads to be distributed efficiently while enabling predictable scaling and infrastructure growth over time.
The platform now supports:
- Concurrent AI batch processing workloads
- Larger production-grade AI models
- Advanced Whisper transcription models for improved accuracy
Business impact
FutureSoft’s GPU infrastructure significantly improved the performance and scalability of its AI services, enabling faster insights and more reliable customer-facing operations.
Key outcomes include:
- Faster delivery of AI-generated insights
- 20%+ improvement in transcription and summarisation accuracy
- Reliable handling of 50+ concurrent batch jobs
- Predictable infrastructure performance and costs
- Scalable growth as adoption of Merlin AI increases
Why local hosting matters
For FutureSoft, hosting AI infrastructure locally in South Africa was a critical requirement. By running AI workloads within xneelo’s data centre environment alongside their existing Future Cloud infrastructure, FutureSoft benefits from:
- Low-latency connectivity between AI and production systems
- Data sovereignty and local data residency
- Enhanced security and compliance control
- Rand-based pricing with predictable operating costs
- Reduced dependence on third-party cloud platforms
- Compliance with cross-border data regulations
- Peace of mind knowing sensitive customer data is securely hosted on South African soil
This local infrastructure model enables FutureSoft to deliver AI capabilities while maintaining control over sensitive financial and customer data.
Full control and support when you need it
“Working with xneelo support was smooth and reassuring. Their guidance made deployment easy and gave us the confidence to push our GPU workloads further,” says Henko Meyer, Software Developer at FutureSoft.
Having full control over their Self-Managed GPU Servers allows the FutureSoft team to fine-tune drivers, AI models and orchestration to match the exact needs of their workloads.
Advice for other businesses
For organisations running sustained AI workloads, Self-Managed GPU Servers provide the control, predictable performance and scalability needed to support production AI systems cost-effectively.
For FutureSoft, locally hosted GPU infrastructure has become a strategic enabler, allowing the company to scale AI capabilities, maintain control over sensitive customer data and deliver AI services with consistent performance and reliability.
