Introduction
Many organisations still rely on employees to manually gather, clean, and report performance metrics from a variety of digital platforms. Serve The Home (STH) recently piloted a solution that swaps this repetitive work for a locally‑hosted AI pipeline powered by Nvidia GB10 hardware.
Why Manual Reporting Was a Bottleneck
Requests arrived as long, unstructured emails asking for data across multiple sources and date ranges. The process required a dedicated reporting role, consumed hundreds of hours each week, and was prone to human error.
AI‑Powered Automation Pipeline
STH built a sequential workflow using n8n pre‑built integrations to connect directly to analytics systems. The pipeline:
- Collects raw data from all relevant platforms.
- Applies consistent time limits, filters, and query parameters.
- Runs multiple AI model calls per request to interpret and format results.
- Delivers final reports without human intervention.
The sequential design simplified testing and troubleshooting during the pilot phase.
Model Comparison and Accuracy
Approximately 1,000 historical requests (2015‑2025) with known outcomes were used to benchmark two open‑source models:
- gpt‑oss‑20b FP8 – performed well on simple queries but showed errors as request complexity grew.
- gpt‑oss‑120b FP8 – achieved >99.9% per‑step accuracy, reducing workflow errors from weekly to rare annual events.
Higher accuracy proved critical because each request required several model calls, and small mistakes compounded quickly.
Infrastructure & Cost Benefits
Two Dell Pro Max servers equipped with Nvidia GB10 units ran the AI locally, keeping all data on‑premises. The hardware investment paid for itself within twelve months by eliminating the need for a full‑time reporting specialist.
Impact on Roles and Business Operations
Automation freed staff to focus on higher‑value activities, such as hiring a managing editor, while maintaining consistent reporting quality. Roles centered on data gathering and summarisation become vulnerable once reliable AI pipelines are in place.
Key Takeaways
- Local AI can replace manual metric collection, delivering speed and accuracy.
- Choosing the right model size dramatically affects error rates in multi‑step workflows.
- On‑premises hardware (e.g., Nvidia GB10) safeguards sensitive data and offers predictable ROI.
- Automation reshapes workforce needs, shifting focus from repetitive tasks to strategic initiatives.
Conclusion
Serve The Home’s experiment demonstrates that with the right infrastructure, model selection, and workflow design, organizations can automate reporting at scale, cut costs, and reallocate talent to more impactful functions.