Key Takeaways
- 1Eliminated $42,000 in dead stock and cut fast-moving SKU stockouts by 28% without buying enterprise software.
- 2Ran IBM Granite Time Series locally on a CPU, avoiding cloud computing costs entirely.
- 3Integrated raw CSV exports from Siigo ERP to generate automated 90-day reorder alerts.
Action Callout: Test your own data this week
Export a two-year CSV of daily sales for your top 50 fastest-moving SKUs from your ERP. Have a junior developer run it through the open-source Granite Time Series zero-shot model using a basic Python script. Compare the model's 90-day prediction against what your Excel sheet says.
Frequently Asked Questions
Can you run IBM Granite Time Series locally?
Yes. The PatchTST-FM-r2 model is roughly 385M parameters and requires no GPU for inference, making it fully operational on a standard laptop CPU.
Do I need a data science team for zero-shot forecasting?
No. Zero-shot models output predictions without needing extensive retraining, meaning you can drop in your historical data exports directly.
Case StorySupply ChainAIOpen SourceLatAm
Share this article
Kyto
AI & Automation Firm
We design and build AI automations and business operating systems. Agency results + Academy sovereignty.
