Archive · 3 Aug 2026
AI Scaling Challenges: Portability Issues Across Industries and Healthcare
AI tools can work well in one place but often fail in another.
Tata Steel studied AI for eight years and found that models built for one machine don’t work on another.
Apollo made a heart‑risk tool that worked well in its own country, but when it was used in Latin America it had to be changed.
The reason is that the chance of heart problems was higher in Latin America.
This shows that AI needs to be adjusted for each new setting.
AI models often fail to transfer between different equipment or settings.
Tata Steel’s eight-year study found models do not travel across assets.
Apollo’s cardiovascular risk tool was >90% accurate domestically but needed recalibration abroad.
Risk prevalence in Latin America was 16% versus 8.07% domestically.
The issue highlights the need for localized AI deployment and recalibration.
- Who
- Bose, a representative at Tata Steel
- What
- AI models do not transfer across equipment and require recalibration
- When
- Over eight years of deployment at Tata Steel; recent deployment abroad
- Where
- Tata Steel facilities and Apollo's domestic and Latin American settings
- Why
- Different operating conditions and varying risk prevalence make models non-portable
- Industry
- Manufacturing and Healthcare
- Key Finding
- AI models lack portability across assets
- Risk Prevalence Domestic
- 8.07%
- Risk Prevalence Abroad
- 16%
- Accuracy Domestic
- >90%