Enterprises are ditching 'general-purpose' LLMs for smaller, targeted AI models in production. The shift signals a new priority: reliability and tighter governance now outweigh raw capability for enterprise AI in production. This directly impacts how product managers architect AI pipelines and investors evaluate AI startups.
How We Got Here
For the last 18 months, the default enterprise AI strategy centred on large, general-purpose LLMs. But as companies moved beyond experimentation, the fundamental question pivoted from raw capability to reliable control and governance.
The Numbers
- SLMs are purpose-built for discrete functions like document processing, classification, and workflow automation.
- Anuj Gupta, founder of KiteFishAI, warns that model size alone does not guarantee safety or tighter control.
- Sourav Bandyopadhyay, chief scientist at Shunya Labs, notes task-specific boundaries make rigorous evaluation possible for SLMs, unlike general LLMs.
- Conventional academic benchmarks often fail to evaluate critical enterprise functions like banking workflows or compliance adherence, which SLMs address.
What Happens Next
🇮🇳 Why This Matters for India
Indian SaaS founders building for specific verticals like healthcare or fintech will find SLMs easier to sell into strict compliance environments.
The Take
The winners here are Indian vertical SaaS companies and AI solution providers who build expertise in narrow domains for enterprise clients. Expect incumbent LLM providers to quickly acquire specialized fine-tuning startups or lose significant ground in regulated enterprise deployment within the next 12-18 months.
Source:
Analytics India Mag ↗