In GenAI, Smaller Is Better – and Smarter

If you keep up with GenAI news – and admittedly, it is hard to avoid – you could get the misimpression that enterprise AI is all about LLMs (large language models) and which multi-hundred-billion-dollar entity is about to unleash AGI (artificial general intelligence) for good or ill or, most likely, both.

Yes, the AI world has spent the past two years obsessed with size: ever-larger language models, record-breaking token counts and increasingly abstract benchmarks that have little relation to most businesses’ reality.

But in enterprise settings – especially in regulated, high-risk industries like financial services and healthcare – that arms race has very little to do with actual value creation.

So what does matter, you ask? Two words: business fit.

Here’s what we’ve learned: The organizations getting the most from AI aren’t chasing scale. They’re building systems that are narrow, focused and deeply aligned to what their business does best.

For most companies, the foundational model itself isn’t the main event. It’s just one component of a well-designed system.

In fact, blindly adopting a massive general-purpose model often slows teams down. In areas that require high domain expertise or precision, these models can create noise, introduce risk or simply produce the wrong results. The effort required to filter and fix their output often outweighs the benefit.

So let’s talk about SLMs – small (or specialized) language models. A prime example of SLMs are the Palmyra models from Writer, the AI platform company. Unlike massive general-purpose models, Palmyra models are designed specifically for enterprise use cases in marketing, HR, legal and customer support functions.

(Microsoft outlines the advantages of SLMs.)

SLMs, by definition, are smaller, and thus more efficient and easier for your business to govern – enabling secure deployment within your company’s environment. The Palmyra models embody the core advantages of SLMs: relevance over scale, control over complexity.

Even better, SLMs hallucinate less – critical in industries such as financial services and healthcare, where “close enough” usually isn’t close enough. They perform more consistently and can be targeted to your business needs with greater precision. When you reduce the “mental clutter” of general-purpose models – the stuff they’ve learned from Reddit or Shakespeare or Wikipedia – you get outputs that are clearer, safer and more useful.

SLMs also help solve another real-world problem: cost. Running massive LLMs at scale isn’t just unnecessary for most enterprise use cases – it’s expensive. If you’re spending six figures a month on tokens, you already know smaller, more precise models are where you would much rather be.

We’re past the novelty phase of GenAI. The question isn’t whether your company will use it – it’s how you’ll do so effectively and intelligently.

For most organizations, the path to value isn’t paved with bigger models, bolder claims or benchmark one-upmanship. It’s about selecting a small model with the perfect domain focus, aligning it to what your business does best and deploying AI-powered applications where they can make a measurable impact.

SLMs aren’t a compromise – they’re a smart strategic choice, a way to bring AI closer to your operations, your compliance needs and your customers. So stop tracking parameter counts. Start looking for fit.

The winners in enterprise AI won’t be the ones who adopted the biggest model – they’ll be the ones who picked the rightmodel.