Built-for-Purpose AI: Positioned to Gain Traction

Technology organizations of all sizes are evaluating how best to adopt artificial intelligence. Many are using general-purpose AI platforms that offer wide functionality out of the box. Others are beginning to explore building and deploying models that are designed precisely for their own data, context, and business goals.

This alternative approach may increasingly serve as a source of differentiation. Startups, mid-market firms, and global enterprises share similar incentives: to maximize return on investment, control costs, and maintain flexibility. The following trends suggest that a built-for-purpose strategy may gradually become more compelling.

Token Costs Are Declining Rapidly

The price of AI tokens has been falling steeply. In early 2023, some models cost around $20 per million tokens. By 2025, prices have dropped to approximately $0.07 per million tokens—a roughly 280-fold decrease in under two years. That shift drastically lowers the entry barrier for AI use across all types of companies (Search Engine Journal).

Another industry analysis found token prices dropping from about $10 per million to $2.50 per million by March 2025—representing a 75 percent reduction in just one year (Ramp).

These declines suggest that AI applications are becoming more affordable at scale, enabling smaller organizations to explore tailored solutions that previously would have been cost-prohibitive.

Small Models Are Getting More Efficient and Effective

In 2024, a model 142 times smaller than a major benchmark model achieved the same level of task accuracy, demonstrating that efficiency can replace sheer size (Search Engine Journal).

Academic research introduced a paradigm in which small models handle simpler subtasks, allowing large models to focus on the harder parts. This approach achieved comparable or even better performance—in one case, improving accuracy from 94.43 percent to 95.64 percent—while reducing model-query cost to about 31 percent of the original (arXiv).

Moreover, independent studies suggest that small-language models can deliver around 94 percent task efficiency compared to large models, making them a viable, cost-efficient alternative (BizzBuzz).

These trends point toward a future where leaner model architectures can deliver strong performance while operating at a much lower cost.

Productivity per Token Counts More Than Volume

Businesses increasingly evaluate AI through a productivity lens—how much value each token generates—not simply how many prompts get processed. Models that are built to address specific workflows often deliver greater productivity per token, because they avoid unnecessary verbosity and deliver more accurate, relevant output.

Organizations that optimize token usage—with techniques like fine-tuning, prompt batching, and prompt caching—have reported significant reductions in cost. One firm optimized its prompts and caching techniques to cut costs by 40 percent or more (Reddit). Fine-tuned models can reduce token usage by 50 to 75 percent for repeatable tasks, yielding long-term savings (10Clouds).

Every Organization Is a Potential Training Ground

Whether an emerging tech startup or a large enterprise, every company has unique workflows, customer interactions, and operational signals. Embedding AI within these environments creates opportunities for models to evolve based on real feedback.

This feedback loop may gradually outperform general-purpose platforms whose development follows external vendor timelines. Over time, organizations may extract more value from models that are trained and refined within their own operational context.

Comparative Perspective

The Forte Perspective

We believe that technology should be designed to serve the context in which it operates. AI is no exception. While general-purpose platforms remain useful, a subtle shift appears to be underway: organizations are exploring AI that reflects their own data, processes, and strategic goals.

This is not yet the dominant mode of adoption. However, as token costs fall and small-model approaches become more powerful and affordable, built-for-purpose AI may emerge as a stronger differentiator across the technology landscape.

Startups may find their edge in lean, efficient models that scale with their pace. Mid-market businesses may increase ROI by tuning AI to domain-specific challenges. Enterprises may unlock sustained advantage by owning and evolving the intelligence that powers their workflows.