Enterprise AI Enters Its Economics Phase

Enterprise AI is moving from experimentation to an economics phase in which measurable returns, lower compute costs, workflow integration and viable pricing determine adoption.

• Legora says customers have shifted from being impressed by demonstrations 18 months ago to demanding an order-of-magnitude view of returns, distinguishing gains in the hundreds of percent from gains closer to 50%.
• Multiverse Computing typically compresses models by 50% to 60% while retaining general-purpose accuracy, with greater reductions possible for specialized applications; smaller models can run on customers’ own hardware when data cannot leave the premises.
• Around 80% of enterprises using AI rely on at least some open-source models, according to Multiverse, and that usage has roughly quadrupled over the past year; a hybrid model is emerging, combining frontier models for critical tasks with local models for routine workloads.
• Specialist applications such as legal technology remain defensible because adoption depends on workflow-specific interfaces and support from domain experts; Legora’s legal engineers are former lawyers, while frontier labs have so far gained little customer traction in legal offerings, according to the company.
• Pricing is still evolving: Multiverse avoids compute costs by running models on customer hardware and charges recurring fees per device or model, while Legora uses consumption-based pricing; legal AI can also be attractive because paralegal work may cost 10 to 100 times more than model usage.

Takeaway: The next source of enterprise AI value is likely to come less from model novelty than from deployment, local execution, workflow adoption and pricing discipline, although the impact on employment and total demand for professional services remains uncertain.