How much Bitcoin should remain locked to cover a liquidity event that may never happen?
Holding the maximum possible reserve is the simplest answer. It is also the least capital-efficient.
While co-designing a product for Bitcoin miners,
@RootstockLabs simulated more than 10,000 possible outcomes to better understand the liquidity required under different conditions.
That gave the team a stronger basis for adjusting the product design, covering potential client needs, and reducing the amount of Bitcoin that would otherwise remain unused.
This is where AI becomes useful for a digital asset company.
Not when it is added to every workflow, but when it helps solve a real business constraint, improve a decision, reduce risk, or change the economics of a product.
The value was not automation for its own sake.
It was better decision-making under uncertainty.
A useful way to assess these opportunities is through a value × feasibility matrix:
• How much value could the use case create?
• And how prepared is the company to implement it responsibly?
AI-assisted reporting, scenario simulation, and stress testing can be quick wins because they build on data and processes that already exist.
Real-time risk pricing, dynamic collateral management, and treasury rebalancing may offer greater strategic value, but they also require stronger data, integrations, controls, and accountability.
The same principle applies to model selection.
Not every problem needs a large language model.
Some are better addressed through forecasting, optimization, anomaly detection, deterministic rules, or a combination of approaches.
The goal is not to use the most advanced model available.
It is to improve a meaningful outcome.
Adapted from a post by
@gca_5772, following insights shared by Andrea Cremonino during the AI in Bank Treasury workshop. Original post linked in comments.