About
Built to be checked.
Utilyzed is an independent analytics platform. This page sets out what that independence rules out.
Independence, concretely
What independence rules out.
No position
Utilyzed holds no market position. We do not trade energy, capacity, or attributes, and we sit on no side of a transaction whose value our engines estimate.
No benchmark
We administer no benchmark, index, or price assessment. Nothing we publish sets a reference that our own analysis would then be measured against.
No stake
We do not sell hardware, controls, or installation. No fee we earn moves with the answer a model returns.
A number that can move a financing decision should come from an analyst with no other interest in the outcome.
What the principles mean in practice.
- Independent
The analysis is the only thing we sell. There is no trading book that benefits from a dispatch assumption, and no development pipeline that needs a project to pencil. We hold no equity and no trading position, so an unfavorable result faces no pressure inside the company.
- Auditable
Every figure records its source: the bill, tariff sheet, or meter file it came from, and the version of the engine that produced it. A reviewer can disagree with a named assumption. The audit trail is recorded as the work happens.
- Reproducible
Engines are versioned, and every result records the version that produced it. A case can be re-run months later on the original inputs and return the original answer. When a number changes, an input or an assumption changed, and the record shows which.
The regulatory frame
Guidance written for models a bank can validate.
In April 2026 the OCC, the Federal Reserve, and the FDIC issued revised guidance on model risk management, replacing the earlier interagency framework. It draws a line that matters for this work: generative AI sits outside the validated-model framework. A system that can return two different answers to the same question is not something a validator can sign off on.
Utilyzed is built for the review that comes after the analysis. Engines are deterministic and versioned, inputs trace to the document they came from, and assumptions are written down where a validator can read them. The same expectation shows up in CFPB Circular 2026-03 on adverse-action explainability: an automated decision should carry a reason a person can state and check. This is guidance, and we claim no certification. It is the frame we design to.
How access works
Access is by invitation.
That keeps the platform close to the people underwriting real projects, and their feedback is what improves it.
A person reads every request
There is no automated approval and no self-serve tier. A request comes to us directly, and the reply is a straight answer about whether the platform fits the decision in front of you.
Scoped to the decision
Engagements are scoped to the decision at hand: a project in diligence, a portfolio under review, a case that has to survive a credit committee. We would rather answer one question well.
The person who answers a request is the person who works the file.
Request access
Tell us what the number has to support.
Describe the project and the decision it needs to support, and we will tell you whether this is the right instrument for it.