Case study 04 · spectral finance
Productising a machine-learning credit model
ML product marketing · Model positioning · Risk models · Product narrative
Role
Early product and narrative advisor
Focus
ML product positioning, model explainability, storytelling, fundraising narrative
Industry
Machine learning, credit risk, decentralised finance
The challenge
The challenge
Credit is built on trust. Spectral was replacing the usual trust inputs with a machine-learning model trained on on-chain behaviour: wallets, repayment history, liquidation events, and transaction patterns instead of names and bureau files.
The concept was powerful, but a learned risk score is hard to sell. Users, partners, and investors needed to understand what the model was learning from, what it predicted, and why they should rely on it.
My role
My role
I advised on early product positioning and storytelling, helping turn a set of model features and on-chain signals into a narrative that customers, partners, and investors could understand and evaluate.
What we did
What we did
Found the familiar reference point
Credit score is a concept people already understand. We used that familiarity carefully, positioning the product as a transparent, model-driven version of a system that typically sits behind closed doors.
Made the model’s inputs and stakes legible
The narrative focused on what the model learned from and what a portable, identity-less risk score could enable: better access, more relevant financial products, and a system where users’ observable behaviour creates value for them.
Turned model performance into a proof point
Instead of leading with architecture, we framed evaluation and prediction quality as the commercial argument: does the score separate good borrowers from bad ones well enough for a lender to act on it?
Outcome
Outcome
Spectral developed a sharper way to communicate a machine-learning credit primitive to a broad market, helping make the category legible as it grew.
What I learned
Selling a model means selling its inputs and its evaluation, not its architecture.