Most cannabis B2B credit decisions get made on gut feel.
Reklaim Credit Solutions built something different.
Here is the technical foundation behind our credit scoring engine.
Not the “secret sauce”. Just the architecture…(sorry, not giving away mother cuts!)
Utilizing proprietary variables, Reklaim Credit Solutions uses a multi-month ensemble model to predict a client’s customer account risk and to set appropriate account credit limits.
Our Ensemble Models utilize Boosting as the methodology, either Decision Tree, Neural Networks or Discriminant Analysis as the classification system and AdaBoost.M1(Freund) or AdaBoost.SAMME as the model weighting methodology.
The model development process creates numerous “weak” models and then weights and combines them to produce a strong model with accuracy at or in excess of 98% on a regular basis.
The model with the smallest validation error gets selected.
Why an ensemble model?
Because the ensemble is stronger than any individual model. That is not a slogan. It is a fact.
The inputs are continuous independent variables drawn from multiple AR aging periods. Total balance. Overdue balance. Monthly payments. Proprietary aging analysis.
Variables get selected through bivariate analysis, means analysis, and correlation analysis. The output is a single dependent variable: GOOD or BAD. Tested and validated against thousands of accounts and growing.
Every account gets a credit score and a recommended credit limit. Predicted GOOD returns a dollar value. Predicted BAD returns zero.
The model does not sit still.
Every month we append the latest AR data, compare prior predictions to actuals, compute prediction error, and roll the window forward. We track override accuracy. We retrain our models when predictive error does not meet out standards.
Three classifications drive operational decisions.
GOOD accounts auto-accept orders within their credit limit.
Predicted BAD accounts require payment before new orders.
Already BAD accounts stay flagged until status changes.
SIMPLE…NOT!
How many other organizations selling credit analysis into cannabis are willing to publish their methodology this openly?
Member-driven credit associations pool tradelines and hand back aggregated reports. Useful for sure, but the underlying scoring logic is not the product.
Other providers wrap their offerings in language like proprietary algorithm or AI-powered without ever telling you what sits underneath.
We believe in transparency.
If we are asking operators to make 8 figure credit decisions based on our scores, they deserve to know how those scores get built.
Architecture. Variables. Validation. Revalidation cadence.
Cannabis operators have been flying blind on counterparty risk for too long.
Receivables aging is the most underused dataset in this industry.
We turned it into a rating engine, and we are willing to show our work.
That is what Reklaim Credit Solutions does.
Is this is just V.1!