Credit and invoice risk, decided on evidence.
Ashcroft Finance advises lenders, factors and financial institutions on credit and receivables risk, and builds the scoring systems that carry the decision.
Our platforms are live, with Python and compiled C++ scoring paths.
{ "risk_score": 94.4, "risk_band": "Low Risk", "default_probability": 0.056, "key_factors_considered": [ "how long you've had an active digital wallet history", "how steady your income is month to month", "your track record repaying any prior small credit" ]}
{ "risk_score": 96.1, "risk_band": "Low Risk", "default_probability": 0.039, "financing_terms": { "decision": "APPROVE", "advance_rate": 0.927, "discount_fee_rate": 0.045 }}
Responses from our live platforms, shortened to the main fields. Input values are illustrative.
Advisory, backed by systems we built ourselves.
We work with institutions that lend against people and invoices. Each engagement ends with a decision you can act on and the reasoning behind it.
Credit risk for the unscored
Applicants with thin files, irregular income or mobile-money histories are often poorly served by bureau-led models. We re-score a sample of your applicants on repayment behaviour and show where your current decisions and ours diverge, and why.
Deliverable. A scored comparison against your current decisions, with the factors behind each difference.
Receivables and invoice-finance risk
An invoice is paid by the buyer, not the supplier. We assess invoices on both sides of the transaction and set advance rates and fees from the result, including when to decline.
Deliverable. Invoice-level financing terms for a sample of your book, set against your current terms.
Decision-system engineering
We design, build and stress-test the systems around a lending decision: scoring engines, ledgers, audit trails and the controls that keep them correct under load.
Deliverable. A working, tested system, or a written technical review of yours.
Platforms behind the work.
Our advisory runs on two proprietary platforms. Both are live and available to institutions through authenticated APIs.
Meridian
Alternative-data credit scoring. Scores applicants from behavioural signals instead of bureau history.
API referenceSupply Chain Finance
Invoice-finance risk. Returns an advance rate and fee, or a decline.
API referenceEngineering record.
Checks run against the platforms, as built and as deployed.
| Check | Result | Verified by |
|---|---|---|
| Engine parity | Independent Python and C++ implementations return identical scores across 5,000 randomised credit applicants and 3,000 randomised invoices. Maximum difference: 0.0000. | Stress tests |
| Decision logic | The advance rate never rises as default probability rises, checked across 1,000 steps. Probabilities at or above 45% return a decline rather than a repriced offer. | Automated and stress tests |
| Directional sanity | With every other input held equal, steadier income never produces a worse credit score. | Automated test |
| Robustness | Extreme and boundary inputs stay within valid ranges, missing fields are rejected, and 200 concurrent Meridian calls return valid results. | Stress tests |
| Access control | Every scoring endpoint requires a valid API key. A missing or invalid key is rejected with HTTP 401 before any scoring happens. | Automated tests and live checks |
| Live parity | Reference cases return the same result on the deployed services as in local testing: a credit applicant at 94.4 and an invoice at 96.1. | Checks on the live services |
Test our judgment against yours.
Engagements are paid, scoped in writing and time-boxed, and most begin as a pilot. The aim is a direct comparison with the process you already run.
Scope and NDA
We agree the evaluation window, the data you will share and how the result is measured. A mutual NDA is signed before any data moves.
Access
Your team receives an API key, the reference documentation and example requests.
Analysis
Your anonymised sample runs through our platforms alongside your current process. Anonymised data only, and we do not retrain models on it without your written consent.
Review
You receive the scored results and a comparison against your current process, with the reasoning behind each difference.
How we work.
Measured before claimed
Each platform ships with automated tests covering edge cases, adversarial inputs and concurrent load, and a record of what those tests showed.
Decisions with reasons
Every output carries the factors behind it, so a credit or risk team can see why a decision came out as it did and challenge it.
Private by default
Engagements use anonymised data, begin under a mutual NDA, and keep your data out of model training unless you agree otherwise in writing.
Begin an engagement.
Write to us with the area you want examined and a line on your current process. We read every enquiry and reply personally.
ashcroftfinance@aldmarch.comWhat helps us scope quickly
- Your market and the borrower or invoice segment you serve
- Rough monthly decision volume
- How you decide today, and what you would compare against
- The data you could share in anonymised form