Finance

Credit for the businesses banks cannot score.

Small firms with no credit history now produce structured invoice data every day, because governments made electronic invoicing mandatory. We turn that stream into cash flow forecasts and credit decisions that lenders can act on, with explanations they can defend, evolved to each sector and region.

The problem

The largest employers get the smallest share of credit.

In fast growing economies, small businesses are most of the companies and a fraction of the lending.

1.3Msmall and medium businesses in one economy, contributing a fifth of output against a national target of more than a third
11.5%of bank lending going to them, against a target of twenty percent
Longestworking capital cycle in its region: the time between paying suppliers and getting paid by customers

A bank cannot lend to a business it cannot score, and a small business often has nothing to score: no audited accounts, no credit file, two years of history at most. So the bank asks for collateral the owner does not have, or declines. Meanwhile the business waits sixty to ninety days to be paid by its own customers, runs out of cash in month two, and either borrows informally at a high price or simply stops growing. The problem is not the business. It is the absence of a signal.

That has changed. Mandatory electronic invoicing now forces every business to business invoice into structured, tax authority stamped records. For the first time, a small firm has a daily, verifiable record of what it sells, to whom, on what terms, and when it is paid. The data exists. What does not exist is the model that reads it, forecasts the firm’s cash position, and tells a lender what the risk actually is.

In plain terms

A bank looks at a small business the way a landlord looks at a tenant with no references. The invoice stream is the references: here is every customer, every bill, every payment, with a government stamp on it. We read the references and say, in numbers, how likely this business is to run short of cash in ninety days and how likely it is to repay.

Why the standard approach fails

Credit scoring was built for people with history, in markets that behave.

01

Scores need history the firm does not have

Traditional models score on years of repayment behavior. A three year old firm with no loans scores as nothing.

What we do. The score is built from the invoice stream: customer concentration, payment delays, seasonality, growth, and the reliability of the firm’s own customers.

02

One model for every sector

A restaurant, a contractor, and a clinic have different cash rhythms. A single model gets all three wrong in different directions.

What we do. Models are evolved per sector and region, with the lender’s own portfolio outcomes as the measure.

03

The economy moves and the model does not

Payment terms stretch when the economy tightens, including payments from large and public customers. A model fit to a good year is wrong in a tight one.

What we do. Delayed payment behavior is tracked as a live signal, and the model is re evolved as conditions change rather than retrained on a fixed calendar.

04

Regulators require reasons

A decline must come with specific reasons. A black box score cannot provide them.

What we do. Explainability is a hard constraint. Candidates whose decisions cannot be traced to named factors are discarded before anything else is scored.

The gap a lender could fill, seen ninety days out. From the firm’s own invoices and obligations, the forecast shows cash dipping below the line in weeks six to eight, before a large customer pays. A short facility sized to that gap is a product. A decline is not.
Minimum safe balance Large customer pays Forecast gap: weeks 6 to 8 TodayWeek 12 Illustrative. Built from invoice due dates, customer payment history, and known obligations.
How we solve it

From invoices to a ninety day forecast and a defensible decision.

Two outputs for a lender: a forecast of the firm’s cash position, and a probability of repayment, each with the factors behind it.

In plain terms

We breed scoring models the way we breed everything else. Dozens of candidates read the invoice histories of businesses whose outcomes are already known, and each is scored on how well it predicted who repaid, how fair it was across sectors and regions, and whether its reasons make sense. Candidates that cannot explain themselves are thrown out. The rest are combined and varied, and the next generation is tested again. The lender gets a model that fits its own market, with a reason behind every decision.

Not one model. A population that is bred. Every generation is made from the last: the strongest candidates become parents, their settings are combined, small variations are introduced, and the children are scored on your own cases. Weak lines die out. Strong ones carry forward.
Generation 1Generation 6Generation 14 48 candidates, all differentWeak ones gone, strong ones combinedYour model, chosen by your team
This is an evolutionary algorithm, not a leaderboard. We do not train many models and keep the best. Each generation is bred from the last. The strongest candidates are selected as parents, their settings are combined to make children, small random changes are introduced, and the children are scored on your own scans. Settings that work together survive together.
1. Scoreon your own scans 2. Selectthe strongest as parents 3. Combineparents into children 4. Varysmall random changes Nextgeneration How a child is made Parent A Parent B Crossover: segments from each parent Child Mutation: one setting changed at random Child Each segment is one setting, such as window or threshold
  1. 01

    Connect the invoice stream with consent

    With the business owner’s permission, the firm’s electronic invoices, bank transactions where open banking allows, and public registry data are brought together. Nothing is scraped. The firm sees what the lender sees.

  2. 02

    Baseline on the lender’s own book

    Our models are run against the lender’s historical portfolio, by sector and region. The baseline shows where current scoring declines good firms and approves bad ones, and the size of the opportunity in each segment.

  3. 03

    Evolve the scorer with explainability as a filter

    Candidates are scored on repayment prediction, calibration, the fairness gap across sectors and regions, and stability under economic drift. Candidates that cannot produce specific reasons for a decision are discarded. Credit committee chooses the operating point: how much risk to take for how much volume.

  4. 04

    Forecast cash, not just risk

    For every firm, a thirty, sixty, and ninety day cash position forecast, from expected receipts and known obligations. This is what turns a decline into a product: a short term facility sized to the forecast gap, offered before the firm misses payroll.

  5. 05

    Re evolve as the market moves

    Every repayment and every delay feeds the next generation. When payment behavior shifts across the economy, the model is re evolved on the new conditions, with every version logged for the regulator.

What the evolution changes

Which signals count, in which sector, in which year.

What we evolveWhy it matters for small business credit
Feature set from the invoice streamCustomer concentration, days to payment, seasonality, growth, and invoice disputes are weighted differently by sector. The weighting is evolved per segment.
Customer quality signalA firm is as healthy as its customers pay. How much a slow paying large customer lowers a score is evolved from outcomes.
Fairness constraintsThe gap in approval rates between regions and sectors at equal risk is an objective, so the model cannot win by favoring the capital.
Drift sensitivityHow fast the model responds to a change in economy wide payment behavior is evolved to balance stability and timeliness.
Explainability filterEvery decision must trace to a handful of named factors. Candidates that cannot are discarded, not penalized.
Operating pointApproval volume against expected loss is a Pareto set. Credit committee chooses the point; the model does not.
What you get

A score with reasons, a forecast with a product attached.

Thin file scoring

A repayment probability for firms with no credit history, from verified invoice data.

Ninety day cash forecast

Expected receipts and obligations, with the gap sized for a facility.

Reasons on every decision

Named factors behind approvals and declines, ready for the adverse action notice.

Fair across your market

Measured and reported approval gaps by sector and region at equal risk.

Early warning on the book

Deteriorating payment behavior flagged before a missed installment.

Deployed inside your perimeter

Runs on the lender’s infrastructure. Data residency and consent handled by design.

Where this fits

Built for lenders in markets where the data just appeared.

  • Banks and non bank lenders with a small business mandate and a thin file problem
  • Invoice finance and factoring platforms sitting on the invoice stream already
  • Franchise networks and distributors extending credit to their own partners
  • Markets where electronic invoicing is mandatory and credit bureaus are young

Everyone can score a firm with ten years of history. The hard version is a firm with two years, no loans, a different rhythm from the next firm over, and a regulator asking why. That is where a model evolved to the segment, with explainability as a rule rather than a feature, is the difference between a decline and a customer.

Show us the segment you decline most. We will baseline our scoring on your book and show you the approvals you are leaving on the table at the same loss rate.
Partner With ALBA