Skip to the flow

Credit decisioning

Credit policy your risk team can change without a deploy.

Bureau pull, affordability, scorecard, limit assignment — drawn as one flow, tested against real applications, and published the day the policy changes.

See the flow

The problem

The policy moves faster than the release train.

A credit policy is never finished. Cut-offs move with loss performance, a new bureau attribute arrives, a segment underperforms and needs its own tier. Each of those is a small change to the logic and a large change to the calendar when the logic lives in a service only two engineers understand.

The cost is not only speed. When the rule is buried in code, the person accountable for the loss rate cannot read it, the person who can read it is not accountable for the loss rate, and the answer to "why was this declined" is reconstructed after the fact from logs.

What the flow looks like

Every step is a node someone can point at.

  1. 01

    Take the application

    The applicant's data arrives against a schema you define, so a missing field is a clear error rather than a silent null halfway down the policy.

    Input
  2. 02

    Pull the bureau

    Call the credit bureau and open-banking providers you already have contracts with. The response lands in the flow as data any downstream node can read.

    Connection
  3. 03

    Derive the attributes

    Turn raw bureau and transaction data into the ratios your policy actually uses — utilisation, affordability, months since delinquency — in sandboxed Python.

    Code
  4. 04

    Score the applicant

    Points per characteristic, laid out as a table a credit analyst can read and change, not a matrix multiplication in a file.

    Scorecard
  5. 05

    Apply the cut-offs

    Score bands, segment, product and channel meet in one table, with an explicit row for every combination instead of nested conditionals.

    Decision Table
  6. 06

    Assign the limit

    Limit and price from the score and income grid your committee signed off, kept as a grid rather than translated into code.

    2D Matrix
  7. 07

    Refer the edge cases

    Applications near the line stop and wait for an underwriter instead of being forced into an automatic answer.

    Manual Review
  8. 08

    Return the decision

    Approve, refer or decline, with the reasons and the version that produced them, ready for an adverse-action notice.

    Output

Data you can pull in

Your contracts, called from inside the flow.

A Connection node calls the providers you already pay for, with your credentials. ArboRule decides which calls to make and what the answers mean together — it does not resell the data or sit between you and your vendor.

Bureaus and business data
  • Experian
  • Creditsafe
  • Codat
Banking and income
  • Plaid
  • Truv
  • LoanPro
Business verification
  • Middesk
  • Baselayer
  • Kyckr

What changes

The policy stops being the bottleneck.

A

Change policy in an afternoon

Edit the table, test the new version against recorded applications, publish. The endpoint your origination system calls does not change.

B

Explain every decline

The version, the inputs, the path through the flow and the reasons are recorded with the decision — available when a customer, an auditor or a regulator asks.

C

Ramp a new policy safely

Route a share of live traffic to the new version, keep it sticky per applicant, and compare outcomes before it carries everything.

Questions

What teams ask first.

Can we keep our existing credit bureau contracts?

Yes. ArboRule calls the providers you already have contracts with — Experian, Creditsafe, Plaid, Truv and others — using your credentials. The flow orchestrates the calls; it does not resell the data.

Who can change a credit policy?

Whoever your workspace permissions say can. The flow is a visual graph, so a credit analyst or policy owner can edit a cut-off table directly, test it in Sandbox, and publish a version without an engineering ticket or a deploy.

How do we prove why an application was declined?

Every decision records the flow version, the input, the output and the node-by-node trace of the path taken. That is the evidence behind an adverse-action notice or a regulator's file review, reproduced from the run itself rather than reconstructed from logs.

Ready when you are

Draw this decision instead of describing it.

Build the flow in Sandbox, run your own cases through it, and see the trace before anything reaches a customer.

See what it costs