Agentic underwriting for alternative lenders

Every credit file worked end to end. Your committee still signs.

RISKW takes small and mid-market credit files from raw documents to a scored, policy-checked credit memo, then follows each loan through the portfolio. Five connected products, built for tickets too small to carry a manual underwriting process.

Request a walkthrough See the suite

Five products on one credit spine

Each product works on its own and plugs into the next. Data feeds the models, the models feed decisions, the agent turns decisions into credit memos, and portfolio outcomes flow back to recalibrate the models.

01 →DataOne governed loan record 02 →ModelsPD, LGD and EAD scorecards 03 →DecisionScores and policy as a service 04 →UnderwriterThe agent that works the file 05 ↺PortfolioCohorts, losses, feedback

RISKW DataOne governed record of every loan

A layered pipeline that turns servicing-platform exports into a reporting snapshot you can rebuild for any past date.

Five layers, from raw ingestion through normalisation and business logic to analytics and reporting
Days past due, default and disposition tracking per payment, with daily and monthly portfolio snapshots
Full change history on every record, so a month-end can be restated exactly as it was
NACE Rev. 2.1 sector mapping and regulatory risk-data formats
Drift reports on incoming data before it reaches a model

OutputDated, versioned reporting schema · warehouse or S3

RISKW ModelsScorecards you can defend in front of a validator

A disciplined build-and-validate workflow for application and behavioural scorecards, with the documentation written as the model is built.

PD, LGD and EAD models with optimal binning, WOE and information value screening
Out-of-time samples, reject inference and correlation and VIF checks before fitting
Validation on KS, Gini, PSI and CSI, calibration and Hosmer–Lemeshow
Model Development Document template and dated decision records for each build
Post-deployment monitoring: rank order, overrides, vintages, actual versus expected

OutputVersioned model · development document · monitoring pack

RISKW DecisionScores and policy in a single call

The production engine that reads financial statements, computes ratios and returns grades and policy outcomes with reasons attached.

Financial-ratio engine across reported periods: leverage, liquidity, coverage, profitability
PD grades and LGD returned together, with each score driver's contribution
Young-company and sector overrides applied as explicit, logged rules
Credit-policy checks that return pass, flag or fail with the threshold that triggered it
REST API, callable from your origination system or a spreadsheet

OutputScore · grade · LGD · policy result with reasons

RISKW UnderwriterThe agent that works the file

An underwriting agent that carries each application through eight steps and drafts the credit memo, leaving the credit decision with a named person.

Intake, document collection and a completeness check that lists exactly what is missing
Data extraction reconciled to source, then scoring and policy through RISKW Decision
A self-contained credit memo with financial tables, charts and score contributions
Overrides against the model grade or policy require an authorised role and a written reason
Raw documents kept byte-for-byte; every artifact hashed; the file locks once decided

OutputCredit memo · review record · offer or decline

RISKW PortfolioSee cohort damage months before the averages do

Portfolio monitoring for mixed books of term loans and revenue-based finance, read by vintage first, with every number drilling down to the facilities behind it.

Executive overview, originations, vintage and cohort, losses and NPL, profitability
A dedicated high-ticket view and a watchlist for the facilities that move the book
Underwriting feedback: broker dispersion, override performance and mix drift
Immature cohorts flagged and left blank, never extrapolated
Dated snapshots with a history selector, plus an encrypted offline copy for investors and committees

OutputPortfolio pack · investor-ready snapshot

Choose how much the agent decides on its own

Every application is created in one of two modes. You can run both side by side: automated for repeat, low-ticket files and human-in-the-loop for larger or unusual credits.

Default

Human-in-the-loop

The agent produces each step's output and waits. An analyst agrees to unlock the next step, or rejects to rewind from that point, correct the inputs and rerun.

0102030405060708

For volume

Automated

Documents are checked on arrival and the assessment runs as soon as the file is complete. The agent stops only for missing documents, a processing error or human review.

0102030405060708

In both modes, a named reviewer approves or declines, the final outcome must match that review, and no offer leaves without a recorded decision.

Built to be examined

Credit committees, auditors, warehouse lenders and rating agencies all ask the same question: how did you get this number? RISKW keeps the answer attached to the number.

Lineage to source

Every figure in a memo or dashboard traces back to the document or source record it came from. Reporting follows BCBS 239 principles.

A consistent default

Default follows the Basel definition, 90 days past due or unlikely to pay, applied the same way in data, models and reporting.

Effective challenge

Model development is kept separate from validation, in line with SR 11-7. PD and LGD estimation follows EBA GL/2017/16.

Override discipline

A decision that departs from the model or policy carries the reviewer's role and reason, and override performance is tracked in the portfolio.

Immutable decisions

Rerunning an assessment clears downstream reviews. Once a decision is recorded, the application is read-only.

Monitoring that triggers action

Population stability, rank order and actual-versus-expected are reported on a schedule, so recalibration starts with evidence.

For lenders whose tickets can't pay for a manual file

The same diligence a large bank applies to a large loan, at a cost that works for small and mid-market credit, and with the data discipline capital-markets funding expects.

SME term lending

Amortising loans underwritten from statutory accounts and management figures.

Revenue-based finance

Merchant cash advance and revenue-based books compared to term loans on a funded-amount basis.

Infrastructure and energy

Small and mid-market project credit with offtake, permits and technical reports in the file.

Private credit platforms

Lenders funding through warehouses, notes or investors who need clean loan tapes and reporting.

Bring one real credit file. We'll run it through the suite with you.

A working session with your own documents: scoring, policy, the drafted memo and where it would sit in your portfolio.

Request a walkthrough info@riskw.xyz