Banking and lendingIRIS8
Own the models your bank runs on
For credit, operations and risk teams: a model trained on your loan files and policies, run inside your VPC or data centre by IRIS8.
How it fitsIRIS8
Where the model sits in your bank
It reads from the systems you already run, prepares the work, and hands it to your underwriter.
Your VPC or data centre
Your people
Your systems
Added by IRIS8
IRIS8 engineerScoped, logged access
Your records
Use casesIRIS8
Where banks and lenders start
Each of these is a place a pilot can begin.
- Commercial loan intake
- Credit memo drafts with source citations
- Financial spreading into your own template
- KYC and account-opening document packs
- AML alert triage
- Loan servicing summaries
- Policy questions answered from your circulars
- 01
Credit package in
- 02
Documents sorted against the checklist
- 03
Tax return missing, flagged
- 04
Sent to underwriter
For your reviewersIRIS8
What model risk and security will ask
These are the questions we are asked first, with the answers we give.
- 01
Does the model make credit decisions?
No. It sorts, extracts, summarises and drafts. The underwriter decides, and the record shows who reviewed each file.
- 02
Where is customer data processed?
Inside your VPC or data centre. Applications, statements, alerts and call transcripts are read where they already live.
- 03
Can we see how the model was built and tested?
Yes. The model record holds the base model, the training data lineage, every test score and each approval. It is yours.
- 04
Who can change the model?
A new version goes live only after it passes your benchmark and your model-risk owner approves it. The previous version is kept for rollback.
- 05
Who at IRIS8 can reach our environment?
Named engineers with scoped, task-bound access. Every action they take is written to the same record.
Model record · loan-file-qa v3
- Base model
- Open-weight, licence checked
- Training data
- 12,400 examples, personal data masked
- Trained
- 14 Oct 2026
- Benchmark
- 95 against a pass mark of 92
- Approved by
- Head of Credit Operations
- Deployed
- Your VPC, 21 Oct 2026
- Access
- IRIS8 engineer, scoped, logged
- Previous version
- v2, kept for rollback
One record of the data, training, scores and approvals.
Take this with youIRIS8
Ten questions for any AI vendor
Ask these before any customer data leaves the bank. They apply to us as well.
Ten questions for any AI vendor
- 1Where, physically, is our data processed?
- 2Is our data used to train a model other customers use?
- 3Who owns the fine-tuned weights?
- 4Who can access the model and our data, and how is that logged?
- 5Can we see the test results on our own cases before go-live?
- 6Who sets the pass mark?
- 7What changes without our approval?
- 8Can we pin a version and roll back?
- 9What records do we get, and where are they kept?
- 10What do we keep if we stop working with you?
iris8.ai/banking.html
Cost and changeIRIS8
Cost and change under your control
High-volume routine work runs on a model whose versions change on your release calendar, with rollback in your hands.
Check the cost on your volumeRelease approved by you
QuestionsIRIS8
Questions
Talk to usQ1How long before something is in production?
The pilot targets one workflow in production in 8 weeks. That is a target, not a guarantee.
Q2Do we need our own ML team?
No. Your experts set the pass mark and review sample outputs. IRIS8 does the training, deployment and operations.
Q3Which model do you start from?
A licensed open-weight model, chosen for the workflow and tested against your benchmark before any training.
Q4Can we keep using the frontier API we have?
Yes. It stays available through your gateway for tasks that need it, and there is no lock-in to one model vendor.
Q5How is the pilot priced?
Fixed price, one use case, credited to year one.
Next stepIRIS8