Fairness and impact tests on each credit model change, before it goes live
TL;DR Modelward replays every new or changed credit model on a lender's own past applications and shows who would gain or lose: by gender, age, region and pin-code proxies. A risk reviewer signs a short impact report before launch. Digital lenders and NBFCs pay per model release.
Lenders retrain credit, pricing and collections models often, frequently with fintech partners. Each change can quietly shift who gets approved and at what rate, and few smaller lenders check outcomes by group before they flip the switch.
In August 2025 an RBI committee published the FREE-AI report, with seven principles and 26 recommendations[1][2]. These include a board-approved AI policy, AI-specific audits, bias testing and telling customers when AI is involved[3].
RBI surveys cited in commentary found large banks and fintechs experimenting while smaller entities held back, naming cost, data quality and governance as reasons[4]. So a model team at a mid-size NBFC has the duty to test but rarely the tooling or the time.
Before a model goes live, the lender uploads the new model's scores on 6–12 months of past applications next to the current model's. Modelward compares approval rates, limits and interest bands across gender, age, state and pin-code groups, and flags inputs that act as stand-ins for protected traits. It also lists which applicants flip from approve to reject. You get a two-page impact report with charts, plus an evidence trail for internal audit.
It sits next to a lender's AI policy and inventory work, which other tools handle, and doesn't replace it.
The seed pitched fully automated impact assessments. I reframed it, because a judgement about harm to borrowers needs a named person and an automated verdict would not satisfy an auditor. Modelward automates the testing and the paperwork, and a named reviewer makes the call.
What it does, in regulatory terms - model testing and decision support for the lender's own risk team. It does not make credit decisions, certify compliance or give legal opinions. The FREE-AI report is a committee report, not a binding direction, so check the RBI site for any directions that follow.
The model risk or credit analytics team uses it and the lender's chief risk officer pays. Your targets are digital lenders, middle-layer NBFCs and housing finance companies that change models several times a year. AI governance platforms don't publish prices[5][6]. My assumption is ₹1–3 L per model release, or ₹8–15 L a year for unlimited releases, which sits below the cost of a consultant's one-off model review.
The FREE-AI report landed in August 2025 with an explicit push for audits and bias testing[1][3]. DPDP Rules now give borrowers' data clear penalties, up to ₹250 crore for security failures[7]. Internal auditors are adding model fairness to their plans this year.
Nobody publishes a figure for what Indian lenders spend on model fairness testing. The narrowest signal is the RBI's own finding, reported in commentary, that smaller regulated entities lag on AI governance because of cost[4]. Next door, governance governance platforms already sell policy packs for the EU AI Act, NIST and ISO 42001[5].
Global vendors sell enterprise AI governance platforms to global firms[5][6]. One compliance platform publishes FREE-AI guidance and sells governance automation[4]. Big consultancies run one-off AI policy projects for large banks. None that I found runs outcome tests on Indian lending groups such as state and pin code as a per-release service. That's the gap.
| Difficulty | Moderate (46/100) |
| Starting capital | ₹55–110 L |
| Time to launch | 6–9 months |
The statistics are standard stuff. The hard part is getting past-application data out of a lender securely, handling missing gender and age fields and writing reports an RBI inspector and an internal auditor will respect. Expect two or three unpaid pilots before any contract.
No licence is needed for testing software, as long as you don't make credit decisions, certify compliance or give legal advice. Your buyers are regulated though.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
AI shouldn't make the final call on whether a model harms borrowers. A named reviewer at the lender decides, and Modelward only does the testing and paperwork.
Want to work out the AI layer for Modelward? Talk to Yash →
Here's what can go south -
I think it's a sound idea for a founder from credit risk modelling or model validation at a bank. With no lending contacts, though, I'd skip it. First prove that three lenders will hand over past-application data and pay for a report before any binding RBI direction.
Plan for roughly ₹55–110 L, mostly statisticians, security work for lender audits and long sales cycles.
No, as long as you don't make credit decisions or certify compliance. RBI's FREE-AI report is advisory, so watch for binding directions.
Yes. Fair lending testing already sells in the US and Europe. Each market has its own protected groups and rules.
The free Modelward validation kit has the line-by-line capital breakdown, the effort scoring, the rules and data checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for risk and model teams, and 5 assumptions to test. Download the kit →
Thinking about building Modelward or something close to it? The make-or-break work is getting lenders to share past-application data and act on what the tests show, and that's where I can help. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.