OpenGrade

One bank connection.
Verified identity, scored trust.

A privacy-first platform on Israeli Open Finance. It proves a person is real, then prices their financial risk, without storing a single document.

Real-person verification 0–100 trust score Zero data stored

Amit Zamir · Noa Elmakies · May Gurevich · Hai Tal opengrade.cs.colman.ac.il

The problem

Not knowing who's real costs billions.

Marketplaces, merchants, lenders and landlords all judge people they've never met, with tools that are opaque, invasive or trivially faked. The bill arrives as fraud losses, compliance overhead and customers who walk away mid-signup.

$56B

E-commerce fraud

Value of fraudulent e-commerce transactions in 2025 alone. Merchants still can't tell a real buyer from a stolen card.

Juniper Research · 2025

$35B

Synthetic identities

Drained from financial institutions in one year by people who never existed, and still climbing. Gen-AI made fabricating them cheap.

Federal Reserve · SIF Toolkit, 2025

$73M

Identity management

Average AML and KYC spend per institution, per year. 70% still lost clients to slow onboarding.

Fenergo · 2025

51%

Bots outnumber people

Of all web traffic is now automated, and account-takeover attacks climbed 40% in twelve months.

Imperva Bad Bot Report · 2025

One root cause behind all four: there is no cheap, unfakeable way to prove a person is real and price their risk. So everyone over-collects documents, over-pays for manual checks, and still gets defrauded.

OpenGradeProblem

This isn't an inconvenience, it's a very expensive one. Fraudulent e-commerce transactions came to fifty-six billion dollars in 2025 alone, because merchants can't tell a genuine buyer from a stolen card or a bot at checkout. Synthetic identities, people who simply don't exist, drained thirty-five billion dollars from financial institutions in a single year, and the Federal Reserve's own toolkit says generative AI has made manufacturing them almost free. On the identity-management side, the average institution burns nearly seventy-three million dollars a year on AML and KYC operations, and seventy percent of them still lost customers in the past year because onboarding was too slow and too invasive. And more than half of all web traffic is now automated, with account takeovers up forty percent. Four different industries, one root cause: there is no cheap, unfakeable way to prove a person is real and price their risk. So everyone over-collects documents, over-pays for manual checks, and gets defrauded anyway. That's the money we're going after.

The product

One bank login. Two answers.

One consented bank login Israeli Open Finance · live today · returns zero identifying data

Verify

1 credit

Is this a real person?

8 evidence checks · pass / warn / fail

Bank authentication plus a 30-day activity peek. No score, no documents.

Score

2 credits

Can I trust them financially?

78/100

7 explainable factors · 6 months of behavior

A trust score with the reasons behind it. Upgradeable from a Verify.

Raw bank data is read in memory and deleted the moment it is used

OpenGradeProduct

Here's the product. One consented bank login, two answers. This is possible today because of regulation: Israel rolled out Open Finance, the local PSD2 equivalent, so a person can consent once and share real bank data through a regulated API — and that API returns no personal identifying information by design, which is exactly why we score behavior and treat the connection itself as proof of a person. From that one connection we sell two depths. Verify answers a narrow question: is this a real, active person? It leans on evidence the consent flow already produces — an email one-time code, a national ID entered at the bank, a completed strong authentication at a regulated bank that has itself KYC'd the applicant — plus a thirty-day peek at account activity that rules out synthetic, mule and dormant accounts. Eight evidence rows, each pass, warn or fail. No score, no ML, no documents, no selfie. That's one credit. Score goes deeper: six months of real banking behavior becomes a transparent zero-to-a-hundred trust score across seven factors, and every score ships with the factors behind it, so it's the opposite of a black box. That's two credits, and a Verify can be upgraded into a full Score while the consent is still live, for the difference. Either way, the raw data is read in memory and deleted the moment it's used.

Why us

Built, validated, and live.

Learned, not guessed

Factor weights are derived by ML per use case, from 2,000 synthetic applicants using non-negative regression.

Validated on real data

Benchmarked on UCI German Credit (1,000 records). Beats the uniform and prior baselines on 4 of 5 use cases.

Zero-PII by design

Raw bank data is deleted in the same DB transaction that saves the score. Provable, not promised.

Deployed & running

Running live on a real bank integration with CI. A working product, not just slides.

7explainable
factors

5use-case
models

0PII / raw data
stored

<2mincreate check
→ score

OpenGradeWhy Us

This is what separates us from a slide. It's built, it's validated, and it's live. First, the scoring isn't hand-waved. An ML pipeline derives the factor weights per use case from 2,000 synthetic applicants using non-negative regression. Second, we tested it honestly: on held-out data the learned weights beat both the uniform and the prior hand-tuned baselines on every use case. We also tried an external benchmark, UCI German Credit, and it did not transfer, so we report that openly rather than burying it. Third, privacy is enforced in code: raw bank data is deleted in the same database transaction that writes the score. And fourth, it's deployed and running on a real bank integration with CI. As a team we're CS finals in the fintech and algorithmic-trading track, and we shipped a working product, not a mockup.

Potential markets

One integration. Three markets.

The same bank connection is a product in every industry that has to answer "is this person real, and can I trust them?" Each one already pays for a worse answer.

$29B

Identity verification

Global market by 2030, up from $14.3B in 2025 at 15.4% CAGR. We enter as the document-free option, at one credit a verification.

MarketplacesDatingGig & P2P

MarketsandMarkets · 2025

$131B

E-commerce trust

Fraudulent transactions per year by 2030, up from $56B in 2025. Merchants pay on every wrong call, so a verified buyer is worth money.

Checkout riskSeller onboarding

Juniper Research · 2025

$206B

Financial trust & KYC

Global financial-crime compliance spend, every year. Scoring replaces statements, payslips and manual review.

RentalsLendingHiring

LexisNexis Risk Solutions · 2023

Beachhead → expansion. Start with Israeli rentals and marketplaces, where we are already live, then open the Verify API to e-commerce and lending platforms. Open Finance is the local flavour of PSD2, so the same product ships into the UK and EU without rebuilding the integration.

OpenGradePotential Markets

So where can this go. One integration, three markets we can sell into, and each of them is already paying for a worse answer. First, identity verification: a fourteen billion dollar market today, twenty-nine billion by 2030, growing at fifteen percent a year. We enter it as the cheap, document-free option for marketplaces, dating apps, gig platforms and bot defense. Second, e-commerce trust: fraudulent transactions are heading from fifty-six billion this year to a hundred and thirty-one billion a year by 2030, and merchants pay on every wrong call, so a verified buyer at checkout is worth real money to them. Third, financial trust and KYC, where financial institutions already spend over two hundred billion dollars a year worldwide on financial-crime compliance; behavioral scoring replaces bank statements, payslips and manual review for rentals, lending, BNPL, hiring and B2B credit. Our route is a beachhead and then expansion: Israeli rentals and marketplaces first, where we're already live, then the Verify API for e-commerce and lending platforms. And because Israeli Open Finance is the local flavour of PSD2, the same product ships into the UK and Europe without rebuilding the integration.

The team

Verify a human. Score the risk. Store nothing.

Amit ZamirBackend & architecture

Noa ElmakiesProblem & market

May GurevichProduct & design

Hai TalScoring & validation

CS finalists in the Fintech and Algorithmic-Trading track. We designed, built, validated and deployed the entire platform ourselves.

What's next : real pilots in rentals and online marketplaces, and an open Verify API for platforms.