Senior Data Scientist, Credit & Fraud Risk

The short version

MarkIII (MKIII) is a profitable, fast-growing company that specializes in off-the-shelf, insurance-backed lending programs. Our products provide lenders with capital relief and risk protection on portfolios originated on leading digital marketplaces (like Credit Karma), underwritten by MKIII's proprietary credit models.

MKIII succeeds by ensuring that all parties make money. We call it "The Formula," and The Formula works when we hit our risk targets — that's where you come in.

Why this role exists

Most lending risk functions are inherited rather than designed. Credit strategy lives in one team, fraud in another, and everything each side learns — the anecdotes, the pattern recognition, the hard-won rules of thumb — stays on its own side of the wall. So nobody can say what a loss is actually made of.

We don't carry the organizational debt that forces most lenders into that shape. What we do have is a real book with real history, and a growing team that mines it endlessly to keep our credit screens, verifications, and funnel tuned against "Risk" with a capital R — credit and fraud in one view — while making sure the best borrowers fund fast.

We want someone who has seen how this gets done at scale, has opinions about why parts of it are wrong, and wants to build the version they'd design from scratch.

Who you'll work with

You'll report directly to the Chief Risk Officer, who is in this data daily — a real counterpart to pressure-test your conclusions, not a vacuum and not a rubber stamp. You'll also work with the C-suite daily. Risk analytics here is a seat at the table, not a request queue. The function is growing, and you'll be a senior voice in how it gets built and who joins it.

The scope below is the team's agenda, not a one-person to-do list.

The work

BackbookBackbook performance and loss forecasting

• Live in the time series. Vintage-level loss curves, roll rates and transition matrices, delinquency migration, prepayment, recovery timing and severity — known well enough that an anomaly announces itself before it reaches a monthly report.
• Forecast cohorts to terminal loss. Lead CNL forecasting by cohort, vintage, and segment: the estimate, the confidence around it, what would have to be true for it to be wrong, and the macro stress cases around it.
• Decompose the drivers. When performance moves, isolate why — credit-band mix, channel and lead source, attribute shifts, sizing and term, seasoning, servicing, macro. Be clear whether you're looking at a mix shift or real credit deterioration.
• Feed the valuation. Your loss and timing curves drive cash-flow valuation of the portfolio and reserve adequacy. You're a contributor to that work, not a data supplier to it.

PolicyCredit policy and buy-box strategy

• Find the expansion. Segment the reject population, size incremental volume against incremental loss, and bring us a swap-set analysis with an honest number attached.
• Size every policy change before it ships. Approval rate, volume, expected loss, blended cohort EL, and the effect on staying inside the insured target band.
• Help manage the box mid-cohort. As early vintages come in, recommend the tightening or loosening that keeps the blended cohort on target — then track whether it worked.
• Keep the policy documents true. Current, versioned, audit-ready: the rules as they actually run in production, the analysis behind each one, and the change history. Our lending partners and our carrier read these during diligence.
• Design the tests. Champion/challenger, holdouts, controlled expansions — measurement plan written before the test starts.

FraudFraud strategy and analytics

Same discipline, second loss channel — and where "the way it's been done" is most obviously broken.

• Build the fraud performance view across both books: application fraud, synthetic identity and ID theft, first-party and never-pay behavior, income and employment misrepresentation, and on the SMB side business verification failures, straw owners, and shell entities.
• Separate fraud loss from credit loss. Own fraud tagging and attribution so each strategy is measured against the losses it actually causes. Get this wrong and both forecasts are quietly corrupted.
• Evaluate the vendor stack on our data, not their deck. Identity verification, device and behavioral signals, bureau fraud scores, KYB and business verification, income and bank-account verification. The case is incremental fraud caught against good applicants lost against cost per decision.
• Place controls in the funnel. Pre-qual, application, pre-funding, or post-funding monitoring — each carries a different cost, a different friction profile, and a different interaction with the credit rules already running there. Own that positioning.
• Own the fraud/credit interaction. No fraud declines silently absorbing credit risk; no credit cuts taking credit for fraud savings. Almost nobody does this well, and it's much of why this role exists.
• Monitor rule performance. Hit rates, false positives, verification pass rates, step-up conversion, rule overlap. Backtest on historical applications before go-live, and retire rules that have stopped earning their friction.

GovernanceGovernance and communication

• Keep methodology, assumptions, data lineage, and results in audit-ready form. Our lending partners, our carrier, and their diligence teams read this work.
• Support model validation and program diligence.
• Turn a complex credit narrative into an executive read: the finding, the number, the confidence, the recommended action.

What we're looking for

We're not asking for an ML research background, an MLOps stack, or a distributed systems résumé. We want someone unusually good at reading a loan portfolio — who also wants to decide how it should be read.

Tools

R and the tidyverse, Python, SQL, Experian credit attributes including trended data, our production decisioning stack, and the marketplace integration layer.

On top of that, we're building an AI-forward analytics function on purpose. Claude and adjacent tooling are in the daily workflow — exploratory analysis, code, documentation drafts, tearing apart a fraud vendor's methodology before we buy it. Use them to move faster, and be rigorous about where they don't belong: a credit decision has to be explainable to a partner, a carrier, and a regulator, and "the model said so" is not a defense. If something that would make this work better isn't here yet, you'll have latitude to bring it in.

Compensation & benefits

Salary

Competitive and commensurate with experience, within the range of $130K–$200K. This is a good-faith estimate of what we expect to pay for this role at hire, and placement depends on depth of portfolio and policy experience and breadth across credit and fraud.

Equity

Company equity, so you benefit directly as the company grows.

Quarterly bonuses

Revenue-based bonuses paid quarterly.

Medical

100% of your medical premiums covered by the company, with the option to pay the difference to upgrade to the next tier.

Holidays

All federal holidays off plus additional company-wide days off.

Flexible time off

Sick leave as needed, and time off when you need or want it.

Location

We'd like you in the Washington, DC area so we can work through this in person — whiteboards beat Zoom for portfolio work. For the right person, we'll go fully remote.

Applications are accepted on an ongoing basis until the role is filled.