01Approval leakage
Good deals leave without a fight
The file was fundable. The right structure just never surfaced. Since the loan never boarded, the lost yield never shows up on a report.
Economic impactMargin never booked
Underwriting decision support for auto finance
Bring your best underwriter’s judgment toevery deal.
Underbot AI scores every auto application the way your best underwriter would, then shows the work. A deterministic, lender-owned scorecard with an advisory AI second read — so good deals stop walking, bad deals stop funding, and every call is one you can defend.
This cycle rewards the most consistent underwriting — not the most underwriters.
Built for subprime and near-prime auto finance companies.
Policyyour rules own the score
Humanyour team keeps the call
Traceevery point is visible
In production with Tracir Financial Services, a multi-state auto lender
A deterministic scorecard — not a black-box model
Onboarding a limited number of lenders at a time
What changed
The lenders who win the next cycle won't have the most underwriters. They'll have the most consistent underwriting.
Files are thinner. Fraud is sharper. Margins are tighter and dealers want decisions faster than ever. The old model — where credit judgment lived in a few veteran heads and got applied a little differently at every desk — can't keep up. The gap between your best underwriter and your newest one stopped being a training problem. It became a line on your P&L.
The old model
Credit judgment lived in a few veteran heads and got applied a little differently at every desk.
What the cycle rewards now
One consistent standard on every file — the same answer whoever is at the desk.
The cost of drift
Your policy can be sound and still leak profit.
The leak starts when the same deal gets a different answer at a different desk. You rarely see the decision that should have changed, only the economics it leaves behind.
Underwriting drift ledger
Where economics escape02Risk leakage
Stacked risk hides inside approved
High LTV, thin cash down, and aging collateral can look tolerable one exception at a time. The stack becomes obvious only after performance turns.
Economic impactLoss arrives later
03Knowledge leakage
Your best underwriter becomes the bottleneck
When veteran judgment lives in memory, new underwriters learn it one file at a time. Speed drops, answers vary, and growth adds inconsistency.
Economic impactCapacity stays trapped
Three leaks. One root cause.Same policy. Different answers.
Underbot turns your best credit judgment into one visible, repeatable standard, so drift becomes something you can find, measure, and fix.
Where this fits
Underwriting intelligence beside your team, not another platform in their way.
What it is
- Decision support: a scorecard plus an advisory AI read
- Deterministic — the same inputs always return the same answer
- Lender-owned — you set every factor, threshold, and rule
- Human-in-control — your team makes the final call, always
What it isn't
- Not a loan origination system replacement
- Not auto-decisioning that funds deals on its own
- Not a black-box risk score you can't explain
- Not a generic chatbot bolted onto your workflow
How the score works
Every point has a reason. Every decision has a trail.
Your best underwriter's logic becomes a visible, repeatable scorecard. Work the deal and every point, tier, and decision updates on the spot.
01Deterministic
Same inputs. Same score.
02Lender-owned
Your weights. Your policy.
03Instant
Restructure in real time.
Illustrative decision trace
Policy v3.4Deal #10482Human controlled
01 / Signals inWeighted risk dimensions
25+ policy factorsCredit8 signals
Bureau and repayment history
FICORiskViewTruDecisionPaid auto historyCharge-offsLatesBankruptcyDelinquency
Capacity4 signals
Ability to carry the payment
DTIPTIGross incomeMultiple income sources
Deal structure5 signals
The terms you can reshape
Cash downLTVLoan termProposed paymentAmount financed
Collateral4 signals
Asset value through payoff
Vehicle ageMileageValueUseful life at payoff
Stability5 signals
Residence and employment depth
HousingTime at addressJob tenureEmployment typeIndustry
02 / Decision outPolicy engine
RecalculatedScore84policy result
Tier BApprove
Clear decision, with every applied point on the record.
Policy baseline100
Factor points+18
Bonuses+6
Penalties−40
Final score84
03 / Policy outcomeOne score. One clear next step.
Illustrative decision bands01A · BApprove
02C · DConditional
03EManager Review
04FDecline
Hard-stop protection stays in control.Deal-killers such as a sub-floor score, active bankruptcy, or capacity beyond policy can override the tier outright.
100base score
5risk dimensions
25+weighted factors
11hard auto-declines
0protected-class inputs
The 10-Deal Second Opinion
Bring 10 deals. See what your own book has been telling you.
Send a small, anonymized mix — a few that funded well, a few that went bad, a few you declined or debated. We'll score every one, explain the call, and show where approvals are walking and where losses are hiding. No integration, no obligation.
Start my deal reviewWhat you'll see
Where approvals are walking, where loss exposure is hiding, and which marginal files a cleaner structure could have funded — on your own deals.
Why you can trust it
The score is deterministic and benchmarked against your funded book. No magic, no model you can't open up and inspect.
How fast
Days, not a quarter. The first review needs no integration and no IT project.
What it costs you
Send a handful of anonymized files, or type them in. Nothing in your core system changes.
Exactly what comes back
Start with 10 applications. Get a deal-by-deal read and the risk patterns your team can act on.
What you get back
A score, tier, and decision for every file
Top risk drivers and compensating factors on each deal
Rescue Path: the structure change that would have funded a declined file
Hidden Exposure: approved files that still stack risk
Point movement from more down, lower LTV, shorter term, or lower PTI
Adverse-action reason support for every decline
A PDF-ready underwriting summary
Portfolio calibration notes across the set
What the output looks like
Move from gut feel to a clear deal-by-deal underwriting answer.
Illustrative dealExample only
- FICO558
- Down payment$750
- LTV119%
- Term72 months
- Vehicle8 yrs · 104k mi
- PTIelevated
Scorecard trail
- Base score100
- FICO 558−14
- Thin credit depth−6
- LTV 119%−10
- PTI elevated−8
- Vehicle 8 yr · 104k mi−7
- Thin cash down $750−3
- Steady employment+6
- Stable industry+4
- Score62 · Tier D
62score
Tier DConditional
Top risk drivers
- High LTV stacked on aging collateral
- Payment-to-income running hot
- Thin down payment behind a sub-560 score
Suggested stipulations
- Verify income with recent stubs
- Proof of residence
- Proof of down payment
Biggest single lever+$1,250 down → +11 pts → into Tier C approval range
What's inside
The factors, thresholds, risk notes, and dealer context behind each underwriting decision.
One standard on every file
A base score plus weighted factors roll up to an A–F tier and a clear decision. Same inputs, same answer — whoever is at the desk.
Compensating factors, made explicit
Real money down behind a thin score earns points on the record, not in a hallway conversation that disappears by funding.
Toxic combinations caught early
Penalties and hard auto-declines flag the dangerous stacks — high LTV, aging collateral, stretched capacity, recent credit stress — before they fund.
Adjust the deal, watch the score move
Change cash down, term, LTV, or payment and the score recomputes live, so you can find the structure that makes a borderline file fundable.
Documentation that defends the call
A factor-by-factor trail, saved run history, and exportable score and analysis records that hold up in an audit conversation.
Dealer context, in the file
Read the application beside the dealer relationship that sent it, so deal quality and source quality get weighed together.
After the score
Two reads on every deal: which to rescue, which to flag, which to fund.
One rules-based, one generative, both advisory. Your underwriter still makes every call.
Rules-based
Deal IntelligenceDeterministic and traceable. Built from your policy, the deal math, and your own funded book.
Rescue PathSave the deal.
The one change — more down, lower LTV, shorter term — that funds a decline. +$1,250 down → Tier-C approval.Hidden ExposureCatch the risk.
Flags files that pass on paper but stack risk — before “approved” becomes a charge-off.Profit CheckApprove for profit.
Ranks each option by your book’s real default rate, loss exposure, and net return.
Generative
AI AnalysisAn experienced underwriter’s write-up on every file — same format, every time.
Quick verdictBottom line first.
Then the risk drivers and what’s working behind it.Stipulations & deal-workTied to this file.
Conditions matched to the real risk — never boilerplate.It never decidesYou keep the call.
Explains the deal; can’t touch the score, tier, or decision.
The receipts
Don't take the scorecard's word for it. Backtest it on every loan you’ve ever funded.
One click in the portal replays your entire funded book through your credit policy — thousands of loans in about a minute — and grades every call against how each loan actually performed. Not a consultant's sample. Not a slide. Your book, your policy, the receipts.
Every loan is scored with day-one data only — the bureau pull as it stood, the deal structure, the application date. The replay never sees the outcome, so it can't grade itself on the answer key.
A default at month 30 already paid for itself. A default at month 8 is a crater. Every loan weighs cash collected against cash advanced, so the report reads like your P&L — not like a model metric.
Click a tier and the actual loans behind it line up, worst first — app number, months on the books, cash in, cash out. Aggregates you can interrogate, not take on faith.
The mirror clause
Sometimes the backtest will tell you your current policy barely separates risk at all. That's not a failure of the tool — that's the most valuable sentence it can say. From that day forward, every threshold you tune is measured against your own history instead of argued in a conference room.
The Policy Backtest is part of the Underwriting Engine — not a separate product. It runs on the same live, read-only LOS + LMS join as the rest of the platform, so once the join is up there is nothing extra to build.
Compliance & audit posture
The real worry isn't speed. It's risk. So we built for that.
The strongest control is the operating model itself: deterministic scorecard first, AI advisory only, your underwriters in control, and a file-level decision trail anyone can inspect.
This supports your compliance and audit readiness. It does not guarantee legal compliance — your lending team and counsel remain responsible for policy and notices.
Deterministic scorecard you control
Lender-owned credit policy
AI advisory only — never decisions
No protected-class inputs
Factor-by-factor decision trail
Adverse-action reason support
Saved run history
Human underwriter keeps authority
Built for the whole credit operation
One underwriting standard for growth, risk control, compliance, and operations.
Grow approvals without loosening the book, and put a measurable floor under loss exposure — without betting the company on a black box.
One standard across every desk, faster file reviews, a shorter ramp for new hires, and a clean justification for every manager override.
Deterministic, inspectable factors, no protected-class inputs, adverse-action reason support, and a decision trail you can hand to an examiner.
Less second-guessing after a loss, fewer rework loops, and a consistent risk vocabulary the whole credit team works from.
Where the ROI comes from
A few cleaner calls each month can change the math.
Two levers move the number: good deals you save with a better structure, and bad deals you catch before they fund. Put your own figures in.
Illustrative annual impact
$188,400
Annual estimate = $15,700 monthly impact x 12 months
Recovered approvals / month$7,200
Losses avoided / month$8,500
Monthly impact$15,700Annualized impact$15,700 x 12 = $188,400
Illustrative only. Actual results depend on your book, policy, pricing, deal mix, and servicing.
A low-risk path to rollout
No leap of faith: review, backtest, shadow, then roll out.
Anonymized files, no integration. We score them, explain each call, and walk your team through what we find.
Once your read-only join is live, one click replays every funded loan you've ever written through the scorecard — and grades every call against how each loan actually performed.
Score live deals beside your underwriters without changing a single decision, until the standard earns their trust.
Turn it on by desk, dealer, or credit tier — only after the workflow has proven itself on real files.
Straight answers
The questions a careful lender actually asks.
Is the AI making our credit decisions?
No — and that's the point. The score, tier, and decision come from a deterministic point system you control. The AI is advisory only: it explains risk, suggests stipulations, and drafts adverse-action reasons for your team to review. It cannot change the number or approve a deal.
Will this replace our underwriters?
The opposite. It captures the judgment your best underwriters already use and makes it repeatable for everyone else. Your team keeps final authority on every file; the engine just makes sure the read is consistent and documented.
Compliance is going to ask hard questions.
Good — it was built for them. Every factor, threshold, and rule is visible and lender-owned. There are no protected-class inputs, the AI never decides, and each file carries a factor-by-factor trail and adverse-action reason support. It supports your compliance posture; your team and counsel still own policy and notices.
We don't want to hand our policy to a black box.
Then you'll like this. Nothing is hidden. You can read and tune every weight, bonus, penalty, and auto-decline rule, and see exactly why any deal landed where it did. The model doesn't drift quietly — you change it on purpose.
Do we have to integrate anything to start?
No. The first review runs on anonymized files or manually entered deals. Integration into your system of record is a later step — and only if the scorecard has already proven useful.
Our team already knows how to underwrite.
Exactly why this works. We don't replace that knowledge — we encode it. Your policy becomes a standard every desk applies the same way, so the gap between your most and least experienced underwriter stops being a line item in your losses.
See it on your own book
Score your last 10 deals.
Send a handful of anonymized files and we'll show how your best underwriting judgment can become a repeatable standard: the score, risk drivers, structure options, and adverse-action support on deals you already know.
No integration to startAnonymized files welcomeAI advisory onlyYou keep the final call
The engine answers "would we fund the submitted deal?" When you're ready for "what fundable deal should we offer instead?", that's Auto-Structure, the counteroffer layer built on this engine.