New startup ideas · Fintech · Fintech
startup idea
Creditome
The foundation model for small-business credit, sold as underwriting software.
Creditome trains a transaction-native foundation model on bank, payments and accounting data, then sells a self-serve SaaS underwriting workbench to small lenders, credit unions and fintech lenders worldwide.
- Software subscription
- Small business
- $1-10B market
- Platform others build on
- Global from day one
3/5
venture judge
27
similar startups, last 2 years (42 all-time)
95%
of 3 nearest real companies still alive
yes
8 matching federal grants and programs
Direction supported by government programs and grants
Test it before you build it
$1,000 · 5 weeks · 15 prospects
For $1,000 in 5 weeks, prove 4 of 15 US small-business lenders will sign a data-sharing agreement and 2 will pay $2,000 up front for a cashflow-model backtest against their own loan book.
Riskiest assumption · US small-business lenders will feed their loan-level performance and transaction data into a third-party model vendor's corpus and pay for the resulting lift, rather than guarding the data as their moat - the exact refusal the card names as the thesis killer.
1Focus group: who and where
The chief credit officer or head of SMB lending at a US fintech working-capital lender, an active SBA 7(a) lender, or a credit union with over $50M in commercial loans, whose current scorecard leans on FICO and time in business and who is watching approval rates fall on young businesses with thin bureau files this quarter
where to find 15 · The SBA's weekly 7(a) lender activity report and an NCUA call-report screen for credit unions with commercial loan books (lists); deBanked's newsletter and community of alternative small-business lenders (community); the LEND360 conference for online lending this fall (event); America's Credit Unions lending council network (channel)
2Sell first, build later
A paid backtest sold before the foundation model exists: the lender shares 24 months of anonymized loan tapes and transaction summaries under NDA and DSA, and within 30 days receives a cashflow-based limit model scored head to head against their current cutoffs, reported as approval lift at constant loss
the ask · $4,000 per backtest, $2,000 invoiced up front; the workbench LOI locks founding pricing at $1.50 per decision with a $500 monthly minimum
a real yes · A real yes is a signed DSA plus $2,000 moved by ACH; 'our data science team would love to see the methodology', unpaid data-sharing offers and requests for a free proof of concept are noes
3Small experiments
The first one attacks the riskiest assumption; each ends with a number that says whether to run the next.
1. Data-sharing agreement asks
$350 · 14 days
List the top SBA 7(a) lenders by volume, the NCUA screen and deBanked-covered fintech lenders, then call 15 chief credit officers with one offer: a paid backtest where they share 24 months of anonymized loan tapes with bank-transaction summaries under a mutual NDA and a one-page data-sharing agreement built on Common Paper templates. The ask on every call is a signature, not an opinion.
keep going if · 4 of 15 lenders sign the NDA and data-sharing agreement in writing
2. Collect the up-front half
$250 · 12 days
Invoice each signed lender $2,000 by ACH, half of the $4,000 backtest fee, due before any data transfer, with the lift report promised within 30 days of data receipt. Count only cleared payments; a signed DSA with an unpaid invoice is a stall, not a yes.
keep going if · 2 of 4 signed lenders pay $2,000 within 14 days of signature
3. Founding per-decision price probe
$400 · 12 days
Show all 15 lenders a Figma mock of the workbench plus a one-page price sheet at $1.50 per decision with a $500 monthly minimum, and ask backtest buyers to lock it in a one-page LOI as founding pricing for year one. This tests whether the platform pricing survives contact with a credit committee, not just the one-off pilot.
keep going if · 3 of 15 lenders sign the LOI locking $1.50 per decision
4Collect a deposit up front
Tesla took $1,000 refundable reservations for the Model 3 and $100 for the Cybertruck before building either: the deposit is the measurement, not the revenue.
$2,000
per prospect, refundable
how · A paid design-partner backtest with 50% invoiced up front by ACH once the NDA and data-sharing agreement are signed - the normal way a credit union or fintech lender pays a small vendor before full procurement - with founding per-decision pricing locked in a one-page LOI set up: Stripe Invoicing ↗
what it reserves · A slot in the five-lender September backtest cohort and the founding price of $1.50 per decision for the first year of the workbench
refund · The full $2,000 is refunded if the lift report is not delivered within 30 days of receiving the lender's data
target · 2 paid backtests and 4 signed DSAs from 15 conversations within 35 days
before taking money · Selling underwriting scores can make the company a consumer reporting agency under FCRA and puts every model under ECOA fair-lending scrutiny, so run backtests only on anonymized historical data under a signed DSA and get counsel before any live decisioning.
Go: build it if
At least 4 of 15 lenders sign the DSA, at least 2 pay $2,000 up front, and at least 3 sign the $1.50 per decision LOI
Kill: stop if
1 or fewer of 15 lenders will sign a DSA, or 4 DSAs signed but zero invoices paid within 14 days - the data refusal the card warns about, confirmed at the smallest scale
5 Scripts to run itoutreach message, landing copy, deposit terms · click to open
outreach message
Your SBA and working-capital book prices small-business risk on FICO and time in business while the borrower's bank transactions sit unused. I am running paid backtests with 15 lenders: you share 24 months of anonymized loan tapes with transaction summaries under NDA, and in 30 days you get a cashflow-based limit model scored head to head against your current cutoffs - approval lift at constant loss. $4,000, half up front, refunded if I miss the deadline. Open to 20 minutes this week?
landing page
See the approval lift hiding in your borrowers' bank data $4,000 backtest, $2,000 up front: your 24-month loan tape scored against a cashflow limit model, lift report in 30 days Reserve one of five September backtest slots
deposit terms
The $2,000 payment is half of the $4,000 backtest fee, invoiced by ACH after both sides sign the NDA and data-sharing agreement, and it reserves one of five September backtest slots plus a locked founding price of $1.50 per decision for your first year on the workbench. If your lift report is not delivered within 30 days of us receiving your data, the full $2,000 is refunded. The balance is due only on delivery.
Would you run this test?
One tap. The yes-share feeds the Demand pillar of this idea's score; nobody sees who answered.
Budgets are out-of-pocket estimates for a team of one to three, US market. Size the deposit to the deal, and check the terms before taking money in a regulated line.
Scorecard
One score that balances how trendy the idea is, the demand for it and its potential for 100x, with competition measured relative to every other idea in the catalog. Recent startup trends first, government priorities second.
62
Idea Score, 0-100 · raw 38.6 x 1.61
Crowded
competition: more crowded than 89% of ideas · headwind x0.56
+5.0
government priorities, secondary (510 matching grants)
Trend
65
Is the wave forming now? 2025-26 entrants vs 2023-24, rounds since 2025, the sector's live-batch direction, the 2026 trend analyst.
- Entrants 2025-26 vs 2023-24 (similar companies)99
- Rounds announced 2025+ in the sector87
- Sector direction (live batch)50
- 2026 trend analyst25
Demand
52
Does anyone want it? YC's current RFS, companies already paid for something similar, the operator judge, founders' yes-rate in decks, readers who would run the test.
- YC asks for it (current RFS: idea / sector)30
- Someone already pays (similar companies, recent / all-time)100
- Operator judge: real pain25
100x potential
66
Can it return a fund? The venture judge (double weight), market-size and moat axes, neighbours still alive, the technologist judge.
- Venture judge50
- Market size axis33
- Moat axis80
- Neighbours still alive81
- Technologist judge100
Score = 100 x cbrt(Trend x Demand x 100x) x (1 - 0.5 x crowding) + government bonus (max 5), calibrated so the 95th-percentile idea scores 90 (order never changes). A geometric mean: a weak pillar cannot be papered over. Percentiles are among the 272 ideas in the catalog; the terms matched were foundation, model, small-business, credit, sold, underwriting, trains, transaction-native.
The idea in full
- What
- Creditome trains a transaction-native foundation model on bank, payments and accounting data, then sells a self-serve SaaS underwriting workbench to small lenders, credit unions and fintech lenders worldwide. Lenders and third-party developers build scorecards, pricing tools and collections logic on top of the model inside the platform, so it becomes the substrate for small-business lending rather than one more score. The road is R&D first: multi-year training runs on owned GPU clusters and licensed credit-reference-agency status in key markets before meaningful revenue.
- Why now
- Prodigy Research (yc S26) is training a foundation model for quantitative finance and zypl.ai is selling synthetic-data credit models to financial institutions, yet only 7 of the 217 fintechs from the last two years even pitch to financial institutions - the model layer for credit is being built now and is not yet crowded.
- Wedge: first customer and entry point
- Start with cashflow-based limit-setting for fintech lenders in two or three emerging markets where bureau data is thin, delivered as a self-serve workbench with usage-based pricing.
- Path to 100x
- Credit decisioning software is a $1-10B market, but the mechanism is platform: if lenders standardize on one credit foundation model the way developers standardized on a handful of LLMs, the winner prices per decision across millions of small-business loans globally. Licensed bureau status in each market compounds into a moat that a model alone never has.
- Ceiling
- If the product stays a decisioning tool and never expands into pricing or capital markets, the software TAM itself caps the outcome near the low billions.
- Closest real companies, as the generator saw them
- zypl.ai generates synthetic data to patch existing bureau models; Pave sells cashflow analytics as a feature. Creditome replaces the model layer itself and takes on bureau licensing, which neither does.
- Main risk
- Incumbent bureaus and large lenders refuse to feed data into a shared model, starving the pretraining corpus that the whole thesis depends on.
Five judges
Each judge scores every idea in the catalog with a named rubric; the venture judge decides whether a card is shown at all (4-5 is venture-grade).
Venture investor
3/5
Platform upside is real, but bureaus refusing to feed the pretraining corpus starves the one asset the entire thesis rests on.
Bootstrapper
1/5
Owned GPU clusters, multi-year training runs and bureau licensing before meaningful revenue is a venture-scale capital problem.
Operator
2/5
Multi-year GPU training before revenue, and the card concedes incumbent bureaus and large lenders will not feed the shared pretraining corpus.
Technologist
5/5
A transaction-native foundation model on owned GPU clusters plus bureau licensing is exactly the deep, compounding stack thin wrappers cannot fake.
Risk
2/5
The entire thesis needs incumbent bureaus and lenders to feed a shared corpus, plus bureau licensing, before any revenue in three years.
trends
2/5
Foundation model for credit is a 2023-vintage thesis wearing a 2026 label, with multi-year training runs and bureau licensing pushing revenue past the claimed window.
Similar startups in the directory
Companies whose pitch matches most of the idea's terms (foundation, model, small-business, credit, sold, underwriting, trains, transaction-native): 42 all-time, 27 from the last two years. Same matching as Idea Check.
Foundation models on enterprise transaction data.
Robots for high skilled labor powering AI infrastructure
Robots for vehicle maintenance.
Titan Foundry provides secure, banking-specific AI models for operations, compliance, and credit analysis.
Othello builds AI models for B2B sales calls, deploying agents to give real-time prompts to sellers.
Making chemicals 100,000x faster
Scale AI scientists whose own breakthroughs accelerate the next.
Speech foundation models that understand people: words, tone of voice, emphasis, intent, and emotion.
Data & Deployment Infrastructure for Physical AI.
AI-native pharma company
Training the world's best AI model for quantitative finance.
MovieBall is an AI-native production OS helping for studios and creators — accelerating the content pipeline and turning gut-feel intuition into data-driven decisions.
The generator's reference companies
Real companies the model named as closest when it wrote the card, with their fate. A check mark is a company the radar could verify in its directory.
Public money in this direction
US federal grants, SBIR/STTR awards and open opportunities from the radar's public-money feed, matched to the idea's terms; the sector totals give the context.
8
grants and programs matching the idea
2
startup-relevant grants in Fintech
$809K
awarded in the sector, tracked
- Development of a milk analyzing technology for the detection of antibiotic contamination using fluorescence spectroscopy and artificial intelligence.awardhigh relevance
NIH / FDA · SBIR phase I · $200K · posted 2026-09-30
- Scalable Free-Text Annotation with Validation: Building Explainable Vision-Language Models in Dermatologyawardhigh relevance
NIH / NLM · STTR phase I · $313K · posted 2026-09-16
- SBIR Fast-Track: Scalable Robotic Skill Learning via Generative Embodied Twins and Vision-Language-Action Models for Task-Oriented Autonomyawardhigh relevance
National Science Foundation · SBIR Fast-Track · $2M · posted 2026-09-09
National Science Foundation · SBIR Phase II · $1M · posted 2026-09-03
NIH / NIGMS · SBIR phase II · $979K · posted 2026-09-01
NIH / NHLBI · SBIR phase I · $290K · posted 2026-09-01
- SBIR Phase II: Leveraging Sequencing to Identify and Predict Multidrug Resistanceawardhigh relevance
National Science Foundation · SBIR Phase II · $1M · posted 2026-08-24
- Implantable Intelligent Engineered System to Enable Virtual Bidirectional Muscle-Prosthesis Mechanical Interaction for Amputeesawardhigh relevance
National Science Foundation · M3X - Mind, Machine, and Motor, Dynamics, Control and System D, Special Initiatives · $809K · posted 2026-08-21
Market signal
What the radar sees in Fintech: new companies by cohort year, the forming YC batch, and outcomes since the February snapshot.
Fintech · 256 → 149 → 182 → 122 → 90 new companies 2022 → 2026 · 89% aliveYC F26: 7 in this cluster, 5% of the batch (was 5% in S26) (W27 is still forming: 5 listed)Since February, of 584 YC companies here: 13 acquired, 9 shut down, 57 rewrote their pitch
Design attributes
The card is one cell of a designed set: every axis below was chosen before the text was written, and the text had to realize it.
- Buyer
- Small business
- Business model
- Software subscription
- Path to 100x
- Platform others build on
- Market size
- $1-10B market
- Capital intensity
- Capital-heavy (hardware, bio, infra)
- Speed to revenue
- R&D first, revenue after 3 years
- Technical depth
- Deep tech: ML, hardware, bio
- Go-to-market
- Self-serve
- Moat
- License or regulatory moat
- Geography
- Global from day one
- Regulation
- Some regulation
- Vibe
- Hot space
Listed under
An idea sits in its own sector and in any sector its text clearly touches.
More ideas like this
Fintech · Fintech
Bookzero
Software that closes a small company's books for a tenth of a bookkeeper's price.
Bookzero is self-serve subscription software that connects to a small business's bank feeds, invoices and payroll and produces a finished monthly close - categorized ledger, reconciliations, financial statements - for roughly $100 a month against the $800-1,500 SMBs pay outsourced bookkeepers.
Fintech · Fintech
Parbook
The self-serve secondary market where institutions trade private credit loan participations.
Parbook is a marketplace where private credit funds, insurers and regional banks list, price and settle loan participations and portfolio slices, using a standardized machine-readable data tape that Parbook generates from each seller's loan documents.
Fintech · Fintech
Perilex
An exchange where machine-priced corporate risk is listed and capital bids.
Perilex is a listing venue for idiosyncratic corporate risk that traditional carriers price badly: fleet and telematics-driven liability, compute and outage exposure, model-failure liability.
Fintech · Fintech
Countersign
Spend controls and audit trail for every dollar an enterprise's AI agents move.
Countersign sits between a company's agents and whatever rails they spend on - issued cards, wallets, stablecoin accounts - enforcing per-agent budgets, merchant allowlists and human approval thresholds before authorization, then reconciling each settled charge back to the agent run and prompt that caused it.
Fintech · Fintech
Underwatt
Underwriting software that prices AI data center risk from live power and thermal telemetry.
Underwatt sells a subscription platform to carriers, MGAs and reinsurers that prices property, business-interruption and outage risk on GPU data centers using a telemetry fleet Underwatt installs itself: power quality recorders, thermal and coolant sensors, and grid interconnect feeds at the insured sites.
Fintech · Fintech
Subrix
AI agents that settle claims between insurance carriers on one shared network.
Subrix runs autonomous agents that handle subrogation - the process where one insurer recovers claim costs from another at-fault party's insurer.
Fictional company written 2026-08-26 from MarkosWeb data; the companies, grants and numbers around it are real and tracked. Treat the idea as a research prompt, not a plan.