M2P Fintech
Fintech is evolving every day. That's why you need our newsletter! Get the latest fintech news, views, insights, directly to your inbox every fortnight for FREE!
US lenders are being asked to move faster while the ground shifts under them. The FDIC's Small Business Lending Survey found that three in four banks approve their typical small business loan within ten business days. Only 3% of banks fully automate the process, and only for very small loans. The survey data dates from 2022, but it shows how much of the work still runs through people and queues.
Three pressures now test that model. Underwriting is still largely manual. Credit conditions are moving differently by product. And the bank data lenders rely on has started to carry a price. This post covers each one and how an origination platform can be built to handle them.
Manual underwriting still sets the pace
In many institutions, the application is digital but underwriting is not. Analysts read bank statements, spread financials, check documents and write credit memos by hand.
The FDIC's survey puts a number on it. Only one in ten banks has a credit-scoring system that can partially or fully automate underwriting for some non-credit-card lending.
Banks that have started automating describe where the effort sat. In July 2026, ConnectOne Bank reported early results from AI tools in commercial lending. Document search time fell by 98%, from about 20 minutes to around 30 seconds, and one custom agent cut the time spent updating relationship records by 60%. These are bank-reported results, but they show how much routine work sits inside a commercial lending day.
The credit cycle is moving product by product
US credit is not getting better or worse as one block. The New York Fed's Q2 2026 household debt report shows products diverging. Overall delinquency improved slightly to 4.7% of outstanding debt, but the flow into serious delinquency for mortgages rose from 1.29% to 1.52% year over year, while credit cards and auto stayed roughly flat. The picture earlier in the year was similar. The Q1 report had mortgage flow into serious delinquency at 1.48% against 1.22% a year earlier, and HELOCs rising from 0.88% to 1.15%. Student loans were harder to read, because the New York Fed noted distortions from the re-reporting of defaulted student debt.
Banks are responding product by product too. In the Fed's July 2026 Senior Loan Officer Survey, a modest net share of banks tightened credit card standards in the second quarter, while standards for auto and other consumer loans were basically unchanged. Compared with July 2025, banks reported easier standards for all categories except consumer loans.
A lender in this environment may need to tighten one portfolio and hold another in the same quarter. That only works if policy can be changed quickly and safely.
Bank data now has a price
For years, aggregators pulled consumer account data from banks at no charge. That is changing. In June 2025, JPMorgan Chase told its largest data aggregators it would start charging for access, citing 1.89 billion data requests in a single month. By November it had updated contracts with the major aggregators covering more than 95% of its open banking data requests. Fees reportedly differ by use case, and overall cost grows with data volume. PYMNTS reporting noted that fees tied to account verification or credit underwriting are likely to be passed on to fintechs and potentially their customers.
Regulation hasn't settled the question. The CFPB finalized its Section 1033 rule in October 2024, but a federal court has enjoined enforcement and the CFPB is rewriting it, including whether banks can charge for access. Until that is resolved, access terms are being set by private contracts.
For lenders, this means the cost and availability of cash-flow data can change. A platform that ties underwriting to one data source makes every such change expensive.
What the three pressures have in common
Manual underwriting, shifting credit conditions and changing data costs look unrelated, but they ask the same thing of an origination system:
Take the manual work out of the workflow, so volume doesn't depend on headcount.
Let Risk change policy without a release, so product-level shifts can be acted on within the quarter.
Keep data sources flexible, so a change in price or access doesn't mean a rebuild.
How M2P's Loan Origination System supports these needs
M2P's AI-orchestrated Loan Origination System runs digital, branch, embedded and partner-led origination on one platform. A configurable Business Rules Engine (BRE) handles credit decisioning, supported by AI agents for documents, fraud, financial analysis, scoring and credit appraisal memos (CAMs). Here is how that maps to the three pressures.
What US lenders are up against | How the M2P LOS supports it |
|---|---|
Manual document and financial analysis | The Document Intelligence Agent classifies 100+ document types with 95%+ OCR accuracy and detects tampering. The Bank Statement Analyser Agent parses statements from 50+ banks, with ABB computation, cash-flow categorisation and bounce detection. The Financial Analysis Agent extracts P&L, balance sheet and cash flow data and calculates 50+ ratios. The CAM Generation Agent produces credit memos from structured templates. |
Credit conditions that differ by product | The BRE supports boolean, financial, gating and weightage-based rules, along with risk-based pricing, custom scorecards and explainable decision outputs. Rules are versioned, auditable and back-tested against historical portfolio data, so policy is enforced consistently across products. |
Exceptions and oversight | Cases that don't go straight through route to the Case Manager, with configurable allocation, multi-level approvals, SLA and turnaround monitoring, and full audit trails. The Policy Compliance Agent validates cases against internal credit policy and flags deviations. |
Dependence on specific data sources | 75+ pre-integrated APIs cover KYC, banking, bureau, documents, payments and fraud services. Multi-bureau connectivity includes soft-pull support for pre-qualification. [Product team to confirm US bureau, KYC and bank-data integrations.] |
Fragmented channels | Digital, branch, embedded and agent-led origination run through one LOS, with partner onboarding and journey customisation. It supports 15+ loan types and hands off to the loan management system with multi-tranche disbursal and co-lending settlement. |
The principle is that AI prepares the work and the business owns the decision. The agents classify documents, parse statements and flag anomalies, while approvals, policies and accountability stay with the business. Role-based dashboards and 100+ configurable MIS reports give Risk, Ops, Sales and Finance one view of decisions, deviations and turnaround times.
Questions worth settling before your next platform decision
Who owns the rule? If Risk wanted to tighten one product and loosen another next quarter, could it do that without an Engineering release?
What happens when data changes? If a data source's price or availability shifted, would it be a configuration change or an integration rebuild?
Where is the effort going? Of your manual underwriting steps, which could move to document, income and policy automation first?
Can you test before you change? Can you back-test a policy change against your own portfolio before it goes live?
Is the trail there? Can you show who changed which rule, when, and what it did to approvals?
Where this is heading
Neither the credit cycle nor the data rules are settled. The CFPB is still rewriting its open banking rule, and bank data access terms are being worked out contract by contract. Credit conditions have shifted from quarter to quarter, with different products moving in different directions.
For lenders, the practical response is an origination platform where policy, data sources and workflows can be configured rather than rebuilt.
Looking to modernise your US lending operations? Talk to M2P about how our Loan Origination System can bring faster, more flexible and auditable origination to your business. Get in touch →