YOUR LOAN MANAGEMENT SYSTEM IS ABOUT TO GET A NEW EMPLOYEE. AND IT DOESN'T NEED A DESK.
AI is moving from answering questions about loan files to performing actual work inside lending systems. Payday, title, and installment lenders need to pay attention.
I've spent a long time in the business of lending money to strangers.
Long enough to know that lenders don't have a shortage of software.
We've got loan management systems. Customer relationship management systems. Underwriting engines. Payment processors. Collections platforms. Reporting dashboards. Document management systems.
Plenty of technology.
And somehow, plenty of employees still spending their days moving information from one place to another.
Opening files. Checking documents. Creating tasks. Calling customers. Updating notes. Chasing missing information. Asking managers for decisions.
Then doing most of it again tomorrow.
Here's what caught my attention this week: AI is beginning to do more than explain what is inside a loan file.
It's beginning to take action.
WHAT nCINO ANNOUNCED ON OCTOBER 12, 2026
On October 12, 2026, nCino announced a new Loan Officer Productivity Suite designed to bring additional artificial intelligence capabilities into its mortgage point-of-sale platform.
The announcement described six AI-powered capabilities.
- Conversational assistance for loan officers.
- Loan eligibility evaluation.
- Automated underwriting summaries.
- Preapproval workflow automation.
- Document and disclosure workflow support.
- Connections allowing AI agents to interact with the mortgage platform.
One capability is particularly interesting.
Loan officers can ask questions about their loan files and receive answers grounded in their institution's guidelines.
That's useful.
But the more important development is that the system can initiate certain actions, including creating tasks and sending reminders, subject to loan-officer approval.
The announcement also describes AI agents connecting to the platform through a Model Context Protocol, or MCP, interface.
Think of MCP as one way to connect AI systems with the tools and information they are authorized to use.
An AI assistant that reads a loan file and tells your employee what is missing is useful.
An AI agent that identifies the missing document, creates the follow-up task, prepares the customer reminder, and updates the workflow is something else.
That's not just answering questions. That's helping perform the work.
nCino also says its automated preapproval workflow can compress certain traditional processes into minutes.
Those are vendor-reported capabilities and performance claims, not independently verified results for every lender.
And let's be clear: this announcement concerns a mortgage lending platform, not a turnkey AI solution for payday, title, or installment lenders.
I'm not suggesting a storefront lender can buy the nCino product tomorrow and plug it into its existing loan management system.
I'm suggesting something more important.
The technology direction is becoming difficult to ignore.
WHAT THIS MEANS FOR PAYDAY, TITLE, AND INSTALLMENT LENDERS
Forget mortgages for a minute.
Think about your own stores.
Think about your CSRs, loan officers, store managers, district managers, underwriters, and collectors.
How much of their work involves making difficult decisions?
And how much involves collecting information, checking rules, sending reminders, updating systems, and chasing exceptions?
The second category is where I see an immediate opportunity.
Not replacing judgment.
Not allowing an AI system to make uncontrolled lending decisions.
Removing routine administrative work so your people can concentrate on borrowers, decisions, and dollars.
FIVE AI WORKFLOWS I'D INVESTIGATE FIRST
1. THE INCOMPLETE APPLICATION CHASER
A customer starts an application.
Income information is missing. A bank statement hasn't arrived. Identification needs review. The customer stops responding.
Today, someone in your store may have to open the file, identify the problem, create a follow-up, and contact the customer.
Tomorrow, an authorized AI workflow could identify missing information, prepare an appropriate reminder, create a task, and escalate the application if it remains incomplete.
Customer communications would still require the proper consent, controls, and compliance-approved language.
What I'd measure: Application completion rate, time to completed file, funded loans per 100 applications, and employee follow-up hours.
2. THE UNDERWRITING EXCEPTION WATCHDOG
Here's one I especially like.
A loan application arrives.
The requested amount exceeds the normal policy limit for the borrower's documented income or other underwriting characteristics.
The loan officer requests an exception.
The system documents the reason, routes it to the correct authority, records the decision, and makes the exception available for subsequent performance analysis.
Now multiply that across 40 stores.
Your question should not simply be:
"How many exceptions did we approve?"
It should be:
"Which exceptions actually make us money, and which ones are creating first-payment defaults?"
AI can help organize and flag the information. Your established underwriting policy, approval controls, and responsible employees still govern the actual decision.
What I'd measure: Override frequency, exception outcomes, first-payment default rates, 30-day delinquency, and eventual charge-offs.
3. THE FIRST-PAYMENT DEFAULT EARLY WARNING SYSTEM
Any operator who has spent enough time in subprime lending knows the significance of the first payment.
Especially for new customers.
Imagine an AI workflow that evaluates available, permissible account signals and flags accounts requiring early attention.
It might identify missing payment arrangements, unusual servicing exceptions, or accounts approaching a critical first-payment date.
It could then prepare an approved servicing task for an employee.
That's different from waiting for the account to become delinquent and hoping somebody notices.
Any predictive model would need validation, monitoring, and appropriate consumer protection safeguards before being relied upon.
What I'd measure: New-customer first-payment default, repeat-customer first-payment default, payment completion rates, cure rates, and net credit losses.
4. THE COLLECTIONS TASK COORDINATOR
Collections departments don't need another dashboard that simply tells them borrowers owe money.
They already know that.
They need help identifying the next permitted, useful action.
Consider a system that reviews account status, authorized contact preferences, applicable restrictions, payment arrangements, and previous collection activity.
It could recommend a compliant next task, prepare an approved communication for review, or identify accounts needing supervisor attention.
The goal isn't more collection attempts for the sake of making more attempts.
It's better-directed activity.
Federal and state collection laws, communication restrictions, consent requirements, and applicable licensing rules must control the process.
What I'd measure: Dollars collected per employee-hour, promise-to-pay kept rate, cure rate, right-party contact effectiveness, and complaints.
5. THE DISTRICT MANAGER'S MORNING EXCEPTION BRIEF
This one could be especially valuable for multi-store lenders.
Every morning, the district manager receives a concise summary of the stores requiring attention.
Not 15 reports.
Not another spreadsheet with 40 tabs.
A useful operating brief.
Store TX-07 approved 68% of new applications yesterday, compared with a company average of 54%.
Its recent 30-day first-payment default rate is 41% above the company average.
Seven policy exceptions and three manager overrides were recorded in the review period.
Recommended action: Review new-customer underwriting, exception approvals, and employee-level loan performance at TX-07.
That's a hypothetical example, not actual store performance data.
But that's what an effective daily operating report should accomplish.
Tell management where to look, why it matters, and what needs to happen next.
What I'd measure: Time to identify exceptions, corrective-action completion, store-level variance, delinquency trends, and net operating income.
LET'S PUT SOME NUMBERS ON THIS
Suppose you operate 40 storefront lending locations.
Each location has employees spending time on routine administrative activity.
What happens if workflow automation saves just 30 minutes per store, per business day?
ILLUSTRATIVE LABOR-CAPACITY CALCULATION
40 stores
× 0.5 hour saved per day
× 22 operating days per month
= 440 employee-hours per month
× $25 estimated loaded labor cost per hour
= $11,000 per month
× 12 months
In potential labor capacity at those assumptions.
That's not necessarily $132,000 in cash savings.
If you don't reduce overtime, avoid incremental hiring, or eliminate actual costs, the benefit is additional productive capacity rather than a direct reduction in operating expenses.
But what happens when your existing employees spend those recovered hours completing applications, serving customers, following up on legitimate leads, and resolving problems?
Now the opportunity extends beyond labor productivity.
WHAT IF IT ALSO IMPROVED LOAN PERFORMANCE?
Let's extend the example.
Suppose 40 stores each fund 100 loans per month.
That's 4,000 loans across the company every month.
Assume an average funded principal of $1,100.
That's $4.4 million in monthly originations.
Now suppose better controls, earlier intervention, and more consistent underwriting reduce realized credit losses by an amount equivalent to 0.3% of funded principal.
That's $13,200 per month.
Or $158,400 annually, assuming that improvement is sustained and the loan volume remains constant.
Combine the two illustrative opportunities:
Potential annual labor capacity value:
$132,000
Potential annual credit-loss improvement:
$158,400
Illustrative annual value before software, integration, implementation, oversight, and ongoing operating costs.
Let me emphasize something.
Those are assumptions, not a vendor ROI claim or a prediction.
The labor benefit and credit-loss benefit are different types of value and should not automatically be treated as realized net operating income.
You would need actual before-and-after data, validated loss measurements, implementation costs, and proof of realized labor or revenue benefits.
But the exercise illustrates why I believe lenders should start evaluating AI at the workflow level.
Even modest improvements can become meaningful when multiplied across a store network.
THE COMPLIANCE PROBLEM NOBODY SHOULD IGNORE
Here's where I'd be careful.
There's a big difference between an AI system that makes a recommendation and one that actually changes a customer's account or loan file.
A wrong answer in a management report is a problem.
An unauthorized action involving a loan approval, payment, customer communication, or account modification can become a much bigger one.
Payday, title, and installment lenders operate under overlapping state and federal requirements.
Depending on the workflow, those requirements may involve consumer protection, fair lending, privacy, adverse action, electronic communications, payment authorization, servicing, and collections.
And the rules vary by product, state, lender structure, and the specific activity.
MY MINIMUM CONTROL REQUIREMENTS
- Defined permissions. An AI agent should access only the systems and information required for its assigned job.
- Human approval for consequential actions. Loan decisions, policy overrides, sensitive account changes, and other material actions require properly designed authorization controls.
- Complete audit trails. Record what information the system accessed, what it recommended, what action was taken, and who authorized it.
- Approved communication rules. Customer messages must respect applicable consent requirements, contact restrictions, and compliance-reviewed language.
- Validated underwriting logic. Don't allow a generative AI system to invent lending rules or substitute unsupported explanations for documented, tested policy.
- Data security and privacy. Protect borrower information, limit unnecessary data exposure, and evaluate vendors' retention and security practices.
- Monitoring and shutdown controls. Management needs the ability to detect errors, investigate unusual actions, and disable a workflow when necessary.
The goal is not to turn an AI agent loose inside your loan management system.
The goal is controlled, measurable automation with clear accountability.
YOUR LOAN MANAGEMENT SYSTEM VENDOR HAS SOME QUESTIONS TO ANSWER
If I were operating a payday, title, or installment lending company today, I'd ask my LMS vendor some pointed questions.
- Can authorized external applications securely read loan and application data?
- Can approved integrations create tasks, update permitted workflow statuses, and record activity?
- Do you provide documented APIs, webhooks, or other supported integration methods?
- Can we restrict permissions by role, action, branch, and data category?
- Can the system provide structured underwriting, exception, and subsequent loan-performance data?
- Can we connect origination decisions to first-payment default, delinquency, and charge-off outcomes?
- Are all externally initiated changes logged and attributable?
- What approval, security, and certification requirements apply before an outside AI workflow can operate?
- What are the integration costs and API limitations?
- What happens if the AI integration fails, produces an invalid request, or needs to be shut off?
I wouldn't accept "we're working on AI" as a sufficient answer.
I'd want to see what the platform can actually do, what it can safely allow, and how management can verify the results.
Nor would I assume every LMS provider has identical capabilities.
Some platforms may support extensive integrations. Others may have restrictions that make certain workflows impractical without vendor involvement.
The capabilities have to be verified.
DON'T START BY SHOPPING FOR AN AI AGENT
This is where I think some operators will get the sequence wrong.
They'll see a demonstration.
They'll hear that AI can make phone calls, send messages, review documents, and update records.
They'll buy something.
And six months later, they'll still be trying to decide whether it saved them any money.
I'd reverse the process.
START WITH THE OPERATING PROBLEM
Find the workflow that's expensive, repetitive, measurable, and sufficiently controlled to automate safely.
Establish a baseline.
Know how many applications are abandoned.
Know how long an employee spends chasing missing documents.
Know your first-payment default rate.
Know your exception approval frequency.
Know what your collectors produce per hour.
Then identify the specific work an AI system could perform.
Test it in a limited environment with actual controls and meaningful performance measures.
Compare the results against the baseline.
Expand only when the evidence justifies it.
Don't buy AI because it looks impressive in a demonstration.
Buy or build a workflow because you can identify what it costs today, what will change, and how you'll measure the result.