A stage-by-stage breakdown of every customer conversation across the loan lifecycle, scored against five criteria and pressure-tested by the people who run these operations.
Most auto lenders have already bought something with AI in the name. The question isn't whether to invest anymore. It's which investments are actually returning something.
AI shows up everywhere in auto lending. It touches underwriting, fraud, and scoring. However, automating conversations across originations, servicing or collections is where a lot of focus we have seen coming from the management.
What follows is a stage-by-stage breakdown of every conversation across the loan lifecycle. We trace the loan from origination through recovery. We give a straight answer on which conversations an AI agent handles, which need a human standing by, and which stay human.
Prodigal mapped every customer conversation across the loan lifecycle, from origination through post-charge-off recovery, and scored each one against five criteria: customer experience, time to value, ROI and volume, complexity, and compliance exposure.
A panel of auto-lending leaders, including CEOs, VP of Collections at national lending platforms, and indirect-lending executives from the country's largest banks, pressure-tested the findings.
This is what we found.
Operations leaders face constant pressure to deploy AI agents from their management. However, it is easier to start and then keep on waiting forever for the ROI, getting again constant pressure from the management to show progress.
Scoping the AI deployment is not an easy feat and understanding what to automate seems like easy straightforward task, however, it requires understanding of what technically possible and as well as understanding consumer adoption and preferences in which point they would like to talk to an AI agent.
Understanding both of these requires iterations or experience of deploying AI agents in the industry, not just a CX AI agent but an actual AI agent handling very specific lending conversation flows and have taken outcome guarantees which are relevant to the industry for example RPC, Promise-to-pay rate, successful auto pay setups while making the onboarding calls not just deflection rates.
With all the experience and time we have spend inside these conversations, we created the library of conversations and rated where it can be automated using AI as of now (June 2026).
This library delivers our honest read on where the line sits. We organize it by how a loan moves through your operation. We track it stage by stage from the welcome call through post-charge-off recovery. Every conversation receives one of three verdicts.
How to decide which lending conversations the AI agent should handle, and what it shouldn't touch
That answer depends on five things:
An anecdote on customer behavior in subprime auto lending: For a customer who's behind, a call from a collector can carry shame. An AI agent doesn't. Plenty of customers let a collector's call ring out, not because they won't pay, but because picking up means being judged and asked very tough questions. The same person will often answer a text, or talk openly with an AI agent, precisely because there's no one on the other end to feel small in front of.
This is the loan lifecycle as your floor actually lives it. Every customer conversation, in the order your team encounters it, with a verdict on each one. Where a use case was scored by the lender panel, the score is on the card.
The relationship starts here.
A loan funds. The customer needs to know their terms, first due date, and how to pay.
The single best moment to set a customer up to succeed. Walk them through the payment date, get autopay set up, answer the early questions. The AI agent does it for every new loan, not just the ones a person has time for. Your panel scored this the top originations use case.
A deal funds. Before it books, residence, employment, references, the vehicle, and the customer's intent all need verifying.
A structured seven-to-ten-question interview, run on every funded deal, not just a sample. Scripted, repetitive, and high-volume: exactly where an AI agent is strongest.
An applicant has questions, or a file is missing a stipulation: income, residence, references.
The AI agent guides applicants through what's needed and chases missing documents so files reach a funding decision faster. It keeps the deal moving without pulling a person off higher-value work.
The customer is paying. This is where you either build the relationship or lose it quietly.
A customer calls to ask what they owe, or what it takes to pay off.
Pure account lookup. The AI agent reads the account and answers instantly. No hold music, no queue, any hour of the day.
Payment is due in three days.
A friendly heads-up before the due date, sent across the channel the customer actually reads: voice, SMS, or email.
A customer wants to pay. At 9am, or at 11pm on a Sunday.
Payments don't keep office hours. The AI agent accepts them around the clock, so a customer who is ready to pay never hits a closed door. Your panel rated this among the highest-value use cases across the entire lifecycle.
A customer wants to change autopay, adjust a due date, or update a phone number.
Small self-service changes that quietly prevent the next delinquency. Every autopay enrollment and every due date aligned with payday is one fewer missed payment later. The AI agent handles them on the spot, inside your rules.
A customer wants to confirm a payment landed, or asks for a receipt.
Customers want certainty fast. The AI agent provides it without a callback.
A customer calls after an accident with their claim and adjuster details.
First-notice-of-loss: capturing the claim number, the adjuster's information, and what happened, then setting clear expectations on next steps. Structured intake an AI agent handles cleanly, so nothing slips in the hours after an accident. The claim and total-loss work that follows stays with your team.
Lending is not a phone channel anymore: Plenty of customers will never pick up a call but will answer a text in seconds or reply to an email at midnight. A real AI agent meets them on the channel they choose, notices which one a given customer actually responds to, and stays there. The conversation moves across channels without the customer ever repeating themselves. For a large share of your book, digital self-serve is not a lesser path. It is the only one that actually works.
Lender panel rated omnichannel outreach 4.0 / 5.
The highest-leverage moment in the lifecycle. Most of these are slips, not crises.
An autopay or ACH bounces. The account just went past due.
Speed is everything here. The AI agent reaches every bounced account the same morning, not three days later when a person finally gets to the queue.
A customer misses for the first time. Usually a slip, not a crisis.
A warm, early nudge cures most of these before they become a problem. The AI agent handles it at full coverage, so no first-miss slips through.
A customer says they can pay Friday. Someone needs to lock it in.
The AI agent captures the promise, explains in plain numbers what waiting actually costs, and sends the payment link on the day. Every time, on every call. Your panel scored this the single highest-value use case across the entire lifecycle.
A promised payment date is approaching, or just passed without a payment.
The reminder nobody on a busy floor has time to send. The AI agent sends it every time, on the right channel, and follows up the moment a promise breaks.
The slip has become a pattern. This is where the conversation gets harder.
A Promise-to-Pay came and went. Time to re-engage.
Still routine outreach. The AI agent re-establishes contact and works toward a new commitment at a volume a human team cannot match across a full book.
A customer cannot fully catch up and asks about moving a payment to the end of the loan.
The AI agent runs the full extension conversation: presenting options, disclosing the real cost clearly, and working through the customer's situation, hardship included. It is built to negotiate. Only the case that falls entirely outside the parameters routes to a person.
A customer offers a partial payment that will not cure the account.
The AI agent explains honestly why a partial keeps a customer stuck rather than digging them out, and works the hardship options with them. It handles the negotiation well. Only a truly exceptional situation needs a person to step in.
Pre-repo. This is where the line gets drawn.
A new phone number surfaces for a customer who had gone dark.
Re-establishing contact is still routine outreach. The AI agent works fresh numbers the moment they land, so no time is lost.
The account is at the edge. The customer is frightened, the car is on the line, and a real workout has to be built in real time.
This conversation runs on human judgment, authority, and the ability to read whether a promise will hold. The math an AI agent can do. This moment belongs to a person. Automating it does not cut cost. It loses the account and invites risk.
A mix of high-volume routine work and low-volume cases that need real judgment.
The customer needs to know how to get their belongings back.
The AI agent identifies the repo company, connects the customer instantly, and sets clear expectations on lot hours, what to bring, and scheduling. A person stays one click away for disputed repos, missing items, or redemption questions.
The auction is over but the balance owed is not.
By the time an account is this deep, the routine paths have been worked. What remains is a smaller queue of harder cases where re-engaging the customer requires a real read of the situation and where trust has to be rebuilt before any plan is on the table. This belongs to a person.
To pressure-test our own read, Prodigal asked a panel of auto-lending leaders to score each use case on how ready it is for an AI agent today, on a scale of 1 to 5. The panel included CEOs of subprime auto finance companies, VP of Collections at national lending platforms, indirect-lending product leads at two of the country's largest banks, Chief Risk Officers, credit union indirect-lending leads, and professionals from captive auto lenders.
An AI agent has no quota to hit by Friday: A collections floor evaluates performance by the month. That pressure is real. It quietly pushes agents toward the payment that closes today. It ignores whether a different plan would secure the account for a year. An AI agent carries no monthly targets and no commission pressure. It consistently chooses what is right for both the borrower and the lender over the life of the loan. It applies the exact extension that fits. It builds the plan the borrower can keep. Applied across thousands of conversations, this long-term alignment drives enterprise returns. It frees your best people for the two moments where human judgment belongs: saving the account and rebuilding it after.
The real payoff: Automating the routine is not about replacing your team. It is about clearing the floor. Your best people should save a car at day 85. They should not be buried under day-5 reminder calls. An AI agent handles those calls hours earlier.
Most implementations slow down and the majority of the blockers can be solved if you chose the right partner for deploying an AI agent — and the right partner can work alongside you to figure out the right use cases to automate. They bring AI expertise as well as the expertise of deploying AI agents in your particular industry, and the ability to integrate with the unique systems of the industry.
Prodigal's approach is domain-specific. Pre-built capabilities tailored to servicing and collections mean less time building infrastructure and more time seeing results. Every automated and human-ready conversation in this library runs on two things working together.
proAgent carries a net promoter score of 9.6 in production. That number comes from the borrowers on the other end of these conversations, the people this library was built around. The investment in AI agents is paying off on both sides of the conversation.
.svg){kind=icon}
.webp)
.png)
.webp)
.jpg)
-min.avif)
.avif)
.avif)
.avif)
.avif)
