Flipping the Model: Digital and AI-Enabled Collections in Consumer Banking

How digital engagement, AI/machine learning, and dynamic decisioning transform collections from an outbound transaction model into an inbound solutions model.

The next credit cycle will stress-test every consumer collections operation—and most are not built to pass that.

Demand is rising into an operation that was never designed to absorb it. Elevated borrowing costs and persistent pressure on household budgets are eroding borrowers’ capacity to pay, and labor-market uncertainty compounds the risk of rising delinquencies and defaults. As of the second quarter of 2026, 4.7% of the $18.8 trillion in US household debt was in some stage of delinquency, and new delinquencies in credit cards and auto loans remained elevated. Lenders should expect not only more delinquent accounts but also a larger mix of loss mitigation requests and complex hardships that standard repayment arrangements cannot resolve.

Regulatory whiplash has exposed institutions. Repeated shifts in supervisory priorities and enforcement expectations have rewarded short-term remediation and deferred structural investment in collections and loss mitigation. Compliance infrastructure, risk management capabilities, and the technology beneath them have often been underfunded, and the resulting control gaps will surface as volumes climb. A change in enforcement posture suspends neither the underlying obligations nor the legal, financial, and reputational consequences of treating similar customers differently. Institutions that treated the lull as a reprieve rather than a window to invest now carry latent risk.

Headcount cannot close the gap. Hiring—the only scaling lever the traditional model offers—is slow, expensive, and subject to diminishing returns. Experienced collectors take months to reach proficiency, attrition is persistently high, and the next cycle will demand more skilled judgment as hardship assessments and workout decisions span multiple products and investor requirements. Meanwhile, every week of delayed intervention deepens arrears, narrows borrower options, and increases eventual losses. A default operation that can scale only by adding agents is a strategic liability.

The outbound-led engagement model is the binding constraint. Roughly 80% of collections contact at most banks is still outbound dialing, directed at borrowers selected solely on a credit score and payment history. Substantial agent capacity is consumed by repeated dial attempts, often before the lender understands the borrower’s hardship or has identified a viable resolution. Contact rates are low and falling, intake is fragmented, and effort flows to borrowers who would have self-cured or cannot or will not resolve, while those most likely to respond to help are reached late or not at all. More dialing expands activity; it does not expand effective loss-mitigation capacity.

Flip the model from outbound pursuit to inbound resolution. The institutions that will lead the next cycle are inverting their contact strategy, moving from an 80% outbound/20% inbound model to an 80% inbound/20% outbound model so that the borrower contacts the bank rather than the bank chase the borrower. Targeted, compliant digital outreach invites borrowers to engage through the lender’s authenticated app, portal, or contact center. Structured intake captures the reason for hardship at first contact, and decision support identifies the workout with the highest propensity to cure within approved policy. Outbound contact is retained where it is effective or required, but borrower-initiated engagement becomes the default path. Four capabilities make the inversion possible:

• Multidimensional segmentation. Machine-learning (ML) models that combine loan, relationship, external, and hardship data to place every delinquent borrower on two axes: willingness and ability to pay.
• Digitally engineered right-party contact. Behaviorally designed, compliance-screened outreach across email, text, app, and online banking that invites the willing but unable borrower to self-identify through an authenticated channel. Every response is, by definition, a right-party contact (RPC).
• AI-driven resolution. A decision engine that uses the hardship information captured at first contact to structure the workout with the highest propensity to cure and learns from every outcome.
• Conditionally prequalified offer. A cure tailored to the customer’s hardship, with the highest propensity to cure, structured and offered prior to customer contact.

Redesign first, then scale with AI. Operational redesign makes the shift practical; artificial intelligence (AI) makes it scalable. Integrated workflows can reduce handoffs and repeated information gathering. AI supports borrower segmentation, offer evaluation, documentation, and exception identification, while specialists concentrate on complex hardships under clear escalation rules and auditable controls. Automating today’s process would only accelerate its weaknesses. Performance is measured by sustained cures and cost to resolve alongside customer outcomes and control effectiveness.

Material value is at stake. BRG client deployments of these capabilities point to RPC improvements of 10% to 15% in initial pilots and 25% or more at scale, alongside lower cost to collect, shorter handle times, faster agent onboarding, fewer compliance failures, and higher customer and employee retention. Independent research reaches similar conclusions: digital-first customers are more likely to pay when reached through their preferred channel, and generative AI can lower collections operating expense by up to 40% while lifting recoveries by around 10%.

The window to act is now. Leaders must engineer compliance and risk management into the design, rather than layer them, while resolving four questions together: What operating model can scale through the next cycle? Which processes must be redesigned before they are automated? Where should digital self-service end and specialist judgment begin? How must workforce roles and skills evolve as technology absorbs routine work?

Loss-mitigation capacity can be built before rising demand exposes the limits of the current operation. Banks that modernize now will enter the next downturn with a scalable, auditable, and customer-centric default operation. Those that do not will meet it with more dialers. The practical starting point is a focused pilot in the segment where the inbound model has the most to prove and the fastest path to value: willing but unable borrowers in early stage delinquency.