What problem does it solve?
Many customers who miss a payment are not refusing to pay: their payday moved, a payment failed or their circumstances changed. A short, timely conversation in the first days of arrears resolves many of these cases, but collections teams often lack the capacity to reach every account in that window, and contact outside office hours is limited. Accounts then roll into later buckets where recovery is harder and more expensive.
At the same time collections is a heavily regulated conversation. Contact frequency, tone, disclosures and the treatment of customers in financial difficulty are all set out in rules, and getting it wrong with a vulnerable customer causes real harm. Automation that only pushes for payment makes this worse; automation that listens and routes well can make it better.
How does it work?
- Reach out at the right time. The agent contacts accounts in early arrears on the channel and at the time each customer is most likely to respond, within contact frequency rules.
- Verify and disclose. It confirms identity before discussing the debt and gives the required disclosures, including that it is an AI agent.
- Understand the reason. It asks why the payment was missed and classifies the answer, such as a failed payment, a changed pay date or a change in circumstances.
- Resolve within rules. It takes a payment, sends a secure payment link, moves a due date or sets up a short arrangement, but only within limits the lender has approved.
- Route hardship and vulnerability to people. Mentions of job loss, illness, bereavement, domestic abuse or distress, or any request for help, go to a trained specialist with a summary, and collection activity pauses.
- Record everything. Every contact, disclosure, promise to pay and arrangement is logged with its basis for audit and complaint handling.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Phone and voice, SMS, WhatsApp, Email, Web chat
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Cost reduction | Too few to pool | 63% to 75% | 2 | 2 vendor |
| Recovery uplift | Too few to pool | 43% to 100% | 2 | 1 organization, 1 vendor |
| Containment rate | Too few to pool | at least 20% | 1 | 1 vendor |
| Interactions handled | Not pooled | at least 400,000 | 1 | 1 vendor |
Value drivers: Lower cost to serve, Risk and loss reduction, Customer experience, Compliance quality.
Indicative value
A lender with 50,000 accounts entering early arrears each year
USD 240,000 to USD 1.7 million
Collections contact cost avoided per year
How this is calculated
Formula: accounts * contactsPerAccount * automatedShare * costPerContact. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Accounts entering early arrears per year accounts, accounts per year | 50,000 | 50,000 | The reference lender. |
| Outbound and inbound contacts per account in early arrears contactsPerAccount, contacts per account | 4 | 8 | Editorial assumption. Replace with your own contact data. |
| Share of those contacts the agent completes without a person automatedShare, fraction of contacts | 0.4 | 0.7 | Editorial assumption, deliberately below full automation because hardship cases must reach people. |
| Cost of a human handled collections contact costPerContact, USD per contact | 3 | 6 | Editorial assumption. For comparison, SameDay Auto Finance's vendor reports 75% lower collection call costs in early delinquency. |
What it leaves out: Contact cost only. It leaves out the usually larger effect of fewer accounts rolling into later arrears and charge off, the cost of running the AI and the payment integration, and the effort of compliance review.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Day Knight & Associates
United States · Professional services · 2026
Day Knight & Associates, a Missouri agency founded in 2001 that collects healthcare and consumer debt, moved from outbound and inbound voice AI to a combined voice and SMS setup. The vendor reports that adding SMS reached consumers that voice alone missed and that, within a month of going multichannel, the cost of collecting a dollar fell from 22 cents to 8 cents; the agency's vice president of business development says collections doubled. It shows how channel choice, not only automation, drives results in collections.
- Cost reduction: 63%, cost of collections, after moving to voice plus SMS
"See how Day Knight & Associates used AI for Debt Collections to double recoveries, reduce collection costs by 63%, and scale multichannel outreach with Voice AI and SMS."
Claimed by: vendor - Recovery uplift: about 100%, collections after moving to voice plus SMS
"After adopting Skit.ai’s multichannel platform, we were able to double our collections and connectivity rate."
Claimed by: organization
SameDay Auto Finance
United States · Banking · 2026
SameDay Auto Finance, a Dallas auto lender whose portfolio sits mostly in early delinquency, moved its early stage outreach to AI voice agents calling around the clock, with SMS for customers who do not answer calls, and redeployed its human agents to inbound returns, skip tracing and complex accounts. The rollout ran in four phases over a year. The vendor reports 43% higher collections and 75% lower collection call costs in the early delinquency buckets.
- Recovery uplift: 43%, early stage delinquency
"SameDay Auto Finance Achieves 43% Higher Collections and 75% Lower Call Costs in Early-DPD using AI for Collections."
Claimed by: vendor - Cost reduction: 75%, collection call cost, early stage delinquency
"SameDay Auto Finance Achieves 43% Higher Collections and 75% Lower Call Costs in Early-DPD using AI for Collections."
Claimed by: vendor
Southwest Recovery Services
United States · Professional services · 2026
Southwest Recovery Services, a Dallas based collections BPO founded in 2004 that works medical, subprime, B2B and B2C debt across several states, deployed Skit.ai's voice AI for both outbound and inbound calling. The vendor reports a two phase deployment automated more than 400,000 calls under a requirement for real time compliance with Regulation F and applicable state rules, and reached a 20 to 25% containment rate with a 50 to 55% inbound right party contact rate, without adding headcount.
- Interactions handled: at least 400,000, two phase inbound and outbound deployment
"A two-phase Al inbound/outbound calling deployment automated over 400,000 calls, resolving the outreach gap and capturing every consumer callback."
Claimed by: vendor - Containment rate: at least 20%, ongoing
"20-25% Containment Rate"
Claimed by: vendor
OTP Bank Romania
Romania · Banking · 2020
When Romania's 2020 legislation allowing customers to postpone loan instalments drove a 125% spike in incoming requests, OTP Bank Romania launched Octavian, an AI agent built on DRUID's platform and embedded on the bank's website. Octavian gives round the clock, personalised guidance on OTPDirekt and OTPay digital services and takes a customer's payment deferral request directly in conversation. Behind it, UiPath robotic process automation validates the submitted data, checks eligibility against the bank's criteria, decides the outcome, and uploads the approved request straight into the bank's business process management system, so a request can be resolved without a human touching it.
- Handling time reduction: 30x
"The results are striking: the average request processing time was slashed from 10 minutes to just 20 seconds, a 30x efficiency boost."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Arrears data per account with contact history, consent and preferred channel
- Written arrangement rules the agent may offer, with limits per product
- Contact frequency and time of day rules per market
- Vulnerability and hardship triggers agreed with the specialist team
Systems to integrate
- Collections or loan servicing system
- Payment gateway for card and bank payments, or payment links
- Telephony and messaging channels, including consent management
- Case management or CRM for hardship referrals
- Complaint handling
Complexity: Medium
Conversation design and integrations (arrears data, payments, arrangement rules, dialler and consent) are manageable. The effort is in conduct rules per market, vulnerability detection and a clean handover to specialists.
- 1
Start in the first days past due
Reminder and resolution conversations in the first bucket are high volume, low risk and the cheapest place to prevent roll forward.
- 2
Write the arrangement rules down
List exactly which arrangements the agent may offer (date change, short plan, split payment), with limits, and what it says when a request is outside them.
- 3
Design hardship routing with the specialists
Agree the phrases and situations that pause collection and route to a person, test them on real transcripts, and err on the side of routing.
- 4
Build compliance in
Encode identity checks, disclosures, AI disclosure, contact limits and quiet hours in the flow, not in the prompt, and log each one.
- 5
Pilot against a control group
Run the agent on part of the book, compare roll rates, promises kept, complaints and satisfaction with a human handled control, then widen.
Guardrails
- Identity verification before any mention of the debt
- Clear disclosure that the customer is speaking with an AI agent
- Arrangements only within preapproved rules; anything else goes to a person
- Immediate handover and a pause in collection on any hardship or vulnerability signal
- Contact frequency, time of day and channel consent enforced by the system
- No threats, pressure tactics or misleading statements, checked by output guardrails
KPIs to instrument
- Roll rate from the first to the second arrears bucket versus control
- Promise to pay kept rate
- Share of conversations routed for hardship, and specialist agreement with the routing
- Complaints and conduct breaches per thousand conversations
- Cost per account resolved
Human in the loop
Trained specialists handle every hardship, vulnerability, dispute and complaint case and every arrangement outside the rules. Quality teams review a sample of AI conversations each week against the conduct standard, and compliance approves every change to scripts or arrangement rules.
Common failure modes
- Missed vulnerability
- The agent keeps pressing for payment when a customer mentions illness or job loss. Route on broad signals and review missed cases every week.
- Arrangements that fail
- Easy plans accepted to end the call and then broken. Check affordability within the rules and track kept rates per arrangement type.
- Contact that becomes harassment
- Automation makes it cheap to call too often. Enforce frequency limits in the system, per customer across channels.
- Payment data in transcripts
- Card numbers read aloud end up in logs. Use payment links or secure capture and mask card data before it reaches the model.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A customer facing collections agent must disclose that it is AI (Article 50). It is not listed in Annex III as long as it applies preapproved arrangement rules and does not itself evaluate creditworthiness; an affordability model that decides who gets which arrangement for individuals should be assessed separately against Annex III point 5(b).
Rules that apply
Guidance
- FG21/1: guidance for firms on the fair treatment of vulnerable customers (Financial Conduct Authority, Europe). Expectations for identifying and supporting vulnerable customers, including in automated channels.
- Debt Collection Practices (Regulation F) (Consumer Financial Protection Bureau, North America). US rules under the Fair Debt Collection Practices Act on communications with consumers, harassment, misleading statements and unfair practices. They govern debt collectors as the FDCPA defines that term, which generally covers third party collectors rather than creditors collecting their own debts.
- RG 96 Debt collection guideline: for collectors and creditors (Australian Securities and Investments Commission and Australian Competition and Consumer Commission, Asia Pacific). Joint ACCC and ASIC guideline on how Australian consumer protection law applies to debt collection, for creditors collecting their own debts and for external collection agencies.
- Regulation (EU) 2024/1689 (AI Act), Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be informed that they are interacting with an AI system unless this is obvious from the context.
- Declaratory ruling on AI generated voices under the TCPA (FCC 24-17) (Federal Communications Commission, North America). AI generated voices count as "artificial or prerecorded voice" under the TCPA, so US outbound AI calls need the consent the TCPA requires.
Controls to put in place
- Approved scripts and arrangement rules under change control
- Logged identity check, disclosures and consent for every contact
- Weekly quality sampling with a specific check for missed vulnerability
- Complaint monitoring linked to AI conversations
- Card data masked or captured outside the conversation
Frequently asked questions
- What results do lenders report from AI collections agents?
The figures on this page come from vendor case studies about US auto lenders and collection agencies. Skit.ai reports 43% higher collections and 75% lower call costs in early delinquency at SameDay Auto Finance, and 63% lower collection costs at Day Knight & Associates, whose own executive says collections doubled. Test against a control group before relying on such numbers.
- Should an AI agent handle customers in financial hardship?
It should recognise them and pass them to a trained person quickly, not negotiate hardship on its own. Hardship and vulnerability need judgment, flexibility and often referral to support that an agent should not decide.
- Is an AI collections agent allowed to call customers?
Generally yes, within the same rules as human collectors: identity checks, disclosures, contact frequency limits, quiet hours and channel consent, plus disclosure that it is AI. Some markets add rules on automated calls: in the US, the FCC treats AI generated voices as artificial voices under the Telephone Consumer Protection Act, which sets consent rules for such calls. Check local telemarketing and collection law.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for early collections and hardship support", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/collections-and-hardship-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 30 September 2026: Added deployments: OTP Bank Romania.
- 29 September 2026: Added deployments: Southwest Recovery Services.
- 27 September 2026: First published