Judged on collections rules, published results, connected systems and price, the best AI voice agents for debt collection in 2026 are Vector by Synthflow for lenders collecting their own loans and Skit.ai for agencies and debt buyers. Salient, Domu and Prodigal also build for lenders, Floatbot for agencies, and Genesys, NiCE and Five9 add AI agents to their contact centers.
Vector, by Synthflow. Synthflow publishes this list and makes Vector, an AI agent that works each consumer-lending account across calls, text and email. In a one week snapshot at a national lender (13,609 accounts, 21,988 dials), Vector reached 4.9% right-party contact per dial vs 3.5% human benchmark (+40%); 49% of conversations ended in a promise to pay, $215,018 promised to pay, at $12 per right-party contact vs $82 fully burdened (the lender's week). RPC to missions (a mission is one account worked): low 1.2%, mid 2.4%, high 4.2%, Vector 7.9%. The low, mid and high bars are rep benchmarks from published industry and regulatory collections data, not a randomized control. Book a Vector demo.
The ten at a glance
Grouped by the buyer each fits. Prices as each vendor's own page states them, checked in November 2026.
| Vendor | Best for | Price, as its page states it | Phone systems it works in |
|---|---|---|---|
| For a lender's own collections | |||
| Vector (Synthflow) | Consumer lenders that want one agent for collections and servicing, in the US or the EU | Not by the minute: "Pricing is built around the outcomes we agree." | Your carrier or Vector's, over SIP trunking; PBX or BYOC with Genesys, Cisco or Avaya |
| Domu | Lenders that take payments on the call through their own processor | No public price | Five9, LiveVox, Genesys, Amazon Connect |
| Prodigal | Consumer-finance teams that also want agent co-pilot and analytics | No public price | TCN, Five9, Genesys, NICE (logos) |
| Salient | US banks, credit unions and auto lenders that also want audit and disputes | No public price; work starts from a signed SOW | None named |
| For agencies and debt buyers | |||
| Skit.ai | Agencies, debt buyers and law firms that pay on what is collected | Contingency on what is collected; no public rates | Your dialer over SIP or call forwarding; TCN, NobleBiz, Twilio, Telnyx |
| Floatbot | Agencies and law firms that want a published plan | From $99 a month billed yearly, for 500 voice minutes or chat sessions; Enterprise by quote | Five9, Genesys, NICE CXone, RingCentral, Avaya, Cisco, TCN |
| Inside the contact center you run | |||
| Five9 | Five9 dialer floors that add AI agents | AI Agents: add-on, no public price; voice seats from $159 a month, 50-seat minimum | Five9's network or BYOC |
| Genesys Cloud | Genesys floors, above all with Latitude by Genesys | From $75 per user a month billed yearly; AI agent 1.2 Experience Units an interaction at $1.00 a unit | Genesys Cloud Voice or BYOC |
| NiCE CXone | CXone floors that want proactive collections conversations | AI agents in the Ultimate Suite: $249 per agent a month plus $0.25 a session | NiCE Voice Services or BYOC |
| A platform you build on | |||
| Retell AI | Teams with engineers that build their own collections agent | $0.07 to $0.31 a minute; Enterprise by quote | Its own numbers, or your carrier over SIP |
We also read InDebted, TrueAccord, Equabli, Symend, Vodex, CollectWise, Moveo.ai, Dapta, Talkdesk, Bland AI and Regal. The ten are six agents built for collections, three contact center platforms with collections offers and one voice AI platform you build on.
How we judged
Four criteria, each a set of checks that a vendor's own pages meet or do not. The second table shows which checks each vendor meets. There is no total score, and the ten are grouped by buyer, not ranked. No vendor paid to appear or saw its entry.
| Criterion | The checks, on the vendor's own pages |
|---|---|
| Rules | A calling window or call cap the product enforces; collections or AI disclosures; TCPA named; SOC 2 Type II or ISO 27001; PCI DSS for card payments |
| Results | A collections result, with the customer named, its counts, its period and a baseline |
| Systems | A named collection or servicing system, phone system and payment processor; voice, text and email in one agent; an API or file feed |
| Price and launch | A public price or pricing model; pilot or trial terms; a stated time to go live |
| Vendor | Rules | Results | Systems | Price and launch |
|---|---|---|---|---|
| Vector | Window, frequency, AI disclosure, TCPA, SOC 2, ISO 27001 | Unnamed lender: counts, period, benchmark | Genesys, Cisco, Avaya; Salesforce, Zoho; voice, text, email; webhooks, REST | Model, pilot steps |
| Domu | Hours, Mini-Miranda, TCPA, SOC 2, PCI | Named (DigniFi): RPC rate | Four phone systems; voice, text, email; SFTP, API | None stated |
| Prodigal | PCI badge | Unnamed: dollars, containment | Finvi; four phone systems; voice, text, email | None stated |
| Salient | Time-zone rule, AI disclosure, TCPA | Named (Westlake): period, baseline | FIS, Fiserv, Jack Henry; four channels; API | Go-live time |
| Skit.ai | Window, cap, disclosures, TCPA, SOC 2, ISO 27001, PCI | Named (SameDay Auto Finance): period, baseline | Collection systems, carriers, gateways; three channels; API, SFTP | Model, pilot, go-live time |
| Floatbot | Window, cap, Mini-Miranda, TCPA, PCI | Unnamed: dollars, period | DAKCS, InterProse; seven phone systems; three processors; three channels | Price, model, 14-day trial |
| Five9 | TCPA, SOC 2, ISO 27001; AI Agents not PCI-compliant | None for AI agents | Own network; CRM adapters | Seat prices only |
| Genesys Cloud | SOC 2, ISO 27001, PCI | None found | Latitude; own voice; digital channels | Price, model, trial |
| NiCE CXone | TCPA, SOC 2, ISO 27001, PCI | Unnamed: rates | Own voice; digital channels | Price, model |
| Retell AI | DNC and TCPA litigator checks; SOC 2, ISO 27001 | Named (Medical Data Systems): counts, period | Carriers over SIP; API | Price, model, trial credit |
For a lender's own collections
Vector (Synthflow)
Vector is an AI agent for collections, applications, refinancing and servicing at consumer lenders; it is "given an account and an outcome", not a single call. A policy engine runs around 70 checks before a call is placed or a message is sent, and its rules name FDCPA, Regulation F, TCPA, UDAAP and state contact rules in the US, and the EU AI Act's Article 50 disclosure in the EU. It speaks English, Spanish, German and Greek, runs in a US region or an EU region in Frankfurt, lists SOC 2 Type II and ISO/IEC 27001:2022, and connects to your telephony, CRM, servicing system and payment processor over webhooks and REST (vector.synthflow.ai). At a Latin American lender over five calling days, Vector's results were 62.5% of contacts reached, and 23.5% of verified borrowers committed to the full overdue amount within seven days.
Domu
Domu's engine is "built for collections, where the rules are strictest" and now also completes payments, originations, renewals and servicing; its agent opens on voice and follows up by text and email in one thread. Its home page shows Reg F as the "approved rule source", the Mini-Miranda, calling hours "controlled end to end", SOC 2 Type II, TCPA and PCI, payments "through your processor, without card data touching the recording", and licenses to make outbound contact in its own right. It names Five9, LiveVox, Genesys, Amazon Connect, a daily SFTP file and an API, and reports a 51.4% RPC rate at DigniFi.
Prodigal
Prodigal calls proAgent "the AI Agent purpose-built for collections": it works servicing and collections across voice, text and email, takes payments on any channel or through your payment portal or IVR, and detects hardship signals and routes complex cases to human agents (proAgent). The same page reports "Over 47%" containment and "Over $5M" collected, with no customer named, and says its guardrails encode "each customer's regulatory requirements" without naming the rules. Its home page adds agent co-pilot, speech analytics and propensity-to-pay scores, and shows TCN, Five9, Genesys, NICE and Finvi logos.
Salient
Salient builds AI for US consumer lenders: it "automates collections, servicing, regulatory audit, insurance claims, and dispute resolution". Its agent Taylor works inbound and outbound across voice, SMS, email and chat inside "real-time compliance guardrails for FDCPA, Reg F, UDAAP, TCPA, and CAN-SPAM" (Taylor), and its privacy policy says each voice agent discloses that it is an AI. Its customers page reports a 34% increase in right-party contact rate at Westlake Financial within 60 days. Most deployments go from signed SOW to production in 60 to 90 days.
For agencies and debt buyers
Skit.ai
Skit.ai sells AI for debt collection to agencies and to "banks, lenders, debt buyers, and law firms". Its responsible voice AI page lists the Reg F 8 a.m. to 9 p.m. window, a "7-in-7 contact frequency cap", the Mini-Miranda and AI disclosure, SOC 2 Type II, ISO 27001 and PCI-DSS. It connects to collection systems such as Finvi, IDMS and Interprose and to payment gateways such as Repay and PayNearMe (integrations), and prices on contingency after a 30 to 60 day pilot on your own accounts. At SameDay Auto Finance, the PTP rate went from 5.7% to 11.5%.
Floatbot
Floatbot's collections agent, LEXI, verifies identity, gives the Mini-Miranda, secures promises to pay and takes "PCI DSS-compliant payments in the same conversation", with "built-in call-frequency and calling-window rules" (collections page). The page names collection systems (DAKCS, InterProse), payment processors (Repay, PayNearMe, Nuvei) and contact centers from Five9 to TCN, and adds a desktop assistant for collectors. It is the only collections specialist here with a published plan. Its featured result, "$1M+" collected in 90 days, is for a healthcare collections firm it does not name.
Inside the contact center you run
Five9
Five9's debt collections page sells predictive, power and preview dialers and names USCB America as a customer. Its AI Agents add-on is in controlled availability, and Five9's documentation says "AI Agents is currently not PCI-compliant", so card payments go to a PCI-compliant IVR (AI Agents security). The platform lists SOC 2 Type 2, ISO/IEC 27001:2022 and PCI DSS v4.0.1, and voice seats start at $159 a month with a 50-seat minimum (pricing).
Genesys Cloud
Genesys sells both the contact center and a collections system: Latitude by Genesys is "the all-in-one solution for accounts receivable management", with pre-built integrations to Genesys Cloud. Its Agentic Virtual Agent speaks 24 languages and is billed in Experience Units. Genesys lists ISO/IEC 27001:2022, SOC 2 Type 2 and PCI DSS for the platform; we found no AI collections result on its pages.
NiCE CXone
NiCE pairs CXone with NiCE Cognigy AI agents and with ContactEngine for Collections, which holds proactive collections conversations "over days, weeks, or months if necessary". Its collections page reports "22% Additional dollars collected" and a "47% Increase in payment arrangements negotiated", with no customer named. Its compliance list includes TCPA, PCI DSS, SOC 2 and ISO 27001.
A platform you build on
Retell AI
Retell AI is "an AI voice agent platform for contact centers": your engineers build the collections agent in its builder or through its API. Its dialer checks the DNC registry, TCPA litigators and line type, its compliance page lists SOC 2 Type 1 and Type 2, HIPAA and GDPR, and its terms class payment card data as prohibited unless authorized in writing. Medical Data Systems collects about $280,000 a month with it, on 30,000 calls a month at 70% containment.
Rules for an AI collections call
The regulators' own text for US calls, in short. This is not legal advice; take your program to counsel.
| Rule | What the text says | What the agent does |
|---|---|---|
| AI voices, FCC 24-17, para. 2 | The TCPA's limits on "artificial or prerecorded voice" cover "current AI technologies that generate human voices", so calls need "the prior express consent of the called party", absent an emergency purpose or exemption | Holds consent for each number before it dials |
| Calling hours, 12 CFR 1006.6(b)(1)(i) | Absent knowledge to the contrary, "a time before 8:00 a.m. and after 9:00 p.m. local time at the consumer's location is inconvenient" | Times the window to the borrower's location |
| Call frequency, 12 CFR 1006.14(b)(2) | A debt collector is presumed to comply if it calls a person about a debt neither "More than seven times within seven consecutive days" nor within seven days after a telephone conversation | Counts every attempt, human and AI, per person and per debt |
| Disclosures, 12 CFR 1006.18(e) | The first communication says the collector "is attempting to collect a debt and that any information obtained will be used for that purpose"; each later one, that it "is from a debt collector" | Gives both and records that it did |
| Massachusetts, 940 CMR 7.04(1)(f) | A creditor may not initiate calls, texts or recorded messages "in excess of two such communications in each seven-day period" to the debtor's home, cell or personal number, for each debt (two in 30 days to other numbers); a reply to the debtor's own request does not count, and a debt here is one "more than 30 days past due" unless the debtor agrees a different period (940 CMR 7.03) | Holds a cap for each state, below Reg F's seven where the state sets one |
Who is a debt collector. Regulation F covers a person who collects debts "owed or due, or asserted to be owed or due, to another", or whose business has collection as its "principal purpose" (12 CFR 1006.2(i)(1)). A lender collecting its own loans in its own name is generally outside that definition, with the exception the same paragraph names: a creditor that "uses any name other than its own that would indicate that a third person is collecting" is a debt collector. The TCPA covers every AI call either way; the Massachusetts rule binds creditors too, and California's Rosenthal Act holds a lender collecting its own consumer debts to the FDCPA's calling-hours and harassment rules (Cal. Civ. Code 1788.2(c), 1788.17).
Questions to send every vendor
Send these to each vendor on your shortlist, and score the written answers before any demo.
| Question | A good answer shows |
|---|---|
| 1. How are calls counted against the seven-in-seven call frequency? | Per person and per debt, across every dialer, collector and AI agent, checked before each dial |
| 2. Whose local time sets the calling window? | The borrower's location, with a rule for accounts with no address on file |
| 3. When does the agent say it is an AI, and when does it give the debt collector disclosure? | Words your compliance team sets, in order, with a record of each |
| 4. How is consent held, and how is a revocation honored? | Consent per number and channel; a stop on any channel takes effect at once |
| 5. Can the agent take a card payment, and in which PCI DSS scope? | Tokenized capture or a hand-off to your PCI IVR; no card numbers in recordings or transcripts |
| 6. What does each call write back to your servicing or collection system? | Disposition, promise date and amount, consent changes, after every call |
| 7. Which decisions stay with people? | Hardship, disputes, bankruptcy, attorney representation and settlements beyond set limits, handed over with the full history |
| 8. How will the pilot be measured? | One segment against your own collectors: right-party contact per dial, promise to pay per right-party contact, kept promises, cost per right-party contact, complaints |
Whom Vector fits
Vector is for consumer lenders that collect their own loans and want one agent to work each account across calls, text and email, in the US or the EU, inside rules their compliance team sets and versions. Synthflow, the platform under it, is sold as an annual enterprise agreement, scoped to call volume, phone system, integrations, security needs and launch support, with no free plan and no rate by the minute. Its real limits, each with what a lender does about it:
- It speaks English, Spanish, German and Greek, so accounts in other languages stay with your collectors.
- Its page names PBX or BYOC telephony with Genesys, Cisco or Avaya and CRMs such as Salesforce or Zoho, but no specific servicing system or payment processor, so a dedicated forward-deployed engineering lead connects yours over webhooks and REST in the build phase.
- Its page lists no PCI DSS scope for Vector; it schedules payments through your payment processor, so keep card capture in that processor's flow.
- It is built for a lender's own book, not for an agency working placements from many creditors; an agency can build its own collections agent on the Synthflow platform (automated debt collection calls).
- It does not decide hardship, affordability or forbearance: it hands those accounts to your team with their history, so keep collectors on them.
FAQ
Is it legal to use an AI voice agent for debt collection calls?
The FCC ruled in February 2024 that AI-generated voices are "artificial" voices under the TCPA, so these calls need the called party's prior express consent unless an exemption applies. For a debt collector, Regulation F's calling hours, call-frequency presumption and disclosures apply to these calls too. State rules can add to these, such as Massachusetts' call limit and California's Rosenthal Act.
Can an AI voice agent take a card payment on the call?
Some can. Floatbot and Domu take payments in the conversation, and Skit.ai takes card-on-call payments through payment gateways; Five9's documentation says its AI Agents are "currently not PCI-compliant" and sends card payments to a PCI-compliant IVR. Ask each vendor whether card numbers stay out of recordings and transcripts.
How much does AI debt collection software cost?
Vendors charge in five ways: by the minute (Retell AI), in plans of minutes (Floatbot, from $99 a month billed yearly), per seat plus per AI interaction (NiCE, Genesys), on contingency (Skit.ai), or around agreed outcomes (Vector). Most collections specialists publish no price and quote after scoping. Compare cost per right-party contact and per kept promise, not per minute.
Do AI collection calls work as well as human callers?
The evidence is mixed. A German bank study of 3,448 borrowers found a 34.4-point increase in timely repayment when a conversation took place (Laudenbach and Siegel, Journal of Finance), and a Dutch field trial with algorithmic call selection delivered 23.4% higher repayment on 8.4% fewer calls (AEA RCT Registry); in both, people made the calls. A randomized test found AI "substantially less effective than human callers" (NBER Working Paper 33669), so judge a pilot on promises kept 30 days on, not only on promises taken.
How long does an AI collections pilot take?
Skit.ai runs 30 to 60 day pilots on live accounts and says they go live in 4 to 6 weeks; Salient says most deployments take 60 to 90 days from signed SOW to production. Vector's first two weeks on live accounts are supervised operation with transcript review. Set your baseline before the pilot starts, so the result is judged against your own collectors.
Read next
- AI debt collection: how it works and where it helps
- Debt collection compliance, the rules in more depth
- Debt collection software in 2026, the collection systems, servicing platforms and dialers an AI agent works beside.
- Alternatives, vendor by vendor: Skit.ai, Salient, Prodigal, Domu, InDebted, TrueAccord
- Talk to Vector: a collections call in your browser, from the opening to the hand-off to a person
Book a Vector demo: bring one portfolio segment, your baseline and the KPI you are judged on.
Sources read on 12 November 2026.
