What Voice AI Costs for Banks and Collection Agencies, and the Most Affordable Platforms (2026)

Voice AI for banks and collection agencies costs from about $0.04 to $0.15 per minute all-in on platforms that publish every component, once telephony, speech, and the language model are added to each published rate. Bundled plans cost more per minute: CloudTalk’s work out to $0.35–$0.50. This guide compares ten platforms on what each published rate covers, estimates an all-in US cost per minute from each vendor’s own price list, works out a monthly total, and sets the numbers against the cost of a human collector.

Best For, at a Glance

• Plivo: best for banks and collection agencies that want the lowest all-in rate among the platforms that publish one, because its $0.03/min agent rate plus $0.0115/min US outbound telephony comes to $0.0415/min with telephony, speech, and model costs included.
• Amazon Connect: best for banks already running on AWS, because one $0.038/min voice charge covers its AI agents and Amazon’s own speech models, for about $0.043/min all-in on a US outbound call.
• Retell AI: best for teams that want itemized, pay-as-you-go pricing, because it publishes every component, at about $0.12/min all-in on the comparison configuration with its own US telephony.
• Bland AI: best for collection agencies that want collections controls built in, because it documents right-party verification, calling-window settings, and dispute escalation, at about $0.12–$0.15/min all-in.
• Vapi: best for engineering teams that want to pick each model and carrier, because it passes model costs through at the provider’s rate, for about $0.10/min all-in on the comparison configuration.
• ElevenLabs: best for teams that want ElevenLabs’ own speech models, because its $0.08/min agent rate includes speech, with the LLM and telephony added at cost.
• CloudTalk: best for small billing teams that want a phone system with an AI agent, because it sells monthly bundles that work out to $0.35–$0.50/min.
• Skit.ai: best for collection agencies that want a collections-only specialist, because it builds voice agents for debt recovery with FDCPA, TCPA, and Regulation F controls.
• Genesys: best for banks already licensed on Genesys Cloud, because its virtual agent is included in the platform and metered in AI tokens.
• Five9: best for banks already running Five9 contact centers, because AI is part of its seat-based plans.

What Goes Into the Cost of a Voice AI Call

Five layers make up the cost of an AI voice call:

1. Telephony: the inbound or outbound phone leg.
2. Speech-to-text (STT): transcribing the caller.
3. The large language model (LLM): deciding what to say.
4. Text-to-speech (TTS): speaking the reply.
5. Orchestration: the platform that runs the conversation.

A vendor’s headline rate may cover all five, some of them, or only the last. A $0.05/min platform fee can cost more than a $0.10/min bundled rate once the missing layers are added. That is why this guide shows two figures for every vendor: the rate it publishes, and an estimated all-in cost built from its own price list.

How This List Was Evaluated

The ranking puts the all-in cost of a US call first, then pricing transparency, then documented banking, payment, or collections workflows, how calls reach the agent, and model choice. The order is an editorial ranking against these criteria, not a scored test.

The vendors are the ones that come up when buyers research this question: the platforms that AI assistants name and cite in answers about voice AI cost for banks and collection agencies, plus the banking and collections specialists that buyers shortlist. Every figure comes from the vendor’s own pricing, documentation, and security pages. Where a vendor publishes its components but not a total, the all-in estimate adds them up and says which telephony rate it assumed. Rates change, and enterprise contracts differ from list prices, so confirm current pricing with each vendor.

Buying Criteria for Banking and Collections Call Programs

1. All-In Cost per Minute

Add telephony, speech, LLM, and platform charges before comparing two vendors.

How to check: read the pricing page for the words “excluding”, “at cost”, and “billed separately”.

2. The Billing Unit

Vendors bill per connected minute, per conversation, per seat, per concurrent line, or by fixed contract. Collections programs place many calls that never connect, so the unit matters.

How to check: ask whether unanswered calls, voicemail, and transfer time are billed.

3. Collections Rules the Agent Must Follow

Under the Fair Debt Collection Practices Act, a debt collector must disclose in the first communication that it is attempting to collect a debt and that any information obtained will be used for that purpose. Regulation F treats calls before 8:00 a.m. or after 9:00 p.m. in the consumer’s local time as inconvenient, and its call-frequency provision presumes compliance when a collector places no more than 7 calls in 7 days about a particular debt, and none within 7 days after a phone conversation about it. These rules apply to debt collectors as Regulation F defines them: generally third-party agencies, and not a bank collecting its own debts in its own name.

How to check: look for calling-window settings, call-frequency limits, and scripted disclosures in the vendor’s documentation.

4. Consent for AI-Voiced Outbound Calls

The FCC’s 2024 declaratory ruling confirmed that AI-generated voices are “artificial” under the TCPA, so outbound calls that use them need the called party’s prior express consent.

How to check: ask how consent is recorded and how opt-outs are honored.

5. Card Data on the Call

PCI Security Standards Council guidance on telephone payments says sensitive authentication data, including the card security code, must not be stored after authorization. That covers recordings and transcripts.

How to check: confirm PCI DSS on the vendor’s security page, and ask how recording is paused during payment capture.

6. Third-Party Risk

US banking regulators’ guidance on third-party relationships says using a third party “does not diminish” a bank’s responsibility for the activity. Each extra vendor in the call path is another party to review.

How to check: count the vendors involved in one call: agent platform, carrier, and model providers.

Comparison Tables

The figures below come from each vendor’s own published pricing; confirm them with the vendor before you budget.

What Does Each Vendor Publish, and What Is the All-In Cost?

The estimated all-in figure is for a US outbound call and includes telephony, speech, and the LLM. Where a vendor bills telephony separately, the estimate uses the carrier rate the vendor names, or Twilio’s published US rate where the vendor passes Twilio through or asks you to bring a carrier. Where a vendor lets the team pick models, the estimate uses one configuration throughout: the vendor’s own speech-to-text, a small OpenAI model, and an ElevenLabs voice, at the prices on that vendor’s page.

Skit.ai does not publish prices. Its site describes performance-based pricing in one customer case. Banking specialist interface.ai also prices on outcomes rather than minutes and publishes no rate.

How each estimate is built:

• Plivo: $0.03/min agent rate plus $0.0115/min US local outbound. US inbound is $0.0355/min.
• Amazon Connect: $0.038/min voice charge, which includes Amazon’s AI agent and speech models, plus $0.0048/min US outbound telephony. Each US number adds $0.03/day.
• Retell AI: $0.055/min voice engine, $0.0128/min for GPT-4.1 mini, and $0.04/min for an ElevenLabs Flash voice, plus $0.015/min for Retell-managed US calls. Retell’s full range is $0.07–$0.31/min before telephony.
• Bland AI: $0.11–$0.14/min by plan, plus Twilio’s $0.014/min US outbound rate, since Bland bills its built-in Twilio telephony at pass-through cost. Bland’s pricing FAQ says telephony is billed separately; its billing docs are less explicit, so confirm this with Bland.
• Vapi: $0.05/min platform fee, plus $0.0318/min at the low end of Vapi’s own calculator for Deepgram speech-to-text, a small OpenAI model, and an ElevenLabs voice (the calculator prices providers rather than individual models, and runs to $0.0789/min), plus the $0.014/min Twilio outbound rate Vapi lists.
• ElevenLabs: $0.08/min, plus $0.014/min for a Twilio line. ElevenLabs prices the LLM per million tokens, not per minute, so no complete per-minute total can be stated.
• CloudTalk: bundles of $99/month for 200 minutes and $349/month for 1,000 minutes; larger packages are quoted. CloudTalk does not say whether carrier minutes are inside the bundle.

What Would 10,000 US Outbound Minutes a Month Cost?

Each total is 10,000 minutes multiplied by the estimated all-in rate, plus any fixed plan fee. Phone number rental is left out; it adds a few dollars a month on every platform.

The ElevenLabs figure uses its $299/month Scale plan, which includes 3,738 minutes, with the other 6,262 minutes at $0.08/min and $0.014/min of telephony on all 10,000. Bland’s Build plan comes to $1,639 at this volume ($0.134/min plus $299/month), so Start is cheaper here.

How Do Calls Reach the Agent, and Can the Team Choose Its Models?

Skit.ai does not publish how calls reach its agents or which models it uses.

Which Card-Data and Security Programs Does Each Vendor Document?

Each “Documented” entry links to the vendor page that states it.

Amazon Connect and Five9 are not in this table; confirm their programs for the voice agent product with each vendor.

How Voice AI Compares With the Cost of a Human Collector

The US Bureau of Labor Statistics reports a median wage of $22.61/hour for bill and account collectors and $21.53/hour for customer service representatives in May 2025. That is about $0.38 and $0.36 per minute of paid time, before benefits, management, and the time between calls.

Every usage-based all-in estimate in this guide is below that: the highest, $0.154/min, is under half of it, and the lowest, $0.0415/min, is about a ninth of it. CloudTalk’s smallest bundle, at $0.50/min, is above it. The comparison is not like for like: an agent handles routine reminders and payment arrangements, while disputes and hardship conversations still need a person. The saving comes from moving the routine share of calls, not all of them.

The 10 Voice AI Platforms Compared on Price

1. Plivo

Best for: banks and collection agencies that want the lowest published all-in rate.

Plivo is a full-stack Voice AI platform: the agent runs on Plivo’s own carrier network, so telephony and the agent are priced by one provider. Teams describe an AI voice agent in plain English with Vibe Agent and review the flow in AI Agent Studio, or build in code with audio streaming, speech-to-speech models, or full-code frameworks such as LiveKit and Pipecat. The team can use Plivo’s models or bring its own at every layer.

Key strengths

• The lowest estimated all-in rate among the platforms that publish one, with no monthly platform fee on pay-as-you-go
• Concurrent calls, simulation testing, transfer to human, and the knowledge base included in the agent rate
• No model lock-in: choose or bring the speech-to-text, LLM, and text-to-speech models
• Compliance set: HIPAA with BAA, SOC 2, PCI DSS, and GDPR, detailed on the security page

Limitations

• The telephony rate changes with call direction and destination, so the all-in figure differs outside the US.
• Pay-as-you-go covers the United States and India. Wider country coverage, BAA, and SSO are part of the Enterprise plan, which starts at $1,000/month.
• Plivo does not publish a named bank or collection agency reference.

Pricing

The AI agent rate is $0.03/min, which covers voice models, transcription, and orchestration. US local outbound telephony adds $0.0115/min, for $0.0415/min all-in. US local inbound is $0.0055/min, for $0.0355/min all-in. See voice pricing for other destinations.

2. Amazon Connect

Best for: banks already running their contact center or data on AWS.

Amazon Connect is AWS’s cloud contact center. Its AI agents and self-service are billed inside a single per-minute voice charge that also covers Amazon’s own speech and conversational models. Amazon sells the phone numbers and minutes, so the call, the agent, and the contact center sit with one vendor. Third-party speech models can be configured at that provider’s cost.

Key strengths

• One voice charge covers the AI agent and Amazon’s speech models
• Phone numbers and telephony from the same vendor
• Fits banks that already run on AWS

Limitations

• It is a full contact center platform, so setup is heavier than a standalone agent
• Third-party speech models, knowledge bases, and some AI tooling are billed separately
• Each phone number carries a daily charge

Pricing

$0.038/min for voice, including AI agents and self-service. US inbound telephony on a direct number is $0.0022/min and US outbound is $0.0048/min, with $0.03/day per US number. That is about $0.0402/min all-in inbound and $0.0428/min outbound. Outbound campaign calls are billed at $0.045/min instead of $0.038/min.

3. Retell AI

Best for: teams that want itemized, pay-as-you-go pricing.

Retell AI publishes each component of its price: platform, voice, and LLM. Agents are built in a no-code dashboard or through an API, and its documentation describes retention settings for transcripts and recordings. It publishes lending and collections customer stories.

Key strengths

• Every cost component is published, including its own US telephony rate
• No-code builder and API
• Documents SOC 2, HIPAA with BAA, and GDPR

Limitations

• PCI DSS is not present on its public compliance pages
• The rate varies more than fourfold with the LLM selected
• States that it does not operate services within the European Union

Pricing

$0.07–$0.31/min depending on the LLM and voice. On the comparison configuration, GPT-4.1 mini with an ElevenLabs Flash voice, the agent is $0.108/min; Retell-managed US calls add $0.015/min, for $0.123/min all-in. Bringing your own carrier has no Retell charge.

4. Bland AI

Best for: collection agencies that want collections controls built into the platform.

Bland AI documents debt collection workflows directly: payment reminders, right-party verification, payment-plan discussions, promise-to-pay capture, and dispute escalation, with calling-window settings and consent tracking. It runs its own speech and language models, so the customer cannot select an outside model.

Key strengths

• Collections is a documented use case with specific controls
• The per-minute rate includes speech and the LLM
• Documents SOC 2, PCI DSS, HIPAA with BAA, and GDPR

Limitations

• Telephony is billed on top, and Bland does not publish a per-minute carrier figure
• Customer model choice is not supported
• The lower per-minute rates carry a monthly plan fee

Pricing

$0.14 per connected minute on the Start plan, and $0.12/min plus $299/month on the Build plan. Bland’s billing documentation also lists a Scale plan at $0.11/min plus $499/month. With Twilio’s $0.014/min US outbound rate passed through, that is about $0.124–$0.154/min all-in.

5. Vapi

Best for: engineering teams that want to pick each model and carrier.

Vapi is a developer platform. The team selects or brings the speech-to-text, LLM, text-to-speech, and carrier, and Vapi charges a platform fee on top. That suits a bank that has already standardized on particular model providers.

Key strengths

• Model choice at every layer
• Model costs passed through at the provider’s rate, shown in a published calculator
• Documents SOC 2, PCI DSS, HIPAA with BAA, and GDPR

Limitations

• The $0.05/min platform fee is roughly a third to a half of the real per-minute cost
• No published banking or collections customer
• The BAA is a $2,000/month add-on

Pricing

$0.05/min platform fee. Vapi’s own calculator adds $0.0318–$0.0789/min for Deepgram speech-to-text, an OpenAI model, and ElevenLabs voices, which puts the agent at $0.082–$0.129/min. The comparison uses the low end, $0.082/min; Twilio outbound adds $0.014/min, for $0.096/min all-in, or up to $0.143/min with the dearer components.

6. ElevenLabs

Best for: teams that want ElevenLabs’ own speech models on servicing and collections calls.

ElevenLabs builds agents on its own speech-to-text and text-to-speech models. Its financial-services page covers account support, lending, outbound servicing, and collections. The customer chooses the LLM or brings its own; ElevenLabs does not sell telephony.

Key strengths

• Speech-to-text and text-to-speech included in the agent rate
• Financial-services use cases include collections
• Documents SOC 2, ISO 27001, PCI DSS, HIPAA with BAA, and GDPR

Limitations

• LLM usage is billed at cost per million tokens, so the per-minute total depends on the model and how much the agent says
• No telephony; a carrier is a separate purchase
• A BAA is an Enterprise plan item

Pricing

Additional call minutes cost $0.08/min on every plan, with plans from $0/month to $990/month. A Twilio line adds $0.014/min, for $0.094/min before the LLM.

7. CloudTalk

Best for: small billing teams that want a business phone system with an AI agent included.

CloudTalk sells a cloud phone system with AI voice agents. It sells its own numbers and also accepts a customer’s carrier. Its entry plan is for inbound calls; the higher plan adds outbound calling and is positioned for billing, invoices, collections, and payments.

Key strengths

• Simple monthly bundles
• Phone system, numbers, and agent from one vendor
• Documents SOC 2, ISO 27001, PCI DSS, HIPAA with BAA, and GDPR

Limitations

• The bundles work out to $0.35–$0.50/min, above most usage-based rates in this comparison
• The page does not say whether carrier minutes are inside the bundle
• Does not publish a bank, credit union, or collections customer result for its AI agent

Pricing

$99/month for 200 minutes on the inbound plan. The plan that covers billing and collections workflows is $349/month for 1,000 minutes. Larger packages are quoted.

8. Skit.ai

Best for: collection agencies that want a collections-only specialist.

Skit.ai builds voice agents for debt collection and says it works with more than 120 collection teams. Its compliance material covers the FDCPA, the TCPA, and Regulation F’s 7-in-7 call cap, with tokenized card capture and card numbers suppressed from transcripts. Skit.ai publishes an unnamed subprime auto lending case in which delinquency fell from 35% to 17%.

Key strengths

• Built only for collections, with collections compliance controls
• Named collections customers, including Pollack and Rosen and Revenue Enterprises
• Documents SOC 2, ISO 27001, PCI DSS, and HIPAA

Limitations

• No price is published
• Positioned for collections rather than general banking service
• Does not publish a choice of models

Pricing

Skit.ai does not publish prices. One of its customer cases describes performance-based pricing, where Skit.ai earns more as more is collected.

9. Genesys

Best for: banks already licensed on Genesys Cloud.

Genesys offers a virtual agent inside Genesys Cloud, its contact center platform. The virtual agent is part of the platform licence, and usage is metered in AI Experience tokens. Genesys lists bank and credit union customers including Kiwibank, TSB Bank, and M&T Bank.

Key strengths

• Virtual agent included in the contact center platform
• Named bank and credit union customers
• Documents SOC 2, ISO 27001, PCI DSS, HIPAA, and GDPR

Limitations

• Token prices are not published, so a per-minute cost cannot be calculated
• Seat prices assume an annual commitment
• Requires the Genesys Cloud platform

Pricing

Seats from $75/user/month to $240/user/month, billed annually. AI Experience tokens are included up to an allowance and otherwise priced by quote.

10. Five9

Best for: banks already running Five9 contact centers.

Five9 is a cloud contact center provider. Its contact center plans bundle AI features, and its AI agents are sold alongside the contact center rather than as a standalone voice agent.

Key strengths

• AI features included in the contact center seat
• One vendor for agents, routing, and AI

Limitations

• No published seat or AI agent prices
• Seat pricing suits contact centers, not usage-based programs
• Every plan is priced by quote

Pricing

Five9 does not publish seat or AI agent prices on its site; every plan is quoted, so confirm current pricing with Five9.

Which Platform Is Most Affordable for Which Buyer

• Lowest published all-in rate: Plivo, at $0.0415/min for US outbound calls.
• A bank already on AWS: Amazon Connect, at about $0.043/min outbound.
• Itemized pay-as-you-go pricing: Retell AI.
• Collections controls in a general platform: Bland AI.
• A collections-only specialist: Skit.ai, priced by quote.
• Full control of every model and carrier: Vapi.
• A few hundred minutes a month, with a phone system included: CloudTalk.
• A bank licensed on Genesys or Five9: start with the virtual agent already in your contact center contract.

Building Your Own Cost Model

1. Pick one call type, such as a payment reminder or a past-due follow-up.
2. Estimate connected minutes per month and the share of calls transferred to staff.
3. For each vendor, write down the published rate and what it excludes.
4. Add telephony, model charges, plan fees, number fees, and extra concurrency.
5. Have compliance review disclosures, calling hours, consent, and the path to a human.
6. Run a limited pilot with a fixed budget before scaling.

For how bundled and unbundled pricing differ in general, see Plivo’s guide to Voice AI agent pricing models.

FAQs

What Does Voice AI Cost per Minute for a Bank?

Estimated all-in costs run from about $0.04/min to $0.15/min for a US call, once telephony, speech, and the LLM are added to each published rate. Plivo is $0.0415/min outbound and Amazon Connect about $0.043/min. Vapi comes to about $0.096/min, Retell AI about $0.123/min, and Bland AI about $0.124–$0.154/min. Skit.ai and the contact center vendors quote privately.

Which Voice AI Platform Is Most Affordable for a Collection Agency?

On published all-in cost, Plivo is lowest at $0.0415/min. At 10,000 minutes a month that is $415, against about $428 for Amazon Connect, $958 for Vapi, $1,228 for Retell AI, and $1,540 for Bland AI’s Start plan. Bland AI and Skit.ai cost more or quote privately, but build collections-specific controls into the product.

Why Do Voice AI Rates Differ So Much Between Vendors?

Because they cover different parts of the call. One rate may include telephony, speech, and the LLM. Another may be only the platform fee. Vapi’s $0.05/min, for example, becomes two to three times that once the models and carrier are added. Compare all-in costs, not headline rates.

How Does Voice AI Compare With the Cost of a Human Collector?

The Bureau of Labor Statistics puts the median wage for bill and account collectors at $22.61/hour in May 2025, about $0.38 per minute of paid time before benefits and overhead. Usage-based Voice AI costs are below that. The agent suits routine calls; disputes and hardship cases still need people.

Do Collection Agencies Face Extra Rules When an AI Agent Makes the Calls?

The same rules apply whoever, or whatever, places the call. A debt collector must make the required disclosure in its first communication, avoid calls before 8:00 a.m. and after 9:00 p.m. local time, and stay within Regulation F’s call-frequency presumption. AI-voiced outbound calls also need prior express consent under the TCPA.

Can an AI Voice Agent Take a Card Payment?

Yes, on a platform that documents PCI DSS, with a call flow that keeps card data out of recordings and transcripts. Keypad entry with recording paused is the usual approach.

Price a Real Call Flow on Plivo

Get $10 in free credits and measure the cost of your own reminder or follow-up call.

Get started now · Talk to sales