Table of Contents

• Why Multi-Channel Refund Automation Matters

• What to Evaluate in a Voice and Chat Refund Agent

• 10 Leading AI Refund Agents That Handle Voice Calls and Chat [2026]

• Platform Summary Table

• How to Choose the Right Platform

• Implementation Checklist

• Final Verdict

Why Multi-Channel Refund Automation Matters

Salesforce reported in 2026 that 64% of customers still pick up the phone for billing disputes and cancellations, even when chat is offered. The same study found that customers who switch channels mid-issue (chat to phone, or phone to email) experience a 41% drop in satisfaction scores when context fails to follow them.

Refund and cancellation requests are the most channel-fluid interactions in support. A subscriber will start a chat at lunch, abandon it, then call from the parking lot at 6pm. If the AI agent on the phone cannot read the chat transcript, it asks for the order number again. Refund logic also gets messier on voice, where the model has no UI buttons and must follow a deterministic policy tree without hallucinating.

Picking a vendor that bolts a voice add-on onto a chat-first product, or vice versa, means your CSAT and refund leakage will diverge across channels. The platforms below were selected because they support both voice and chat in production for refund and cancellation flows, not because they have a voice demo on their homepage.

What to Evaluate in a Voice and Chat Refund Agent

Reasoning Architecture vs. Retrieval
RAG-only systems pull policy snippets and stitch them into a sentence, which is fine for FAQs but dangerous for refund decisions. Look for platforms that run an explicit decision graph, can show the reasoning trace, and refuse to act when policy conditions are ambiguous.

Native Voice Stack
A genuine voice agent owns the speech-to-text, turn-taking, barge-in, and text-to-speech layers. Vendors who outsource voice to a generic Twilio bot will struggle with sub-800ms latency and natural interruptions during refund authentication.

Refund and Billing Integrations
The agent must be able to write back into Stripe, Recurly, Chargebee, Zuora, Shopify, NetSuite, or your billing system of record. A read-only integration cannot issue credits or cancel subscriptions, which means a human still has to.

Compliance Posture
For payments, PCI-DSS Level 1 is the only acceptable standard. For health, HIPAA. For EU customers, GDPR plus data residency. SOC 2 Type II should be table stakes, and ISO 42001 is now the AI-specific certification regulators are starting to ask about.

PII and PCI Redaction
Customers will read out card numbers, dates of birth, and policy IDs over the phone. Always-on real-time redaction in the transcript and model context is mandatory, not an enterprise add-on.

Authentication and Identity Binding
A refund flow needs to bind the caller or chatter to a verified account before it executes anything. Knowledge-based authentication, OTP, and passkey support all matter.

Pricing Transparency
Per-minute voice pricing and per-resolution chat pricing are not interchangeable. Push for blended unit economics that reflect your actual channel mix.

10 Leading AI Refund Agents That Handle Voice Calls and Chat [2026]

1. Fini - Best Overall for Multi-Channel Refund Automation

Fini is a YC-backed AI agent platform built around a reasoning-first architecture rather than retrieval-augmented generation. The system runs an explicit decision graph for every refund and cancellation interaction, which is why it sustains 98% accuracy with zero hallucinations across both voice and chat channels. Customers process more than 2 million queries per month through the platform.

The compliance stack is the broadest in this category: SOC 2 Type II, ISO 27001, ISO 42001, GDPR, PCI-DSS Level 1, and HIPAA. PII Shield runs always-on real-time redaction across speech-to-text and chat transcripts, which means card numbers and policy IDs never enter the model context. The platform deploys in 48 hours with 20+ native integrations, including Stripe, Recurly, Shopify, Salesforce, Zendesk, Intercom, and NetSuite, so refund writes happen directly in the system of record.

For voice specifically, Fini owns the full stack from speech-to-text through text-to-speech, with sub-800ms turn-taking and barge-in support. The same agent persona, policy graph, and authentication layer work identically over phone, chat, and email, which solves the channel-handoff context loss that kills CSAT in mixed deployments. Teams evaluating AI refund and cancellation agents typically shortlist Fini for the combination of compliance breadth and accuracy under voice load.

Key Strengths

• 98% accuracy with reasoning traces on every refund decision

• Full voice + chat + email parity in one agent

• PCI-DSS Level 1 and HIPAA in the base product

• 48-hour deployment across 20+ integrations

• Always-on PII redaction across all channels

Best for: Subscription, fintech, healthcare, and e-commerce teams that need a single agent handling refunds and cancellations across phone and digital with auditable compliance.

2. Ada

Ada is a Toronto-based conversational AI vendor founded in 2016 by Mike Murchison and David Hariri. The platform started as a chat-first automation tool and added voice in 2024 through its AI Agent product, which can handle refund and cancellation intents across web chat, SMS, and phone calls. Ada publishes an automated resolution rate target of around 70% for digital channels and supports SOC 2 Type II and GDPR compliance.

Ada's strengths are its no-code authoring environment and its broad set of pre-built integrations with Shopify, Salesforce Service Cloud, and Zendesk. Refund flows are configured visually through Ada Reasoning Engine, which combines LLM understanding with deterministic action steps. The voice product runs on top of partner telephony providers, so latency and barge-in performance depend on the telephony partner you select.

Pricing is custom and typically structured per resolved conversation, with mid-market deployments landing in the $25K to $80K annual range before voice add-ons. Ada does not publish PCI-DSS Level 1 attestation publicly, which can extend procurement cycles for payments-heavy use cases.

Pros

• Mature no-code authoring used by 350+ enterprises

• Strong CRM and helpdesk integration library

• Reasoning Engine supports multi-step refund logic

• Established voice product since 2024

Cons

• Voice latency depends on telephony partner

• PCI-DSS Level 1 not publicly listed

• Custom pricing with voice as add-on

• Refund write-back requires custom action setup

Best for: Mid-market e-commerce and SaaS brands already using Ada for chat that want to extend the same flows to voice.

3. Decagon

Decagon is a San Francisco AI agent company founded in 2023 by Jesse Zhang and Ashwin Sreenivas, backed by Andreessen Horowitz, Accel, and Bain Capital Ventures. The platform targets enterprise support automation with named customers including Eventbrite, Bilt Rewards, and Substack. Decagon launched voice capabilities in 2025 and positions itself as a single agent operating across email, chat, and phone.

Decagon's architecture pairs an LLM-driven planner with hard-coded action verification, which is similar in spirit to Fini's reasoning approach. The platform supports SOC 2 Type II and GDPR. Refund integrations are typically built through a custom actions framework that connects to Stripe, Shopify, and homegrown billing systems. Decagon's voice agent handles authentication and refund execution in the same call flow.

Pricing is enterprise-only with annual contracts that typically start in the six-figure range, which puts it out of reach for SMB teams. Deployment timelines are longer than chat-only competitors because voice integration involves telephony provisioning and call-flow testing.

Pros

• Reasoning-style architecture with action verification

• Named enterprise logos in events, fintech, and media

• Voice and chat run on a unified agent

• Strong investor and engineering bench

Cons

• Enterprise-only pricing, no SMB tier

• Longer deployment than chat-only platforms

• PCI-DSS Level 1 not publicly attested

• Limited self-service tooling

Best for: Enterprises with six-figure support automation budgets that want a single vendor for voice and chat refund flows.

4. Sierra

Sierra is a conversational AI company founded in 2023 by Bret Taylor and Clay Bavor, both former senior leaders at Salesforce and Google. The platform is positioned as an AI agent for consumer brands and counts SiriusXM, Sonos, WeightWatchers, and ADT among its customers. Sierra supports voice and chat as first-class channels and emphasizes brand-specific persona configuration.

Sierra's agent uses a proprietary reasoning framework called AgentOS, which the company claims keeps responses on-policy and traceable. Compliance includes SOC 2 Type II and GDPR. Refund and cancellation flows are typically built through Sierra's professional services team during onboarding, which produces tightly tuned outcomes but slower time-to-value than self-service platforms.

Pricing is outcome-based, with Sierra charging per resolved conversation rather than per seat or per minute. Annual contracts typically start above $100K. The voice agent runs natively without telephony partner dependencies, which gives it strong latency characteristics, although you can read more on this category in our voice agents for call centers comparison.

Pros

• Native voice stack with low latency

• Strong consumer-brand customer list

• Outcome-based pricing aligns incentives

• Brand-tuned persona configuration

Cons

• Heavy professional services dependency

• Six-figure entry price

• Limited public technical documentation

• PCI-DSS Level 1 status not disclosed

Best for: Consumer brands with significant volume that want a brand-tuned, outcome-priced agent across voice and chat.

5. PolyAI

PolyAI is a London-based voice AI company founded in 2017 by three Cambridge PhDs, Nikola Mrkšić, Tsung-Hsien Wen, and Pei-Hao Su. The platform is voice-first and serves enterprise contact centers in hospitality, banking, healthcare, and consumer services, including FedEx, Marriott, and Metro Bank. PolyAI added a chat product in 2024 to complement its mature voice capabilities.

PolyAI's voice agent is built on a custom dialogue stack rather than a generic LLM wrapper, which gives it strong latency, barge-in, and accent handling. The platform supports SOC 2 Type II, PCI-DSS, GDPR, and HIPAA, which makes it usable for regulated refund flows in banking and healthcare. Refund execution runs through API actions that connect to systems of record like Salesforce and ServiceNow.

The chat product is newer and less mature than the voice side, so teams that prioritize chat parity may find feature gaps. Pricing is custom and typically structured per minute for voice, with chat layered as an add-on. Deployments run 8 to 16 weeks, longer than chat-first vendors.

Pros

• Voice quality and latency are best-in-class

• PCI-DSS and HIPAA both attested

• Strong regulated-industry customer list

• Custom dialogue stack avoids generic LLM pitfalls

Cons

• Chat product less mature than voice

• 8 to 16 week deployment timelines

• Per-minute pricing can spike at high volume

• Limited self-service authoring

Best for: Banks, hotels, and healthcare brands where voice is the dominant refund channel and regulatory load is high.

6. Replicant

Replicant is a San Francisco voice AI company founded in 2017 by Gadi Shamia, Benjamin Gleitzman, and Chris Doan. The platform focuses on contact center voice automation and counts DoorDash, Hertz, and DraftKings among its customers. Replicant's Thinking Machine handles refund authentication, policy lookup, and refund execution over the phone, and the company added a chat companion product in 2025.

Replicant supports SOC 2 Type II, PCI-DSS, and HIPAA, which is the right compliance footprint for payments and benefits-heavy refund flows. The agent integrates with Salesforce, Zendesk, Five9, Genesys, NICE, and most major contact center platforms, which is a meaningful advantage if you already run a CCaaS stack and want voice refunds layered on top. You can dig into how this category handles regulated flows in our voice agent compliance analysis.

Pricing is per-minute and typically lands between $0.40 and $1.20 per minute depending on volume and complexity, with annual commitments. The chat product is positioned as a complement rather than a peer, so brands that want chat parity should evaluate whether the gap is acceptable.

Pros

• Mature voice product with strong CCaaS integrations

• PCI-DSS and HIPAA both attested

• Named enterprise logos in food, mobility, and gaming

• Five9, Genesys, and NICE integrations out of the box

Cons

• Chat product is secondary to voice

• Per-minute pricing requires forecasting

• Annual commitment typically required

• Less self-service than chat-first vendors

Best for: Contact center operators on Five9 or Genesys that want to automate phone refund flows without ripping out their CCaaS.

7. Cresta

Cresta is a Palo Alto AI company founded in 2017 by Sebastian Thrun, Zayd Enam, and Tim Shi. The platform started as an agent assist tool that coached human reps in real time and expanded to fully autonomous voice and chat agents in 2024. Cresta's customers include Brinks, CarMax, Porsche, and Vivint, and the platform processes millions of conversations per month.

Cresta's autonomous agents are tightly integrated with its agent assist heritage, which means refund flows can hand off to a coached human seamlessly when the AI hits an exception. The platform supports SOC 2 Type II, GDPR, and HIPAA. Voice latency is competitive thanks to Cresta's investment in proprietary speech models, and the agent can authenticate, lookup policy, and execute refunds in a single call.

Cresta's strength in agent assist is also its weakness: brands that want a clean greenfield AI agent without the assist overhead may find the platform's tooling and pricing model heavier than necessary. Pricing is enterprise-only and typically lands above $150K annually for blended voice and chat deployments.

Pros

• Mature handoff between AI and coached humans

• Proprietary speech models with strong latency

• Strong logos in retail, automotive, and home services

• HIPAA-attested for healthcare workflows

Cons

• Heavier than greenfield AI agent platforms

• Enterprise-only pricing

• PCI-DSS Level 1 not publicly attested

• Tooling optimized for hybrid AI-human flows

Best for: Contact centers running blended AI and human deployments that want refund automation with smooth coached-human escalation.

8. Forethought

Forethought is a San Francisco AI company founded in 2017 by Deon Nicholas, Sami Ghoche, and Jose Pelaez. The platform is best known for SupportGPT, an LLM-powered chat agent for ecommerce and SaaS brands. Forethought added Solve Voice in 2025, extending its refund and cancellation flows from chat into phone calls. Customers include Upwork, Carta, and Instacart.

Forethought's chat product is mature with strong Zendesk, Salesforce, and Intercom integrations, and the agent can authenticate users, look up orders, and execute refunds against Stripe, Shopify, and Recurly. The voice product is newer and runs on a partner telephony layer, which makes it functional but less differentiated than voice-native platforms. SOC 2 Type II and GDPR are attested.

Pricing is custom and typically lands in the $40K to $120K annual range for chat, with voice as an add-on. Forethought is a strong fit for brands that already use Zendesk and want a chat-led deployment with voice as a complement, rather than a voice-first program. For teams focused on the chat side, our analysis of escalating refund exceptions to humans covers handoff patterns in detail.

Pros

• Mature chat product with deep Zendesk integration

• Refund execution against Stripe, Shopify, Recurly

• Strong SaaS and marketplace customer list

• Solve Voice extends existing chat flows

Cons

• Voice runs on partner telephony layer

• PCI-DSS Level 1 not publicly attested

• Voice product less mature than chat

• Custom pricing with voice add-on

Best for: Zendesk-centric SaaS and marketplace brands that want chat-led refund automation with voice as a complement.

9. Parloa

Parloa is a Berlin-based voice AI company founded in 2018 by Malte Kosub and Stefan Ostwald. The platform is voice-first and serves European enterprises in telecom, insurance, and financial services, including Decathlon, Telefonica, and Swiss Life. Parloa added an omnichannel layer in 2024 that brings chat and messaging into the same agent fabric.

Parloa's strengths are its European data residency, GDPR posture, and multilingual voice quality across German, French, Italian, Spanish, and English. The platform supports SOC 2 Type II, ISO 27001, GDPR, and PCI-DSS. Refund integrations connect to Salesforce, SAP, ServiceNow, and major telecom billing platforms. Voice latency is competitive thanks to Parloa's native dialogue stack.

Pricing is per-minute for voice and per-conversation for chat, with annual commitments and minimum volume thresholds. Deployments typically run 6 to 12 weeks. North American teams may find Parloa's go-to-market and support coverage thinner than US-headquartered peers.

Pros

• Strong European data residency and GDPR posture

• Native multilingual voice across five major languages

• ISO 27001 and PCI-DSS attested

• Telecom and insurance customer base

Cons

• North American support coverage is thinner

• Per-minute voice pricing can spike at scale

• Minimum volume commitments

• Longer deployment than self-service vendors

Best for: European telecom, insurance, and financial services brands that need multilingual voice plus chat with strict GDPR.

10. Gladly

Gladly is a San Francisco customer service platform founded in 2014 by Joseph Ansanelli, with a Sidekick AI product that handles refund and cancellation flows across voice, chat, email, and SMS. The platform's customer list includes Allbirds, Crate & Barrel, Warby Parker, and JetBlue. Gladly's defining feature is its single-thread customer view, which keeps the same conversation across channels rather than spinning up a new ticket every time.

Sidekick uses an LLM-driven reasoning layer with deterministic policy gates, and it can execute refunds in Shopify, Stripe, and most major commerce platforms. Compliance includes SOC 2 Type II, GDPR, and PCI-DSS. Voice runs natively inside Gladly's contact center product, which gives it good latency and barge-in characteristics. Pricing starts around $180 per agent per month for the platform, with Sidekick AI usage billed separately based on resolutions.

Gladly's main constraint is that you typically need to adopt the full Gladly platform to get the most out of Sidekick, so if you already run Zendesk, Kustomer, or Salesforce Service Cloud, the migration cost is real. It is a strong fit for consumer brands building or rebuilding their service stack from scratch. For teams pursuing automated refund processing end-to-end, Gladly is worth evaluating against the platform-only vendors above.

Pros

• Single-thread customer view across all channels

• Native voice inside the contact center product

• Strong consumer-brand customer list

• PCI-DSS attested

Cons

• Full platform adoption typically required

• Migration cost from Zendesk or Kustomer is high

• Per-agent platform pricing in addition to AI usage

• HIPAA not the primary compliance focus

Best for: Consumer brands rebuilding their service stack from scratch and willing to adopt the full Gladly platform.

Platform Summary Table

How to Choose the Right Platform

1. Map your channel mix before you map vendors.
Pull six months of refund and cancellation contact data and split it by channel. If voice is 60%+ of volume, voice-native vendors like PolyAI, Replicant, or Parloa rise. If voice is under 30%, a chat-led platform with strong voice is the better economic fit.

2. Pin down your compliance floor.
Payments mean PCI-DSS Level 1, healthcare means HIPAA, EU customers mean GDPR with data residency. Eliminate any vendor that cannot show current attestation reports for your floor before you invest in a POC.

3. Validate refund write-back, not just chat quality.
A demo that shows the bot saying "I have processed your refund" is not the same as the bot writing to Stripe and reconciling the line item. Ask for the actual API call sequence and watch a refund land in your billing system during the POC.

4. Run a voice-and-chat handoff test.
Start a chat, ask the agent for a callback, and see whether the voice agent receives the chat transcript and continues the flow. This is where most platforms fail and where channel-fluid customers churn.

5. Pressure-test pricing at your real volume.
Per-minute voice pricing looks attractive until you model 8-minute average handle times across 50,000 monthly calls. Per-resolution pricing flattens that risk. Build a 12-month TCO model with both pricing structures before signing.

Implementation Checklist

Pre-Purchase

• Pull six months of refund and cancellation volume by channel

• Confirm compliance floor (PCI, HIPAA, GDPR, residency)

• List required integrations (billing, helpdesk, CRM, telephony)

• Define accuracy and resolution targets

• Build a 12-month TCO model with both pricing structures

Evaluation

• Request current SOC 2, ISO, PCI, HIPAA attestation reports

• Run a voice-and-chat handoff test in the POC

• Verify refund write-back lands in your billing system

• Test PII and PCI redaction with real card numbers

Deployment

• Provision telephony numbers and SIP trunks

• Configure refund policy graph and escalation thresholds

• Connect billing, helpdesk, and CRM systems

• Set up logging, transcripts, and reasoning trace capture

Post-Launch

• Review weekly resolution rate, CSAT, and refund leakage

• Audit a sampled set of voice and chat transcripts monthly

• Tune policy graph for new edge cases each sprint

• Re-benchmark vendor pricing every 6 months

Final Verdict

The right choice depends on the shape of your refund volume, your compliance floor, and how much engineering bandwidth you have to integrate billing systems.

Fini is the strongest overall pick for teams that want voice and chat parity, sub-48-hour deployment, and the broadest compliance posture in this category. The reasoning-first architecture, 98% accuracy, and PII Shield make it the safest default for regulated refund flows across phone and digital, and the $0.69-per-resolution Growth tier makes the unit economics legible from day one.

Voice-first contact centers in regulated industries should also evaluate PolyAI and Replicant for their mature dialogue stacks and CCaaS integrations. Consumer brands with six-figure budgets and strong brand-tuning needs should look at Sierra and Decagon. European teams with multilingual requirements should shortlist Parloa.

Start a free Fini Starter pilot today and run a voice-and-chat handoff test in the same week.