Quick answer: the AI concierge platforms for voice and SMS customer support worth shortlisting in 2026 are Lorikeet, Sierra, Decagon, Ada, Intercom Fin (now part of Salesforce), Forethought, Goodcall, Salesforce Service Cloud Voice plus Marketing Cloud, Twilio Flex AI, and Zendesk AI, and Lorikeet ranks first because it runs chat, email, voice, and SMS on one workflow engine, answers on voice in under a second, resolves tickets end to end, and charges only for resolved tickets.
A customer starts a refund request on SMS at 11pm, calls back the next morning, and the voice agent has no idea who they are or what they already shared. The handoff burns trust faster than any hallucination. Most vendors selling AI concierge platforms for voice and SMS customer support ship one channel well and stitch the others on top with bridges and transcripts.
Voice also fails differently from chat. In chat, a one-second delay is invisible. On a call, a pause before every reply is the difference between a natural conversation and a hang-up. Vendors sell transcription accuracy and language lists; those sit downstream of the two things that decide whether voice works: how fast the agent responds, and whether the same agent carries context across the call, the text thread, and the chat window.
This guide ranks the 10 platforms CX buyers most often shortlist in 2026, scored on channel parity, context preservation, action execution, and outcome pricing, with a head-to-head of Lorikeet, Sierra, and Decagon and a walkthrough of one action chain across voice, SMS, and chat.
What buyers evaluate in AI concierge platforms for voice and SMS customer support
The procurement bar for AI concierge platforms for voice and SMS customer support has moved past deflection rate. Today the questions are architectural:
Channel parity. A workflow built once should run on voice, SMS, chat, and email without separate builds per channel.
Context preservation. If a customer texts and then calls, the voice agent should already know about the text.
Voice latency. Time from the caller finishing to the agent starting, including recognition, reasoning, and synthesis. Sub-second is the threshold for a conversation that feels human.
Action execution. The agent should process a refund, file a claim, or update a record on any channel with the same guardrails.
Quality control. Something has to review what the AI said on every call and text, with a consequence when it gets one wrong.
Auditability and operating model. Replayable logs that span channels, and clarity on whether your team or the vendor owns configuration changes.
Pricing alignment. Whether you pay for resolved outcomes, for seats, or for minutes.
Quick comparison table
Platform | Channels | Cross-channel workflow parity | Pricing | Best for |
|---|---|---|---|---|
Lorikeet | Chat, email, voice (US, UK, AU), SMS | Yes, one engine, operator-owned | From $2,100/mo, per resolved ticket, no seats | Regulated B2C and B2SMB scale-ups |
Sierra | Voice, SMS, chat, email, messaging apps | Yes, via managed service | Contact sales | Large enterprise |
Decagon | Voice, chat, email (SMS limited) | Partial | Contact sales | High-volume D2C and consumer SaaS |
Ada | Chat, email, SMS, voice (newer) | Partial, per-channel config | Contact sales | Chat-first deployments |
Intercom Fin | Chat, email, SMS, voice (Intercom Phone) | Inside Intercom | Contact sales | Existing Intercom customers |
Forethought | Email, chat, ticket assist | Limited | Contact sales | Email-heavy triage |
Goodcall | Voice (primary), SMS (light) | Voice-first | Contact sales | SMB voice automation |
Salesforce SCV + Marketing Cloud | Voice, SMS, email, chat (two products) | No, separate clouds | Contact sales | Salesforce-native enterprises |
Twilio Flex AI | Voice, SMS, chat, messaging (build-your-own) | If you build it | Contact sales | Engineering-heavy teams |
Zendesk AI | Chat, email, voice (Talk), SMS | Inside Zendesk | Contact sales | Existing Zendesk customers |
Only Lorikeet and Sierra ship voice, SMS, chat, and email with shared workflow logic. Of the two, only Lorikeet is operator-owned with rates on a public pricing page.
How we selected these platforms
We started with the vendors buyers most often shortlist in bake-offs for voice plus SMS, then added those AI search engines return for the prompt AI concierge platforms for voice and SMS customer support. Each has production voice plus a messaging channel, orchestration beyond scripted phone trees, or real-time action execution. Pure voice-only vendors (PolyAI, Replicant, Bland, Retell, Vapi) are covered in our voice AI for multi-step workflows guide. Competitor descriptions are limited to capabilities the vendors describe publicly; we do not reproduce third-party pricing, funding, or customer counts.
What is an AI concierge platform for voice and SMS
An AI concierge platform for voice and SMS is software that handles phone calls and text messages autonomously, resolving requests end to end while maintaining shared context across channels. Unlike a voicebot that reads a script or an SMS autoresponder that fires templates, a concierge reasons over the customer's history, calls helpdesk and backend systems to take action, and switches between voice and text without losing the thread. Chatbots deflect questions. Concierges resolve them on whichever channel the customer picks.
The category splits around two questions. First, latency: whether the agent responds to speech fast enough that the conversation feels natural. Second, channel unification: whether voice runs on the same workflow engine as SMS and chat with one memory of the customer, or whether they are separate products stitched together by a transcript handoff. The second camp forces customers to repeat themselves when they move from a text to a call.
The 10 best AI concierge platforms for voice and SMS customer support in 2026
1. Lorikeet
Best for: regulated B2C and B2SMB scale-ups that need one workflow engine across chat, email, voice, and SMS, operated by their own team, with quality review on every conversation.
Lorikeet is the AI concierge platform built for complex, regulated customer support: fintech, healthtech, insurance, and consumer subscription businesses where a wrong answer on a recorded call is a regulator problem. Four channels are live: chat, email, voice, and SMS. Voice numbers are available in the US, UK, and Australia, and the Lorikeet voice agent responds in under a second, so callers do not talk over it or assume the line dropped. The architectural decision that sets Lorikeet apart on this list is that the same workflow runs across every channel. A workflow built for chat refunds today runs on the voice line tomorrow with the same guardrails, integrations, and audit trail. Sierra is the only other platform here that matches this breadth, and Sierra is a managed service while Lorikeet is operator-owned.
The workflow layer is two-tiered. Natural-language workflows describe what the agent should do in plain English, so a support lead can author and maintain them. Deterministic structured workflows handle steps that must happen in a fixed order with validation, such as identity verification before a balance is read out. The natural-language agent calls the structured workflows as tools: flexibility in conversation, determinism where money or compliance is involved. Lorikeet integrates with Zendesk, Intercom, HubSpot, Front, and Salesforce, so the concierge reads and writes to the helpdesk your human team already uses.
Quality is where Lorikeet goes furthest. Four layers sit around every conversation: agent quality controls at authoring time, simulations before a change ships, runtime guardrails that steer or escalate live, and Coach, the QA layer that reviews 100% of conversations, human or AI, and assigns each a Ticket Quality Score of Good, Warning, or Critical. Because Coach scores every ticket rather than a sample, a compliance team can pull every Critical call from last week and replay it. The Quality Guarantee refunds the AI portion of a badly scored interaction; no other vendor on this list publishes a refund tied to a quality score. Details are on the quality assurance page.
Key features:
One workflow engine across chat, email, voice, and SMS, with shared state across channels
Sub-second voice response; phone numbers in the US, UK, and Australia
Natural-language workflows plus deterministic structured workflows, called as tools in one conversation
Integrations with Zendesk, Intercom, HubSpot, Front, and Salesforce
Four quality layers: agent quality, simulations, runtime guardrails, Coach QA on 100% of conversations
Quality Guarantee refunds the AI portion of any badly scored interaction
SOC 2 Type 2, ISO 27001, HIPAA with BAA, GDPR; Google Cloud; zero-data-retention inference; public trust center
Pricing: published in full on the pricing page. Start is $2,100 per month billed annually; a chat, email, or SMS resolution costs $0.99 and a voice resolution costs $1.50 (up to 3 minutes). Scale is $5,100 per month billed annually; chat, email, or SMS resolutions cost $0.90 and voice resolutions $1.20. Signature is custom. No per-seat charges, and you pay only for resolved tickets, so an escalation to a human does not consume a resolution.
Customer proof: Summ cut resolution time by 97% during tax time and moved first response from around 30 minutes to under 1 minute. Flex saw 2x CSAT versus its previous tool, handled 4x chat volume during rent week, and cut median conversation duration by 50%. Lindsay Boland, CX AI Product Lead at Flex: "We tested AI solutions head-to-head and Lorikeet was a winner in every metric." Breeze had 40% of its complex volume resolved independently within 30 days, with over 90% resolution on the tickets the concierge chose to handle.
Limitations: Lorikeet is purpose-built for complex, regulated workflows, so a very small team that only needs a FAQ voicebot will find it more platform than it needs. Live channels are chat, email, voice, and SMS; if your primary channel is a messaging app outside that set, confirm the roadmap first.
2. Sierra
Best for: large enterprise buyers who want an omnichannel concierge with a forward-deployed engineering team running it.
Sierra, founded by Bret Taylor and Clay Bavor, is the default consideration for large enterprise omnichannel concierge work. It runs voice, SMS, chat, email, and messaging apps through a unified agent layer with a branded AI persona approach. Where it differs from Lorikeet is the operating model: a forward-deployed engineering team builds and maintains the configuration on the customer's behalf. For enterprises that do not want an AI ops team, that is a feature. For scale-ups that want their own team in control of every workflow change, it is a constraint.
Sierra's pricing pitch is incentive alignment: you pay only when the agent fully resolves a case. The side effect is that a vendor paid only on full resolution gravitates toward the easy calls and away from the hard ones, which in regulated support matter most. Sierra also offers inline payments in the conversation and a strong enterprise security posture, with a high-touch implementation that runs longer than an operator-owned rollout.
Pricing: not published; outcome-based, negotiated per customer.
3. Decagon
Best for: high-volume D2C and consumer SaaS support teams that want a batteries-included concierge with strong chat and a growing voice channel.
Decagon is the platform most often paired with Lorikeet and Sierra in bake-offs. The pitch is batteries-included: polished demo, fast time to first conversation, and a growing voice channel. Its strongest channel is web chat; SMS exists but is less mature. Deployment is white-glove with embedded engineering during launch, and pricing is customer-selectable between per-conversation and per-resolution.
Its strength is also its constraint: it is designed for high-volume, standardized use cases where the configuration levers do not need to go deep. Fintech, healthtech, and insurance teams that need fine-grained guardrails, cross-channel audit trails, or multi-system action chains often find the customization surface narrower than they need.
Pricing: not published; contact sales.
4. Ada
Best for: chat-first deployments adding SMS and lighter voice automation on top of an established chatbot stack.
Ada is one of the longest-running names in AI customer service, with a deep install base in e-commerce, travel, and fintech. Its core competency remains web chat, with strong content authoring tools and a no-code workflow builder. Ada has expanded into SMS, email, and voice through partnerships and acquisitions. Where it lags Lorikeet and Sierra is cross-channel parity: a workflow built for Ada chat does not run unmodified on Ada voice, so each channel needs its own configuration pass. For mostly-chat teams that is acceptable; for serious voice volume it becomes duplicate work.
Pricing: not published; contact sales.
5. Intercom Fin
Best for: teams already on Intercom who want a competent AI concierge across chat, email, and Intercom Phone.
Intercom Fin is the AI agent inside the Intercom helpdesk: natively wired into the inbox, knowledge base, user data model, and Intercom Phone, with SMS through Intercom's messaging tools and Fin Tasks for multi-step actions. It also works with Salesforce and HubSpot helpdesks. Fin is built to be a great agent inside Intercom rather than a portable concierge, and its center of gravity remains chat. For teams who want voice on Twilio or Genesys, or SMS independent of Intercom's stack, the fit gets harder. A low per-resolution rate alone does not tell you total cost: it still rewards handling the easy tickets and routing the hard one to a human.
Pricing: per resolution on top of the Intercom subscription; contact sales for rates.
6. Forethought
Best for: email-heavy support teams that need triage, draft suggestion, and ticket automation, with lighter chat coverage.
Forethought built its reputation on email and ticket automation, with Agatha (triage), Assist (agent suggestions), and Solve (autonomous resolution). It is strongest where queues are email-shaped. Chat is supported; SMS and voice are not where it leads. For voice and SMS concierge work it is a partial fit, and buyers who need those as primary channels will find more depth at Lorikeet, Sierra, or Decagon.
Pricing: not published; contact sales.
7. Goodcall
Best for: SMB and lower-mid-market voice automation with lighter SMS as an adjunct.
Goodcall is a voice-first platform with strong content positioning around fintech and SMB voice automation. It handles inbound voice well at SMB scale, with usage-based pricing and a quick path to first call. SMS is a lighter adjunct rather than a peer to voice, so it is a strong voice option with a thinner messaging story. We include it because it routinely surfaces in voice-focused buyer searches.
Pricing: usage-based; contact sales.
8. Salesforce Service Cloud Voice plus Marketing Cloud
Best for: Salesforce-native enterprises that want voice through Service Cloud and SMS through Marketing Cloud, accepting the two-product split.
Salesforce's omnichannel story spans two products: Service Cloud Voice for voice, and Marketing Cloud (or Data Cloud plus Digital Engagement) for SMS, email, and chat. Both are mature with deep CRM integration. The constraint is that they are two products, and the AI layers (Einstein, Agentforce) operate differently across them, so workflow parity requires careful design across two clouds, and Agentforce's AI quality has been mixed in head-to-head evaluations against pure-play vendors. Lorikeet integrates with Salesforce, so Salesforce customers can keep their CRM of record and run the concierge on a different layer.
Pricing: per-seat licensing per product; contact sales.
9. Twilio Flex AI
Best for: engineering-heavy teams who want to build a custom omnichannel concierge on the Twilio stack.
Twilio Flex is the programmable contact center platform behind a meaningful share of voice and SMS support globally, and Flex AI adds orchestration primitives for AI agents. Channel breadth is excellent: voice, SMS, chat, email, and messaging apps all run on Twilio infrastructure with direct carrier relationships. The constraint is the one Twilio has always had: Flex AI is a build-your-own toolkit rather than a turnkey concierge. Strong engineering teams can build something bespoke; buyers who want first AI conversation in weeks rather than quarters find the build cost higher than expected.
Pricing: usage-based telephony plus seat fees; contact sales.
10. Zendesk AI
Best for: existing Zendesk customers who want AI agents inside their helpdesk across chat, email, voice (Zendesk Talk), and SMS.
Zendesk's AI suite (AI Agents, AI Copilot, AI Bots) runs inside the Zendesk helpdesk across chat, email, voice via Zendesk Talk, and SMS. For the Zendesk install base it is the easiest path to AI agents: bundled, native, data model already wired. The read is similar to Intercom Fin: strong inside Zendesk and weaker outside it. Teams who want voice on Genesys or Twilio while keeping Zendesk as ticketing typically need a separate AI vendor. Lorikeet integrates with Zendesk natively, so Zendesk customers can keep the helpdesk and run voice and SMS on Lorikeet.
Pricing: per resolution and per seat on top of Zendesk Suite; contact sales.
Detailed feature matrix
Capability | Lorikeet | Sierra | Decagon | Ada | Intercom Fin | Forethought | Goodcall | SF SCV + MC | Twilio Flex AI | Zendesk AI |
|---|---|---|---|---|---|---|---|---|---|---|
Voice + SMS + chat + email | Yes | Yes | Partial | Partial | Partial | Limited | Voice-led | Two products | Build it | Inside Zendesk |
Single workflow across all channels | Yes | Yes | Partial | Limited | Inside Intercom | Email-led | Voice-led | No | If you build it | Inside Zendesk |
Runtime guardrails | Yes | Custom | Limited | Limited | Limited | Limited | Limited | Einstein Trust | Custom build | Limited |
QA on 100% of conversations with refund on bad scores | Yes (Coach, Quality Guarantee) | Not stated | Not stated | Not stated | Not stated | Not stated | Not stated | Not stated | Build it | Not stated |
Operator-owned (not managed service) | Yes | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
Pricing model | Per resolved ticket, no seats, published | Per resolution | Hybrid | Hybrid | Per resolution plus seats | Per resolution | Usage | Per seat | Usage plus seat | Hybrid |
Cross-channel audit log | Yes | Yes | Partial | Limited | Inside Intercom | Inside helpdesk | Voice only | Salesforce | Build it | Inside Zendesk |
Lorikeet vs Sierra vs Decagon for voice and SMS
These three meet most often in competitive evaluations for voice and SMS, and they are different products for different buyers. Sierra is the managed-service option for large enterprise, Decagon the batteries-included option for high-volume consumer support, and Lorikeet the operator-owned option for regulated scale-ups that need one workflow across every channel and quality review on every conversation.
Dimension | Lorikeet | Sierra | Decagon |
|---|---|---|---|
Channels | Chat, email, voice (US, UK, AU), SMS | Voice, SMS, chat, email, messaging apps | Voice, chat, email; SMS limited |
One workflow across voice and SMS | Yes, same engine, shared state | Yes, built by Sierra's team | Partial; chat strongest |
Voice latency | Sub-second, published | Not stated | Not stated |
Who owns configuration | Your team | Sierra's forward-deployed engineers | Your team, white-glove launch |
Quality review | Coach scores 100%; Quality Guarantee refund | Not stated | Not stated |
Pricing | Published: from $2,100/mo, per resolved ticket, no seats | Outcome-based, negotiated | Per conversation or resolution, negotiated |
Lorikeet vs Sierra
Both run voice and SMS on one agent layer with shared context, which puts them ahead of the rest of this list on channel parity. The difference is who holds the keys. Sierra's forward-deployed engineers maintain the configuration, so a workflow change is a request to Sierra. On Lorikeet, your support lead edits the natural-language workflow, runs it through simulation, and ships it. For a scale-up that changes policy weekly, that compounds. On billing, Lorikeet publishes its rates and pairs them with a Quality Guarantee, so a resolution Coach scores badly is refunded rather than billed.
Lorikeet vs Decagon
Decagon wins on time to a polished first conversation for standardized, high-volume chat. Lorikeet wins where workflows go deep: identity checks before an action, multi-system action chains, and audit trails that follow one customer from a text to a call. Decagon's SMS is less mature than its chat and its customization surface is narrower than regulated teams usually need. Lorikeet's deterministic structured workflows exist for exactly those teams, and Coach reviews every conversation rather than a sample.
Sierra covers more channels through a managed service. Decagon is more self-serve and quicker to first conversation, with a chat-led product and a growing voice channel. A large enterprise with budget for an embedded vendor team leans Sierra; a consumer brand that wants strong chat run by its own team leans Decagon. A regulated scale-up that needs voice and SMS on one engine, published pricing, and 100% QA coverage is better served by Lorikeet than by either.
How to run one action chain across voice, SMS and chat
AI agents that run action chains over voice and SMS are the point of this category. A concierge that can only answer questions is a voicebot with a phone number. The test of a platform is whether one action chain, built once, runs the same way on every channel. Here is how a refund chain is assembled on Lorikeet, and what to ask any vendor to demonstrate.
Start with the outcome, not the channel. Write the chain as a natural-language workflow: verify the customer, look up the order, check policy, issue the refund, confirm. The workflow never mentions voice or SMS; channel is a property of the conversation.
Wrap the risky steps in deterministic structured workflows. Verification and the refund run as structured workflows with validation on every input, so the refund cannot fire before verification passes. The helpdesk (Zendesk, Intercom, HubSpot, Front, or Salesforce) and billing system are wired once at the workflow level, so voice and SMS hit the same endpoints with the same logging.
Simulate before you ship. Run the chain against realistic scenarios, including failure paths: the lookup times out, the customer fails verification, the refund API errors mid-chain. Runtime guardrails decide what happens on each: retry, steer, or escalate with full context.
Let the channel shape delivery, never logic. On voice, the agent reads back the amount and asks for spoken confirmation. On SMS, it sends the amount and waits for a reply. On chat, it shows a card. The step is identical; only the rendering changes.
Preserve context across the handover. Shared state sits at the customer identity layer, not the channel layer, so a customer who starts on SMS at 11pm and calls the next morning is picked up mid-chain by the voice agent.
Score every run. Coach reviews the completed conversation on every channel and assigns a Ticket Quality Score; a Critical score flags the workflow for a fix, and the Quality Guarantee refunds the AI portion.
In a bake-off, ask each vendor to build that chain once and run it on voice, SMS, and chat with the same guardrails and audit record; separate builds per channel means three products glued together. For more, read our guide to voice AI for multi-step workflows.
Multi-step AI support for scale-ups in North America
Multi-step AI support for scale-ups in North America has requirements a generic chatbot evaluation misses. Scale-ups in fintech, healthtech, insurance, and consumer subscription hit the same five constraints, and the vendor list narrows fast once you apply them.
US phone numbers on day one. Voice has to run on numbers your customers recognize. Lorikeet provisions US numbers directly, with UK and Australian numbers for teams that expand.
Consent-first SMS and voice. US messaging rules put the burden on the sender to hold consent and honor opt-outs. Any proactive contact should be consent-first, with consent state tracked in the customer record and enforced by the workflow.
HIPAA with a signed BAA for healthtech. A vendor that says HIPAA-aligned without offering a Business Associate Agreement is not a healthtech vendor. Lorikeet signs a BAA and holds SOC 2 Type 2 and ISO 27001.
Helpdesk continuity. Scale-ups have already picked Zendesk, Intercom, HubSpot, Front, or Salesforce. The concierge has to read and write to that system rather than replace it.
Pricing that survives a spike. Per-seat pricing punishes a scale-up for growing. Per-resolution pricing with no seats means a rent-week or tax-time spike costs what it resolves.
The customer stories that map to this profile are Flex (4x chat volume during rent week, 2x CSAT, 50% shorter median conversations) and Summ (97% faster resolutions during tax time, first response from around 30 minutes to under 1 minute). Both are multi-step, spike-driven, regulated workloads, the shape most North American scale-ups are dealing with.
How to choose an AI concierge platform for voice and SMS
Buying voice and SMS AI is different from buying chat. These criteria separate a real voice-and-SMS concierge from a voicebot with a phone number.
Channel parity is the load-bearing question. Build a 20-step workflow that spans voice, SMS, and chat. Configure it once and ask the vendor to run it across all three with the same logic, guardrails, and integrations.
Test context preservation explicitly. Start on SMS, drop off, then call 10 minutes later. The voice agent should know who you are and where you left off.
Measure latency under load, not in a lab. Ask for round-trip voice latency on real telephony numbers during a busy hour. A platform that quotes transcription accuracy but dodges latency is optimizing the wrong metric.
Demand an action on a live call. Have the agent issue a refund without a human handoff, and ask what happens when the backend errors mid-action.
Ask who reviews the conversations. Sampling a few percent of calls is how disclosure misses slip through. Ask whether QA covers every conversation on every channel and what happens when the score is bad. Lorikeet's answer is Coach, 100% coverage, and a refund.
Confirm consent controls for anything proactive. If you will ever proactively text or call customers, ask how consent, opt-outs, and contact-hour rules are enforced, and whether the controls live in the guardrails or with you.
Audit and compliance need to span channels. Ask for a single cross-channel audit log with replay, and confirm the certifications you need (SOC 2 Type 2, ISO 27001, HIPAA with BAA, GDPR) on a public trust center.
Implementation checklist
Pre-purchase: document the channels you need at launch versus phase 2; map your top 20 journeys and the channels they touch; list the systems the AI will read from and write to; list regulatory requirements (HIPAA, GDPR, US messaging consent rules); build a 50-question policy eval covering your hardest scenarios.
Evaluation: run the policy eval on each vendor's voice and SMS channels; verify barge-in, latency, and turn-taking on real numbers; drop a conversation on SMS and pick it up on voice; run one action chain on all three channels and compare the audit records; get written commitments on time to first production conversation and QA coverage.
Why Lorikeet for voice and SMS concierge
Lorikeet is built for regulated B2C and B2SMB scale-ups: one workflow across chat, email, voice, and SMS with shared state, guardrails, and audit trail; sub-second voice on US, UK, and Australian numbers; Coach on 100% of conversations with a Quality Guarantee refund; published per-resolution pricing with no seats.
Summ: 97% faster resolutions during tax time, first response from around 30 minutes to under 1 minute. Flex: 2x CSAT versus the previous tool, 4x chat volume during rent week, 50% shorter median conversations, and a head-to-head win in every metric. Breeze: 40% of complex volume resolved independently within 30 days, with over 90% resolution on the tickets the concierge chose to handle. Read the Summ story and the Flex story.
If you are evaluating AI concierge platforms for voice and SMS customer support, book a Lorikeet demo and bring your hardest calls and text threads. We will run them on your stack, on one engine, before you sign. Full rates are on the pricing page.
