# Decagon review (2026): What it is and why enterprises choose Bland
By Samuel Oram-Jones
Decagon is a customer service AI platform that runs chat, email, and voice agents and sells them per conversation or per resolution with no published rate. This review covers how Decagon works, what its pricing means for a budget and which outside AI companies sit behind it, and why enterprises often choose Bland for phone calls, as of November 2026.
## What is Decagon and how does it work?
Decagon sells AI agents for customer support across chat, email, and voice, and it describes one intelligence layer behind all three: "Decagon unifies chat, voice, and email within a single intelligence layer, ensuring customer experiences stay consistent across every channel." Phone is one channel among several, so the product is shaped around support tickets and conversations more than around a phone line.
Agents are written as Agent Operating Procedures (AOPs), workflows that a team drafts in natural language, and Decagon says "Non-technical teams can architect and iterate on agent logic with Agent Operating Procedures (AOPs) and Duet." Decagon's Watchtower scores conversations after launch.
Decagon builds some models itself and buys the rest. It says "We use both models from the frontier labs, along with our models developed by Decagon Labs that are purpose-built for CX performance and latency." Its voice product says voices are "Built on Chord, our own speech model," and Decagon's Chord write-up adds that outside speech models are "still" used "across our stack."
Decagon uses 27 subprocessors, the outside companies that handle a customer's data on a vendor's behalf. They include OpenAI, Anthropic, Cohere, and xAI for language models; ElevenLabs and Cartesia for text-to-speech; and Soniox for speech-to-text. Each is a separate contract for a security team to review. Bland is full stack, so a call stays inside one AI company.
### What features does Decagon offer?
Decagon offers chat, email, and voice, and its llms.txt file adds SMS: "across various channels like chat, SMS, email, and voice." Voice agents handle inbound calls and, Decagon says, can start outbound ones: "Unlock proactive engagement by initiating outbound calls for sales, support, and customer updates."
Decagon says agents "Support customers in 70+ languages with automatic detection and language switching," and that a team can "Choose from hundreds of voices, including custom-tuned ones, and tailor it to sound like your brand."
Decagon doesn't offer iMessage agents. Bland runs branded iMessage agents natively on Enterprise, on the same pathways as voice.
## How much does Decagon cost?
Decagon publishes no rate, and it has no pricing page at all. It describes two ways to pay. Per-conversation pricing is "A fixed rate for every incoming conversation, with flexible pricing for higher volumes," and per-resolution pricing is "A higher fixed rate for each fully resolved conversation, with no charge for escalations." Decagon adds "Larger resolution commitments lower the rate," and neither model comes with a figure.
Per-resolution pricing depends on what counts as resolved, and Decagon is candid about the weak spot. It says "Defining what a resolution is can be tricky," that "Gray areas can lead to billing disagreements," and that "Resolutions may vary month-to-month, making it tough to forecast costs compared to usage-based pricing."
Decagon's own pricing write-up goes further: "You'll get into situations where a user is upset, leaves, and that gets counted as a resolution." A buyer therefore needs written answers to four questions before signing. These are what counts as a conversation, what counts as resolved, whether a transfer to a person is billed, and what happens to the rate when volume lands under a commitment.
### How is Bland priced?
Bland's self-serve Start plan is $0.14 per minute with no platform fee, and self-serve Bland Build is $0.12 per minute with a $299 monthly platform fee, which is lower than Start above about 14,950 minutes a month.
Either rate covers Bland's language model, transcription, and speech with no token charges. Phone is billed separately, on a customer's own carrier or Bland's at pass-through cost. Enterprise pricing is custom, set with Bland's sales team.
Enterprise is the plan regulated teams buy. It adds native iMessage agents alongside SMS and web chat, warm transfers, guardrails, a BAA, SSO, data residency, dedicated orchestration with dedicated GPUs run by Bland, and forward-deployed engineers who work beside a customer's team, with concurrency contracted to its volume and custom pricing.
A minute is a unit both sides can count, while a resolution is a unit the vendor defines. Because Decagon shows no rate, no per-minute comparison is possible, and Bland's published rates are what a team can model against its call center costs today. See [pricing](https://www.bland.ai/pricing) for each plan.
## How easy is Decagon to use?
Decagon's pitch is that a support team can build and tune agents without engineers. AOPs and Duet are the non-technical route, and technical teams "retain full visibility and control over guardrails, integrations, and versioning."
Decagon publishes no latency figure and no uptime number for voice. Bland publishes sub-200ms latency and a 99.9% uptime SLA on every plan, an SLA being the written uptime promise a vendor puts in a contract.
Bland moves an agent from staging to production only when chosen simulated test cases pass chosen judges, and its alarms fire on latency, long silences, transcription quality, and API errors, sent by email, phone call, or API.
Because the models underneath change, an agent's behaviour can too. Decagon says it will "continue to build on leading models while developing specialized models," so whatever sits under an agent moves when the leaders move. A buyer should ask which models answer calls today and how a swap is announced.
## Which sounds more human, Decagon or Bland?
Decagon's Chord speech model has no Audio Realism Bench score. On Design Arena's Audio Realism Bench on 4 August 2026, Bland Speech v3 ranked 1st of 13 AI voices at 1365, so no direct comparison with Chord is possible.
In Design Arena's blind test, vetted native speakers hear two voices read one script and choose the one that sounds more like a real person. Decagon's voice partners ElevenLabs and Cartesia supply text-to-speech behind some of its voices, and the results are public in [Design Arena's published results](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/). A team can listen to both on its own script before choosing.
## Is Decagon secure enough for enterprise calls?
Decagon reports SOC 2 Type II, ISO 27001, PCI DSS 4.0.1, GDPR, and HIPAA, which covers the usual checklist. Decagon's details sit in three places: what a private install contains, when personal details come out, and which agreement covers health calls.
On health data, Decagon reports HIPAA compliance. A BAA is the contract HIPAA requires before a vendor handles patient information, and a health-call buyer asks Decagon for one in writing. Bland includes a BAA on Enterprise.
Redaction happens late. Decagon says "Sensitive data is automatically detected and redacted using Google’s DLP service, sanitizing logs and transcripts just after a conversation ends." That pass starts after a caller has hung up, so raw details sit in transcripts and logs until it finishes. Decagon's description covers logs and transcripts. Bland offers real-time PII redaction on call recordings and transcripts.
On privacy with AI providers, Decagon says it "enforces zero-data retention with all AI providers like OpenAI and Anthropic," which is a promise a security team can ask to see in a contract.
Decagon says "Cloud-prem / VPC means Decagon's full stack runs inside the customer's own cloud environment," and it supports on-premise "for more specialized requirements."
### How does Bland handle security?
Bland is full stack, so no outside AI lab sits in its call path. It includes a BAA on Enterprise, and its [security program](https://www.bland.ai/trust-security) self-attests to [PCI DSS](https://www.pcisecuritystandards.org/standards/pci-dss/) v4.0.1 and holds FedRAMP 20x Class A. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. Bland also offers a 99.9% uptime SLA on every plan.
Bland also runs its own SIP signaling and media servers, so no third-party telephony provider sits in the call path.
## Is Decagon worth it for enterprises?
For most enterprise phone calls, Decagon isn't the best choice as of November 2026, because pricing is set by a resolution definition that Decagon itself calls tricky, and a call depends on several outside AI companies.
### Decagon pros and cons
Pros
- One platform across chat, email, and voice, with SMS listed in llms.txt
- Agent Operating Procedures let non-technical teams write and tune agent logic
- 70+ languages with automatic detection, and hundreds of voices including custom-tuned ones
- SOC 2 Type II, ISO 27001, PCI DSS 4.0.1, GDPR, and HIPAA reported
Cons
- No published rate, and per-resolution billing that Decagon says can lead to disagreements
- Language, speech, and transcription come partly from outside companies beside Decagon's own models
- Redaction runs just after a conversation ends, not during the call
- No iMessage agents
Pros
- One AI company on every call, so a security team reviews one vendor
- One published rate for all of its AI, with no token charges
- Dedicated orchestration and dedicated GPUs run by Bland, on Enterprise
- Native iMessage agents on Enterprise, sharing pathways with voice
- Staged releases that pause a new agent version when it performs worse than the last
- Its own SIP and media servers, with no outside telephony provider in the call path
Cons
- Overkill for a small prototype, because Bland is built for production calls at volume
What these rows mean
- **AI on each call:** This row shows whether the language model, speech, and transcription behind each call belong to the platform or to other companies: Full stack means all three are the platform's own, Outsourced means its language model and transcription come from other companies, and Mixed means it runs some of its own models alongside outside ones. Bland builds and runs its own language, speech, and transcription models, so a call never passes through an outside AI provider. Decagon builds its own Chord speech model and trains its own language models, and its agents also call outside providers such as OpenAI and Anthropic.
- **Audio Realism Bench score, 4 August 2026:** Audio Realism Bench is Design Arena's blind listening test, where listeners pick whichever of two voices sounds more like a real person, and a higher score means listeners picked that voice more often. In its published results of 4 August 2026, Bland Speech v3 ranked first among AI voices. Each rival number is the highest-scoring voice model in those results available on that platform. Not ranked for Decagon means Decagon's own Chord speech model wasn't ranked, though Cartesia and ElevenLabs, the outside voice companies on its subprocessor list, were.
- **iMessage:** iMessage lets an agent text customers on a channel they already use. Bland offers it natively on Enterprise.
- **HIPAA with a BAA:** A Business Associate Agreement, or BAA, is the contract that lets a vendor handle patient information under HIPAA. Bland includes its BAA on Enterprise. On Decagon, HIPAA listed, no BAA stated means Decagon reports HIPAA compliance and doesn't publish a BAA.
- **PCI DSS:** PCI DSS is a security standard from the card industry for handling card details. Bland self-attests to v4.0.1 and shares its attestation under NDA on Enterprise.
- **FedRAMP:** FedRAMP is the US government's cloud security program. Bland holds 20x Class A.
- **AI rate per minute:** Headline rates cover different things, so a buyer should check what each one includes before comparing them. Bland's self-serve rate covers all of its AI: language model, transcription, and speech. Phone is billed separately at carrier cost, and Enterprise pricing is custom, set with Bland's sales team. Decagon prices per conversation or per resolution and publishes no rate.