Book a demo

Product

AI AgentsAI SafetyHelpdeskSelf-Service PortalSocial Media AIReturns & Exchanges

Results

Customer Stories50% Performance GuaranteeTrust Center

More

PricingMigrationPartnersLog in
Book a demo
For teams evaluating Decagon, and teams already running it

Decagon has no helpdesk under it. That one fact decides which alternative you need.

Decagon is a capable AI agent layer with brand-name deployments and a Series D behind it. It is also, by design, not a helpdesk: its own integrations page describes routing email and chat through Zendesk, Intercom, Salesforce and Kustomer, and never mentions an agent inbox of its own. That architecture splits the alternatives question into two questions with two different answers. If you have not bought yet, you are shopping around gated pricing and a three to six week implementation. If you are already live, your setup cost is spent and the live issues are the second bill underneath the layer and the share of conversations the AI does not resolve. This guide defines six operational criteria before scoring anyone, runs a matrix across eight platforms with every price fetched from the vendor's own page on 28 July 2026, names the situation where each one beats Richpanel, and ends with a decision tree whose last line says stay where you are.

By Amit RG, Founder, Richpanel Published 2026-07-28 Updated 2026-07-28 ~14 min read
View as Markdown →
AR
Amit RG is the founder of Richpanel, the AI-native helpdesk serving 3,000+ brands. He sits in vendor bake-offs against most of the platforms compared here. Two sources ground this piece beyond the usual desk research: every vendor price below was fetched from that vendor's own public page on 28 July 2026, and the sections on what running an agent layer actually looks like draw on a July 2026 screen-share walkthrough of a live Decagon production tenant, conducted with the operator who runs it. That account is anonymized throughout at their request, and no figure from it is published here. On X: @realamitrg.
The short answer, up front

Two different problems, two different shortlists.

Almost every published Decagon alternatives list answers a question nobody asked, which is "who else sells an AI agent." The useful split is by where you are standing.

The table below scores eight platforms on six criteria: architecture, published unit economics, what happens to the un-resolved share, who authors the tests, time to production, and channel and action depth. Each is defined operationally in the next section so two evaluators would score a vendor the same way.

Platform Architecture Published price (28 Jul 2026) Billing unit The un-resolved share Time to production
Richpanel AI agents and the helpdesk are one product. No second platform underneath. $99/seat/mo annual, plus ~$0.20 per AI-resolved conversation, $200/mo minimum Seat plus conversation Worked by your team in the same helpdesk, with an AI copilot; QA AI reviews closed conversations 30-min proof of concept on your data during the demo, 2-week pilot, 4-week deployment
Decagon Agent layer. Requires your existing helpdesk or CRM underneath (names Zendesk, Intercom, Salesforce, Kustomer) None published; /pricing returns 404. Reported $95K to $590K+/yr, ~$386K median (third party) Per conversation or per resolution (described publicly, rate not published) Escalated into the helpdesk you buy separately "3-6 weeks from kickoff to production" (Decagon's own docs)
Sierra Agent layer over your CRM, helpdesk and data systems None published. Reported ~$150K/yr floor (third party) Per successful outcome. "If the conversation is unresolved, in most cases, there's no charge" Escalated into your helpdesk 4 to 10 weeks across Sierra's own published case studies
Lorikeet Agent layer on Zendesk, Intercom or Front Start $1,500/mo (18,000 credits/yr), Scale $4,000/mo (48,000/yr), Enterprise custom. No seat fees, no implementation fees Per resolved ticket: $0.95 chat/email/SMS on Start, $0.80 on Scale. Voice $1.50 and $1.20 up to 3 min Escalated into your helpdesk. Automated QA billed separately at $0.30 and $0.25 Not published
Fin (Intercom) Runs inside Intercom, or standalone on Zendesk, Salesforce, HubSpot, Freshworks, Gorgias $0.99 per resolution. Intercom seats from $29/seat/mo. Standalone: no seat cost, 50-outcome monthly minimum, no setup or platform fees Per outcome. Unresolved handoffs to a team are not charged. Qualifications $9.99 Your team, in whichever helpdesk you run. Copilot $35/user/mo Not published; self-serve start
Ada Agent layer on your existing stack None published; pricing page is a demo booking Not published Escalated into your helpdesk Not published
Gorgias Ecommerce helpdesk with AI Agent included on every plan Plans priced by ticket volume, 50 to 5,000 tickets/mo, "never per agent". AI Agent $0.90 per resolved interaction, $1.00 on Starter Ticket plan plus per resolved interaction. Overage $0.83 to $2.00 by tier and term Worked by your team inside Gorgias Not published; self-serve start
Zendesk Helpdesk with AI agents included across Suite plans Suite Team $55, Suite Professional $115 per agent/mo paid yearly. Copilot +$50, Workforce Engagement +$50, Contact Center +$83 Seat, plus automated resolutions beyond plan allowances (no public per-resolution rate) Worked by your team inside Zendesk. QA sits in the Workforce Engagement bundle Not published

Every cell above was read from the vendor's own public page on 28 July 2026, except the two marked "reported", which come from third-party procurement aggregators rather than the vendor. If your reading of any cell differs from current reality, email amit@richpanel.com and we will correct it.

The honest read: three of the eight publish no per-unit rate at all, and the platforms divide less on AI quality than on whether they bring a helpdesk with them. Decagon, Sierra, Lorikeet and Ada are agent layers, which means a second contract underneath. Richpanel, Gorgias and Zendesk are the platform and the agent together. Fin is the interesting hybrid: it will run standalone on your existing helpdesk with no seat cost, or inside its own. That split, not the model quality, is what actually moves a support budget.

Defined so two evaluators would score the same

Six criteria, and the two that carry the weight.

The weights below reflect the reader this guide is written for: a CX or support leader at a mid-market or enterprise brand who is either pricing Decagon or already running it, and whose real constraint is the total cost of the whole support stack, not the sticker on one line item.

01

Architecture: does it bring the helpdesk? (weight: high)

Operational test: after the AI escalates, which product does your human agent open, and who invoices you for it? If the answer is a different vendor, you are buying a layer, and the agent contract is not your bill. This is the criterion that reorders every published Decagon ranking, because most of them compare agent layers to agent layers and never surface the platform underneath.

02

Published unit economics (weight: high)

Is there a rate on a public page, and what exactly is the unit: a conversation, a resolution, an outcome, a seat, a credit? Operational test: can you model a monthly bill for your own volume without booking a sales call? Three of the eight platforms here fail that test outright, which is not disqualifying but does change how long your evaluation takes.

03

What happens to the share the AI does not resolve (weight: high)

Every vendor sells you the resolved share. Ask what they give the humans handling the rest: the inbox, the context, the copilot, the QA loop, the reporting. Operational test: name the product your team uses for that work and the line item it sits on. For an agent layer the honest answer is "your helpdesk", which means a second product and a second renewal.

04

Who authors and owns the tests (weight: medium)

Can your own CX team write test cases, run them against a change, and see the regression, without filing a vendor ticket? Worth saying plainly: Decagon scores well here. It ships customer-configured Testing and QA, natural-language QA criteria in Watchtower, and an admin copilot, Duet, whose Autopilot mode proposes its own fix and stages it for human review. Any comparison claiming Decagon keeps evaluation vendor-side is out of date.

05

Time from kickoff to production (weight: medium, or zero)

Vendor-published or case-study-evidenced, not a sales claim. The weight on this criterion drops to zero if you are already live on an agent layer, because that cost is already spent. Treating a sunk cost as a live comparison is the most common mistake in switching evaluations, and vendors lean on it because it is the easiest thing to market.

06

Channel and action depth (weight: medium)

Does the agent handle voice natively, and does it execute real operations (refunds, cancellations, order edits, subscription changes) as typed, policy-bounded tool calls rather than free text? Free text saying "I have refunded you" with nothing behind it is how an AI agent commits a company to something it did not authorize. See our breakdown of how to defend against AI hallucinations in support.

Criteria 1 through 3 decide the budget. Criteria 4 through 6 decide the fit once the budget works. If you take one thing from this section: score the stack, not the agent.

Credit first, then the structure

Decagon is good at the hard part. The gaps are structural, not quality.

Anything written by a competitor should start by saying what the competitor is genuinely good at, so here it is, sourced from Decagon's own product pages rather than from a rival's characterization of them.

The admin-side tooling is the strongest part of the product. AOPs, Decagon's Agent Operating Procedures, are natural-language instructions compiled into structured, validated workflows, and Decagon's own documentation is explicit that CX teams author the logic in everyday language while engineers handle guardrails, security and integrations.[2] Duet, its admin-facing copilot, reads past interactions to draft those procedures, detects gaps in guardrails and entry criteria "before they reach customers," generates tests that stress edge cases, gives real-time tracing on failures, and in Autopilot mode turns production signals into proposed updates that are staged for human review with a health report.[1] Watchtower runs always-on QA, and Experiments does live A/B testing. That is a serious, well-built improvement loop.

Two things you will read in other comparisons are simply wrong. The first is that Decagon keeps evaluation locked behind a vendor dashboard. It does not: customers configure their own tests and QA criteria, and Duet's auto-QA authors a proposed procedure change with a reviewable diff. The second is that Decagon is an unproven startup. It raised a Series D at a reported $4.5 billion valuation in January 2026, and the customers it names on its own homepage include Chime, Duolingo, Rippling, Cash App, Fanatics and Live Nation, with published outcomes such as 70% chat and voice resolution at Chime and an 80% deflection rate at Duolingo.[1] Deployments at that scale are not a fluke, and a maturity argument against Decagon would be dishonest.

Its integration model is also a genuine strength for the right team. Rather than a fixed catalog of pre-built connectors, Decagon leans on customer-authored tools against any API, plus an MCP path. Teams with engineering capacity treat that as an advantage: if a system has an API, they can build the tool themselves, in an afternoon, without waiting on a vendor roadmap. If your team writes its own integrations, weigh that seriously, because no opinionated platform will match arbitrary-API reach.

Now the structure. Decagon's own integrations page describes taking action "across Zendesk Sunshine, Salesforce, and more," handling "email inquiries through Zendesk and Intercom, with smooth handoffs when needed," and routing calls with "seamless transfers to human agents." Kustomer appears under knowledge base syncs. Nowhere does it describe an agent inbox of its own, because that is not what the product is.[1] Two consequences follow, and neither is about how good the AI is.

Consequence one: two systems of record. The agent layer holds the conversations it handles; the helpdesk holds the ones your humans handle. Where a chat is resolved entirely by the AI, it may never land in the CRM at all, which quietly breaks the customer history your team relies on and any reporting built on it. Teams solve this with custom sync work, and that sync becomes something you own and maintain.

Consequence two: nothing addresses the share the AI does not resolve. Take Decagon's own published customer numbers at face value: at 70% resolution, three in ten conversations still reach a person; at 80%, one in five does. In the deployments we have seen up close the human share runs closer to one in three. Those conversations are the expensive ones, commonly $2 to $10 each in fully loaded human cost against roughly $0.20 for an AI-resolved conversation on our own pricing, and they arrive pre-touched by an AI, which makes them harder for an agent to pick up cold. An agent layer, by construction, does not staff that work. That is not a criticism of Decagon. It is the definition of the category.

Where each one wins

The situation where each platform beats us.

For every platform in the matrix, here is a specific situation where it is the better choice than Richpanel. If your situation matches one of these, take it seriously and stop reading.

Decagon

The deepest admin-side agent-improvement tooling in the category, and an integration model that reaches any API your engineers care to wire. Choose Decagon over Richpanel if you have in-house engineering capacity that wants to author its own tools rather than consume a connector catalog, you are already standardized on a helpdesk you are content to keep paying for, and you want Duet-grade tracing, simulation and auto-QA on the agent itself. One more honest line: if you are mid-deployment on Decagon and it is hitting your number, the right answer is usually to stay. We will make that case in a section of its own below rather than pretend otherwise.

Sierra

Outcome-based pricing stated in the vendor's own words, with real teeth: "If the conversation is unresolved, in most cases, there's no charge," and the same for escalations.[4] Plus deep managed-deployment muscle and native voice. Choose Sierra over Richpanel if you want a fully managed engagement where the vendor's team does the build, you want to pay only when a task completes, and you have budget for a six-figure floor (reported around $150K a year, not vendor-published). For a large enterprise that would rather buy an outcome than run a platform, that is a coherent purchase and we lose those deals.

Lorikeet

The only platform in this comparison that publishes a complete rate card: $1,500 or $4,000 a month, $0.95 or $0.80 per resolved chat, email or SMS ticket, voice at $1.50 or $1.20 for resolutions up to three minutes, automated QA at $0.30 or $0.25, "Per seat charges: None" and no implementation or platform fees, billed only on successfully resolved tickets.[5] Choose Lorikeet over Richpanel if you are in a complex or regulated vertical where deterministic workflow execution matters more than breadth, you already run Zendesk, Intercom or Front and intend to keep it, and you want to model your exact bill from a public page this afternoon. That transparency is a real advantage over most of this field, including the parts of it that are not Richpanel.

Fin (Intercom)

The lowest published per-outcome rate among the dedicated agent layers, at $0.99 per resolution, with "no additional costs, such as integration fees, setup fees, or platform charges" and "unlimited teammates" when it runs standalone on Salesforce, HubSpot, Freshworks, Gorgias or Zendesk, on a 50-outcome monthly minimum.[6] Unresolved handoffs are not charged. Choose Fin over Richpanel if you want the cheapest published per-outcome rate on the helpdesk you already own, with no platform fee and no seat cost for the agent. The variable to weigh is roadmap: Salesforce signed a definitive agreement to acquire Fin for approximately $3.6 billion on 15 June 2026, expected to close in Salesforce's fiscal Q4 2027 subject to regulatory clearance, and the stated plan folds it into Agentforce.[6] Signed is not closed, but a multi-year decision should price it in. We wrote the full read in what the Salesforce and Fin deal means for support teams and a wider comparison in best Intercom Fin alternatives.

Ada

A mature, no-code agent builder with a long track record in multilingual enterprise support. Choose Ada over Richpanel if your team wants to configure and maintain the agent without writing anything, you want a managed enterprise engagement, and gated pricing is acceptable in your process. Ada publishes no rate as of 28 July 2026: the pricing URL is a demo booking page, so budget the evaluation time accordingly.[7]

Gorgias

The most ecommerce-native platform here, with the widest Shopify app marketplace, AI Agent included on every plan, and a pricing model that is deliberately "never per agent" so growing the team does not grow the bill.[8] Choose Gorgias over Richpanel if you are a Shopify-first brand whose top priority is the breadth of the pre-built app catalog and you want a self-serve start rather than a guided deployment. The trade-off to model is that AI is billed at $0.90 per resolved interaction on top of a ticket-volume plan, with overage from $0.83 to $2.00 depending on tier and billing term. If you are weighing Gorgias directly, see our Gorgias alternatives comparison.

Zendesk

The consolidation option, and the most underrated line on this list for anyone already running Decagon on top of Zendesk. Autonomous AI agents are now included across Suite plans rather than sold as a separate agent add-on, the ecosystem is the largest in the category, and it is the easiest name to get through a procurement committee.[9] Choose Zendesk over Richpanel if your leadership wants fewer vendors above all else and deleting the agent-layer contract in favor of the bundled agent gets you most of the way there. The trade-offs to model: automated resolutions bill beyond plan allowances with no public per-resolution rate, and Copilot, Workforce Engagement (which is where QA lives) and Contact Center are separate add-ons at $50, $50 and $83 per agent per month. Our Zendesk alternatives comparison works the full cost stack.

Richpanel

Stated as plainly as the others. Richpanel is an AI-native helpdesk: the AI agents and the helpdesk your humans work in are one product, priced at $99 per seat per month on annual billing plus about $0.20 per AI-resolved conversation with a $200 monthly minimum, carrying a 50% resolution guarantee in 30 days or your money back. Choose Richpanel if you want the resolved and the un-resolved share of your volume living in one record, one QA loop and one bill, with a published unit price you can model today. In production that has looked like a wellness brand where AI sends 63% of every customer message and closes routine conversations end to end with no human touch, at 4.39 out of 5 CSAT against its own team's 4.33, and a subscription tea brand where AI closes about 37% of volume at CSAT parity with its human team.[10]

Where we are honestly weaker. Three things, named. We do not host native voice: we integrate with Aircall, Dialpad and JustCall, while Decagon, Sierra, Lorikeet, Ada, Gorgias and Zendesk all ship voice natively in some form, so a voice-first support model should weigh them above us. We ship a native connector catalog for the common commerce, subscription and logistics stack and build the rest on request, which does not match Decagon's arbitrary-API reach if your team wants to author a tool against any endpoint in sixty seconds. And on vendor scale we are smaller than Decagon at a reported $4.5 billion valuation or Sierra at a reported $15.8 billion, which is a real consideration if your committee treats funding as a risk control. Our answer to that last one is references at your scale plus the audit reports (SOC 2 Type II, HIPAA-audited, GDPR) on our trust portal, not an argument.

The section nobody writes

Already live on Decagon? Time to value is not one of your questions.

Most switching content is written for buyers, which makes it useless to the larger group: teams already running an agent layer in production. The pitch that lands with a buyer bounces off an operator, so start by crossing things off.

Cross off time to value. If Decagon is live and hitting a number, the implementation cost is spent. Nobody re-buys a deployment to save weeks they have already paid for, and any vendor leading with speed at your account has not understood your situation.

Cross off feature novelty. At this stage the question is not what else a platform can do. It is how to do what you already do with fewer tools and less money.

Cross off the pilot as proof. You have already watched an AI agent work on your tickets. A second bake-off proves something you know.

That leaves two questions, and they are both about the shape of the stack rather than the quality of the model.

One: count the contracts the agent layer is holding up. A mature agent-layer deployment usually sits on a helpdesk or CRM, plus a separate QA tool, a separate CSAT tool, and a separate knowledge tool. Write down all five, their annual cost, and their renewal dates. The agent contract is often not the largest number on that page. The helpdesk underneath, increasingly used as a pipe for conversations the AI already touched, frequently is.

Two: look hard at the share the AI does not resolve. Even at Decagon's own published customer figures, one in five conversations reaches a human, and in practice the human share tends to run higher. Those are the expensive conversations, they arrive already handled once, and your agents are picking them up cold. Nothing in an agent layer is built to make that work cheaper, faster or better, because that work happens in a product it does not own.

The move that changes the arithmetic is not a better agent. It is putting the agent and the helpdesk in the same product, so the resolved and un-resolved halves of your volume share one customer record, one QA loop and one bill, and the four satellite tools collapse into the platform. On the commercial side, we buy out the remaining term on your existing contracts, and your current helpdesk keeps serving until the Richpanel setup is verified, then you cut over once.

And if that does not change your number, stay. That is a legitimate outcome of this section, and it is the outcome for plenty of the teams we talk to. A working deployment is worth more than a marginally cheaper one.

A decision tree, not a verdict

Find the line that describes you.

There is no single best Decagon alternative. There is a best one for where you are standing, what you already own, and which half of your volume you are trying to fix.

The answer flips on facts about you, not on who wrote the article. Note that two of the seven lines send you somewhere other than Richpanel even when the fit is good, and the last one tells you to do nothing. If a single name appeared on every line, you would be reading marketing again.

Run these before you sign anything

Six questions that cut through the pitch.

Whichever shortlist you land on, these six separate the platforms that change your cost structure from the ones that demo well. For the long version, use our 40-question vendor RFP template.

1. After the AI escalates, which product does my agent open?

And who invoices me for it? This one question separates an agent layer from a platform, and it is the difference between one contract and two.

2. What is the per-unit rate, in writing, and what is the unit?

Conversation, resolution, outcome, credit or seat. Then ask what happens to that unit when a conversation reopens, or when a human joins halfway through.

3. What do you give the humans handling everything the AI does not resolve?

Name the product and the line item. If the answer is "your existing helpdesk," you now know which half of your support cost this purchase does not touch.

4. Run the agent on 100 of my historical tickets and show per-response accuracy.

Plus a walkthrough of the failures, not just the aggregate. A vendor that will not do this before a contract is selling you a demo.

5. Show me a refund executed as a typed, policy-bounded tool call.

Not free text that claims a refund happened. Ask to see the parameters, the validation and the limits the agent cannot exceed.

6. Give me three reference customers at my volume and vertical, on a call this month.

References at your scale predict your outcome better than any ranking, this one included. Ask them specifically about the un-resolved share.

How this comparison is limited

What this guide cannot tell you.

An honest comparison names its own blind spots. Four apply here.

The claim this guide will stand behind is narrow: Decagon's AI quality is not why teams shop it, the missing helpdesk and the un-resolved share are, and the single most predictive thing you can ask any vendor on this list is what they give the humans handling the conversations their AI does not close.

The wider field, one line each

Five more we did not score.

Each of these comes up in Decagon evaluations. The descriptions come from each vendor's own positioning; verify current pricing and capability directly before shortlisting, because none of these were re-fetched for this table.

Salesforce Agentforce

Enterprise agents locked to the Salesforce stack (Service Cloud plus Data Cloud). Two pricing models are live and mutually exclusive: roughly $2 per conversation, or Flex Credits at about $0.10 per standard action. Both figures are third-party reported. Strong if you are already all-in on Salesforce, a poor fit if you are not.

Cresta

Combines AI agents with real-time guidance for human agents and conversation intelligence. Worth a look precisely because it targets the half of the volume an agent layer ignores, though it is oriented toward large contact centers rather than mid-market CX teams.

Kore.ai

Broad enterprise automation platform with pre-built connectors for Salesforce, SAP and ServiceNow. Strongest where support is one workflow among many across a large enterprise, rather than the main event.

Rasa

A developer platform that keeps language understanding separate from deterministic business logic, so rules execute as flows rather than as model output. The right answer if you want to own, host and audit the agent yourself and you have the engineering team to do it.

Gladly

A people-centered B2C platform where the unit of work is the customer rather than the ticket, with voice as a first-class channel. Fits relationship-led consumer brands whose highest-CSAT channel is the phone.

None of them change the question this guide keeps returning to: when the AI cannot resolve a conversation, what product does your team work in, and who is billing you for it?

Frequently asked

Leaving Decagon, in plain English.

Why do teams look for Decagon alternatives?

For two different reasons, and they lead to two different shortlists. Teams that have not bought yet run into gated pricing (there is no pricing page on decagon.ai as of 28 July 2026), reported contract sizes in the six figures, and Decagon's own documentation putting initial implementation at three to six weeks from kickoff to production. Teams already running Decagon shop for a different reason: Decagon is an AI agent layer, not a helpdesk, so they are paying a second bill for the ticketing system underneath it, plus separate QA, CSAT and knowledge tools, and nothing in the agent layer addresses the share of conversations the AI does not resolve. If you are evaluating, compare on published unit price and time to production. If you are already live, compare on how many tools the layer is holding up and what happens to the un-resolved share.

How much does Decagon cost in 2026?

Decagon does not publish pricing. The decagon.ai/pricing URL returns a 404 as of 28 July 2026, and the models described publicly are per conversation or per resolution. Third-party procurement data cited across the market reports a range of roughly $95,000 to $590,000 a year with a median near $386,000, but that is reported by aggregators rather than published by the vendor, so treat it as directional and get a quote for your own volume. What matters more than the sticker is that the figure covers the agent layer only. Decagon routes escalations into an existing helpdesk or CRM, so a full budget has to include the ticketing platform underneath, and usually the QA, CSAT and knowledge tools alongside it.

Does Decagon include a helpdesk?

No. Decagon is an AI agent layer that connects to the helpdesk or CRM you already run. Its own integrations page names Salesforce, Intercom, Zendesk and Kustomer, and describes handling email through Zendesk and Intercom with handoffs to human agents, plus live chat escalation through Zendesk Sunshine. There is no mention of a proprietary agent inbox for your human team. This is a design choice rather than a flaw, and for a large enterprise already standardized on Salesforce or Zendesk it is often the right one. It matters for budgeting because the agent layer contract is not the whole bill: the platform your humans work in is a separate product and a separate renewal.

Should I switch off Decagon if it is already working?

Often, no. If Decagon is hitting your resolution target and your team has already absorbed the setup cost, that cost is spent, and time to value on a new vendor is not a reason to move. Cross it off the list. The two questions that do survive are about structure rather than capability. First, count the contracts the agent layer sits on top of: the helpdesk or CRM, plus any separate QA, CSAT and knowledge tools, and note their renewal dates. Second, look at what happens to the conversations the AI does not resolve, because those are the expensive ones and an agent layer does not staff them. If collapsing that stack into one platform does not change your number materially, staying is the right call.

What is the cheapest published alternative to Decagon?

On per-outcome rate alone, Fin at $0.99 per resolution is the lowest published figure among the dedicated agent layers, with no seat cost when it runs standalone on your existing helpdesk and a 50 outcome monthly minimum. Lorikeet publishes the most complete rate card, at $0.95 per resolved chat, email or SMS ticket on its Start plan and $0.80 on Scale, with no seat fees and no implementation fees, but with a $1,500 or $4,000 monthly plan underneath. Richpanel is about $0.20 per AI-resolved conversation plus $99 per helpdesk seat per month, which is a different comparison because the seat price replaces the helpdesk you would otherwise buy separately. Compare total cost of the whole stack rather than the per-outcome rate, since an agent layer rate excludes the ticketing platform underneath it.

Which Decagon alternative is best if I want one platform instead of a stack?

Only a few options collapse the agent and the helpdesk into one product rather than layering one on the other. Richpanel is built that way: the AI agents and the helpdesk your team works in are the same platform, priced at $99 per seat per month on annual billing plus about $0.20 per AI-resolved conversation, with a 50% resolution guarantee in 30 days or your money back. Gorgias is the ecommerce equivalent, with AI Agent included on every plan and billed at $0.90 per resolved interaction on top of a ticket-volume plan. Zendesk now includes AI agents across its Suite plans, which makes consolidating onto the bundled agent a real option for a team already running Decagon on Zendesk. Decagon, Sierra, Lorikeet and Ada all require a helpdesk underneath by design.

Sources & references

Where the claims come from.

Inline citations [1]–[10] map to the entries below. Every vendor page was fetched on 28 July 2026. Where a figure is third-party rather than vendor-published, the entry says so.

  1. Decagon product pages. Homepage (named customers and published outcomes including 70% chat and voice resolution at Chime, 80% deflection at Duolingo); integrations page (Salesforce, Intercom, Zendesk, Kustomer, Zendesk Sunshine, Amazon Connect, RingCentral; "email inquiries through Zendesk and Intercom, with smooth handoffs when needed"; no proprietary agent inbox described); Duet product page (admin-facing copilot, Autopilot staging updates for human review with a health report). decagon.ai
  2. Decagon, "From SOPs to Agent Operating Procedures." Verbatim: "Initial AOP implementation typically takes 3-6 weeks from kickoff to production," with CX teams authoring workflow logic in natural language and engineers handling guardrails, security and integrations; notes that custom databases and legacy systems require more engineering effort. decagon.ai/resources
  3. Decagon pricing, reported. No vendor pricing page exists; decagon.ai/pricing returned HTTP 404 on 28 July 2026. The $95K to $590K+ range, the ~$386K median and the January 2026 Series D at a reported $4.5 billion valuation are third-party procurement and market reporting, not vendor-published figures.
  4. Sierra, "Outcome-Based Pricing for AI Agents" (published 10 December 2024). Verbatim: "you pay only when the software achieves specific, valuable outcomes" and "If the conversation is unresolved, in most cases, there's no charge." No dollar figure is published. Deployment windows of four to ten weeks come from Sierra's own published case studies; the reported ~$150K floor is third party. sierra.ai/blog
  5. Lorikeet pricing page. Start $1,500/mo (18,000 credits/yr), Scale $4,000/mo (48,000 credits/yr), Enterprise custom; chat, email and SMS resolutions $0.95 and $0.80; voice $1.50 and $1.20 for resolutions up to three minutes; routing and tagging $0.30 and $0.25; automated QA $0.30 and $0.25; "Per seat charges: None"; no implementation or platform fees; billed only on successfully resolved tickets. lorikeetcx.ai/pricing
  6. Fin pricing page and Salesforce announcement. $0.99 per resolution (charged once per conversation regardless of actions taken); procedure handoffs and disqualifications $0.99; qualifications $9.99; unresolved handoffs to a team not charged; Intercom seats from $29/seat/mo; standalone on Salesforce, HubSpot, Freshworks, Gorgias and others at $0.99 per outcome with a 50-outcome monthly minimum, "no additional costs, such as integration fees, setup fees, or platform charges" and "unlimited teammates"; Copilot $35/user/mo. Salesforce signed a definitive agreement to acquire Fin for approximately $3.6 billion on 15 June 2026, expected to close in Salesforce fiscal Q4 2027 subject to regulatory clearance. fin.ai/pricing
  7. Ada pricing page. No pricing published; the page is a demo booking form with no rate, model or minimum stated, as of 28 July 2026. ada.cx
  8. Gorgias pricing page and AI Agent pricing article (updated 28 May 2026). Helpdesk scales from 50 to 5,000 tickets per month and is "never per agent"; AI Agent included on every plan and billed per resolved interaction at $0.90 on most plans and $1.00 on Starter; overage $1.50 per interaction on Support plus Shopping Assistant plans and $0.83 to $2.00 on Support-only plans by tier and billing term; "You only pay when the AI fully resolves a conversation on its own." gorgias.com/blog/ai-agent-pricing
  9. Zendesk pricing page. Suite Team "US$ 55 agent/month paid yearly", Suite Professional "US$ 115 agent/month paid yearly", Enterprise contact sales; AI agents included across Suite plans with usage billed as automated resolutions beyond plan allowances and no published per-resolution rate; Copilot "US$ 50 agent/month paid yearly", Workforce Engagement Bundle "US$ 50", Contact Center "US$ 83". zendesk.com/pricing
  10. Richpanel pricing and production case studies. $99 per seat per month on annual billing, about $0.20 per AI-resolved conversation, $200 monthly minimum, 50% resolution guarantee in 30 days or money back. Wellness brand: AI sends 63% of every customer message (5,753 of 9,138), closing routine conversations with no human touch, at 4.39/5 CSAT against the team-wide 4.33. Subscription tea brand: AI closes about 37% of volume at CSAT parity with the human team. richpanel.com/case-studies/wellness

Version history, v1.0 (2026-07-28): initial publication. Matrix cells are a snapshot of public vendor product and pricing documentation as of the publication date. Competitor pricing in this category changes monthly; verify before you model.

An agent layer only fixes half your volume. See what one platform does to the bill.

Richpanel is the AI agents and the helpdesk your team works in, on one platform. Bring your five contracts and your real ticket mix, and we will run the AI on your own tickets in a 30 minute demo.

Book my 30-min demo →