Decagon and Fin — the company that was Intercom until it renamed itself in May 2026 — are the pair that lands on most AI-support-agent shortlists. Both resolve a customer conversation end to end across chat, email, and voice, and both take the backend action (a refund, an identity check, a plan change) rather than answering from a help center and handing off. One clarification before the routing rule, because it trips buyers: this is one vendor with two names. The agent is Fin, the helpdesk it grew out of is still sold as Intercom, and ooligo carries the vendor at /en/tools/intercom/.
The axis that used to route this decision — a standalone AI-first agent versus an agent bolted onto a helpdesk you already pay for — stopped being true on 9 June 2026, when Fin shipped as a Service Agent running on top of HubSpot and Freshworks, with APIs, MCP, and a CLI, explicitly so buyers could deploy it “without migrating off your helpdesk.” Fin is no longer a reason to run Intercom’s inbox. What is left to decide is narrower and easier to answer with your own numbers: the price shape at your conversation volume, how much of the agent’s behavior your ops team authors versus configures, and whether you want an independent vendor or one that will sit inside Salesforce once the deal closes.
Where Decagon wins
Your ops team authors the behavior, in plain language. Agent Operating Procedures are Decagon’s differentiator: non-technical CS and support-ops staff define multi-step workflows in natural language rather than coded decision trees. Once the engineering-led integration is wired, the people who own the customer experience change what the agent does directly. For a support org that expects to rewrite policy monthly — return windows, escalation thresholds, identity-verification steps — that authoring surface is the reason to pay the floor.
QA and experimentation are native, not a separate purchase. Watchtower runs always-on QA across conversations and Experiments does live A/B testing of agent behavior, so a team that treats resolution quality as a metric to move can tune inside the platform instead of exporting transcripts. Fin reports quality in aggregate; Decagon hands you the loop to change it.
Independence, and enterprise proof at the top of the market. Decagon raised a $250M Series D on 27 January 2026 led by Coatue and Index at a $4.5B valuation, tripling its valuation in under six months, and added more than 100 new global enterprise customers in the fiscal year — Avis Budget Group, Block, and Deutsche Telekom among them. Published customer results include 70% chat and voice resolution at Chime, 80% deflection at Duolingo, and 50%+ voice deflection at Valon. If your procurement process weights vendor independence and regulated-enterprise references, that is the concrete side of the ledger.
Where Fin wins
The price floor is roughly two orders of magnitude lower. Fin is metered at $0.99 per outcome with a 50-outcome monthly minimum — about $50 a month, or roughly $600 a year, with no seat purchase required if you keep your existing helpdesk. Decagon publishes no price at all; its reported structure starts at a ~$50,000 annual platform fee before any usage. For anyone who cannot commit five figures to find out whether an agent works on their traffic, that difference decides the evaluation before features do.
It goes live in hours, and you can test it without a contract. Fin is self-servable — integration and deployment inside an hour, per the June 2026 platform post — and the 14-day trial carries unlimited outcomes. Decagon’s onboarding runs about six weeks from signature to production, and the first weeks are engineering-led. When the question is “does this deflect our tickets,” Fin answers it this week on real traffic.
Scale evidence and a proprietary model. Fin reports a 76% average resolution rate across 12,000+ customers, running on Apex 1.0 and Apex Flash, models trained on customer-experience interactions rather than a general-purpose LLM behind a prompt. Decagon’s public proof is deeper per logo; Fin’s is broader across the market, which is the more useful reference class if you are a mid-market SaaS company rather than a Fortune 500 brand.
Pricing reality
Compare by shape and floor, because only one side publishes. Fin: $0.99 per resolution outcome, $9.99 per qualification outcome, 50-outcome monthly minimum; Intercom helpdesk seats at $29 (Essential), $85 (Advanced), and $132 (Expert) per seat per month if you want the inbox too; Copilot at $35 per user per month. Decagon: quote-only, reported at a ~$50K annual platform fee plus usage billed either per conversation or per resolution, with conversation minimums on enterprise contracts. Third-party procurement marketplaces put real Decagon contracts in the low-to-mid six figures — median snapshots cluster around $390K–$430K a year with a range from roughly $100K to $900K+ — but these are buyer-panel estimates, not a vendor rate card, and they vary by channel mix and volume.
That gives you an actual crossover to compute rather than a preference. At Decagon’s reported median contract of roughly $400K a year, the same budget buys about 404,000 Fin outcomes annually — near 33,000 resolutions a month. Below that line, Fin is cheaper for the same resolved volume and you are paying Decagon for authoring depth and QA tooling, not for throughput. Above it, Decagon’s platform-fee-plus-usage structure amortizes and the per-unit negotiation starts to favor the enterprise contract. The term that decides both: what counts as a billable resolution. Fin bills an outcome; Decagon’s definition is negotiated per contract. Pin that language down before you compare quotes, because it moves the effective rate more than the headline number does.
The Salesforce overhang
On 15 June 2026 Salesforce signed a definitive agreement to acquire Fin for approximately $3.6 billion, expected to close in the fourth quarter of Salesforce’s fiscal year 2027. Eoghan McCabe stays on as CEO and Des Traynor continues to lead R&D after close. Price this honestly in both directions rather than treating it as a disqualifier. It raises the odds that Fin’s roadmap bends toward Service Cloud and that its openness to competing helpdesks gets re-evaluated under new ownership — so if you are a HubSpot or Freshworks shop buying Fin specifically because it runs on your stack, put contractual protection around that, and negotiate term length and price locks that survive the close. If you already run Salesforce, the deal is an argument in favor: you are buying an agent that is heading into your CRM anyway. Decagon’s counter-position is that it will still be independent in 18 months, and for a support org that just finished one vendor migration, that is worth real money.
Implementation effort
Neither one is only a switch, but they are not the same size of project. Fin is self-serve to first value and the meaningful work is policy — deciding what the agent is allowed to do without a human, and writing the escalation rules. Decagon front-loads engineering: API connections to CRM, helpdesk, and data sources come before any CX person can author an AOP, so budget ops and engineering time on top of the license across roughly six weeks. For either, scope one high-volume journey first, set a containment and accuracy baseline against your current tooling, keep hard limits and human review on irreversible actions like refunds above a threshold, and log every agent-executed transaction for reconciliation from day one.
Verdict
Pick Decagon when your support org authors and rewrites agent policy continuously and wants non-engineers doing it in plain language, when native QA and live A/B testing of agent behavior are decision criteria, when your volume is high enough that a five-figure platform fee amortizes, and when vendor independence carries weight in your procurement review. It is the control-first, ops-authored, independent pick.
Pick Fin when you want resolution economics you can model before signing, when you need to prove deflection on your own traffic in days rather than after a six-week implementation, when you want to keep the helpdesk you already run, and when a metered floor near $50 a month beats a $50,000 commitment to answer the same question. It is the low-floor, fast-to-production, portable pick.
If you cannot decide, default to Fin. It costs about $600 a year to find out whether an AI agent resolves your tickets, and that evidence is the input to every other decision here — including whether Decagon’s floor is justified. Run Fin on your real volume for a quarter, then take your measured resolution rate and volume curve into a Decagon evaluation. Flip to Decagon when you hit roughly 33,000 monthly resolutions, when Fin’s configuration surface stops being deep enough for the policy changes you need, or when the Salesforce close creates a roadmap risk your contract cannot cover.
Pick neither when your volume is a few hundred tickets a month and a human team plus a help center still clears it — an agent adds a policy-maintenance burden that only pays back at volume. Choose Salesforce Agentforce instead if you are Salesforce-resident and would rather consolidate agents inside the CRM than buy a specialist that is being absorbed into it anyway, or Sierra if voice is your primary channel and the agent has to take payments inside PCI-scope flows.
Decagon and Fin — the company that was Intercom until it renamed itself in May 2026 — are the pair that lands on most AI-support-agent shortlists. Both resolve a customer conversation end to end across chat, email, and voice, and both take the backend action (a refund, an identity check, a plan change) rather than answering from a help center and handing off. One clarification before the routing rule, because it trips buyers: this is one vendor with two names. The agent is Fin, the helpdesk it grew out of is still sold as Intercom, and ooligo carries the vendor at /en/tools/intercom/.
The axis that used to route this decision — a standalone AI-first agent versus an agent bolted onto a helpdesk you already pay for — stopped being true on 9 June 2026, when Fin shipped as a Service Agent running on top of HubSpot and Freshworks, with APIs, MCP, and a CLI, explicitly so buyers could deploy it “without migrating off your helpdesk.” Fin is no longer a reason to run Intercom’s inbox. What is left to decide is narrower and easier to answer with your own numbers: the price shape at your conversation volume, how much of the agent’s behavior your ops team authors versus configures, and whether you want an independent vendor or one that will sit inside Salesforce once the deal closes.
Where Decagon wins
Your ops team authors the behavior, in plain language. Agent Operating Procedures are Decagon’s differentiator: non-technical CS and support-ops staff define multi-step workflows in natural language rather than coded decision trees. Once the engineering-led integration is wired, the people who own the customer experience change what the agent does directly. For a support org that expects to rewrite policy monthly — return windows, escalation thresholds, identity-verification steps — that authoring surface is the reason to pay the floor.
QA and experimentation are native, not a separate purchase. Watchtower runs always-on QA across conversations and Experiments does live A/B testing of agent behavior, so a team that treats resolution quality as a metric to move can tune inside the platform instead of exporting transcripts. Fin reports quality in aggregate; Decagon hands you the loop to change it.
Independence, and enterprise proof at the top of the market. Decagon raised a $250M Series D on 27 January 2026 led by Coatue and Index at a $4.5B valuation, tripling its valuation in under six months, and added more than 100 new global enterprise customers in the fiscal year — Avis Budget Group, Block, and Deutsche Telekom among them. Published customer results include 70% chat and voice resolution at Chime, 80% deflection at Duolingo, and 50%+ voice deflection at Valon. If your procurement process weights vendor independence and regulated-enterprise references, that is the concrete side of the ledger.
Where Fin wins
The price floor is roughly two orders of magnitude lower. Fin is metered at $0.99 per outcome with a 50-outcome monthly minimum — about $50 a month, or roughly $600 a year, with no seat purchase required if you keep your existing helpdesk. Decagon publishes no price at all; its reported structure starts at a ~$50,000 annual platform fee before any usage. For anyone who cannot commit five figures to find out whether an agent works on their traffic, that difference decides the evaluation before features do.
It goes live in hours, and you can test it without a contract. Fin is self-servable — integration and deployment inside an hour, per the June 2026 platform post — and the 14-day trial carries unlimited outcomes. Decagon’s onboarding runs about six weeks from signature to production, and the first weeks are engineering-led. When the question is “does this deflect our tickets,” Fin answers it this week on real traffic.
Scale evidence and a proprietary model. Fin reports a 76% average resolution rate across 12,000+ customers, running on Apex 1.0 and Apex Flash, models trained on customer-experience interactions rather than a general-purpose LLM behind a prompt. Decagon’s public proof is deeper per logo; Fin’s is broader across the market, which is the more useful reference class if you are a mid-market SaaS company rather than a Fortune 500 brand.
Pricing reality
Compare by shape and floor, because only one side publishes. Fin: $0.99 per resolution outcome, $9.99 per qualification outcome, 50-outcome monthly minimum; Intercom helpdesk seats at $29 (Essential), $85 (Advanced), and $132 (Expert) per seat per month if you want the inbox too; Copilot at $35 per user per month. Decagon: quote-only, reported at a ~$50K annual platform fee plus usage billed either per conversation or per resolution, with conversation minimums on enterprise contracts. Third-party procurement marketplaces put real Decagon contracts in the low-to-mid six figures — median snapshots cluster around $390K–$430K a year with a range from roughly $100K to $900K+ — but these are buyer-panel estimates, not a vendor rate card, and they vary by channel mix and volume.
That gives you an actual crossover to compute rather than a preference. At Decagon’s reported median contract of roughly $400K a year, the same budget buys about 404,000 Fin outcomes annually — near 33,000 resolutions a month. Below that line, Fin is cheaper for the same resolved volume and you are paying Decagon for authoring depth and QA tooling, not for throughput. Above it, Decagon’s platform-fee-plus-usage structure amortizes and the per-unit negotiation starts to favor the enterprise contract. The term that decides both: what counts as a billable resolution. Fin bills an outcome; Decagon’s definition is negotiated per contract. Pin that language down before you compare quotes, because it moves the effective rate more than the headline number does.
The Salesforce overhang
On 15 June 2026 Salesforce signed a definitive agreement to acquire Fin for approximately $3.6 billion, expected to close in the fourth quarter of Salesforce’s fiscal year 2027. Eoghan McCabe stays on as CEO and Des Traynor continues to lead R&D after close. Price this honestly in both directions rather than treating it as a disqualifier. It raises the odds that Fin’s roadmap bends toward Service Cloud and that its openness to competing helpdesks gets re-evaluated under new ownership — so if you are a HubSpot or Freshworks shop buying Fin specifically because it runs on your stack, put contractual protection around that, and negotiate term length and price locks that survive the close. If you already run Salesforce, the deal is an argument in favor: you are buying an agent that is heading into your CRM anyway. Decagon’s counter-position is that it will still be independent in 18 months, and for a support org that just finished one vendor migration, that is worth real money.
Implementation effort
Neither one is only a switch, but they are not the same size of project. Fin is self-serve to first value and the meaningful work is policy — deciding what the agent is allowed to do without a human, and writing the escalation rules. Decagon front-loads engineering: API connections to CRM, helpdesk, and data sources come before any CX person can author an AOP, so budget ops and engineering time on top of the license across roughly six weeks. For either, scope one high-volume journey first, set a containment and accuracy baseline against your current tooling, keep hard limits and human review on irreversible actions like refunds above a threshold, and log every agent-executed transaction for reconciliation from day one.
Verdict
Pick Decagon when your support org authors and rewrites agent policy continuously and wants non-engineers doing it in plain language, when native QA and live A/B testing of agent behavior are decision criteria, when your volume is high enough that a five-figure platform fee amortizes, and when vendor independence carries weight in your procurement review. It is the control-first, ops-authored, independent pick.
Pick Fin when you want resolution economics you can model before signing, when you need to prove deflection on your own traffic in days rather than after a six-week implementation, when you want to keep the helpdesk you already run, and when a metered floor near $50 a month beats a $50,000 commitment to answer the same question. It is the low-floor, fast-to-production, portable pick.
If you cannot decide, default to Fin. It costs about $600 a year to find out whether an AI agent resolves your tickets, and that evidence is the input to every other decision here — including whether Decagon’s floor is justified. Run Fin on your real volume for a quarter, then take your measured resolution rate and volume curve into a Decagon evaluation. Flip to Decagon when you hit roughly 33,000 monthly resolutions, when Fin’s configuration surface stops being deep enough for the policy changes you need, or when the Salesforce close creates a roadmap risk your contract cannot cover.
Pick neither when your volume is a few hundred tickets a month and a human team plus a help center still clears it — an agent adds a policy-maintenance burden that only pays back at volume. Choose Salesforce Agentforce instead if you are Salesforce-resident and would rather consolidate agents inside the CRM than buy a specialist that is being absorbed into it anyway, or Sierra if voice is your primary channel and the agent has to take payments inside PCI-scope flows.