Competitor Dossier — Enterprise CX AI-Agent Incumbents

Compiled: 2026-06-21 · Method: parallel sourced web-research (official product/pricing/security/trust pages + reputable secondaries), synthesized. Every factual claim is cited inline. Honesty discipline: vendor-published resolution rates, ARR, and per-resolution prices are marketing/secondary-sourced unless noted — they are flagged where soft. Pricing for most of these is sales-gated; published numbers are rare and noted as such.

Why this dossier exists: these are the platforms an OpenAlice enterprise-CX prospect is most likely to also be evaluating. The strategic punchline is at the bottom (§"What OpenAlice should learn / where we win"). The one-sentence version: the entire incumbent field is US-cloud-only SaaS with no self-host and (mostly) bolt-on EU residency, and none has an embodied avatar — those two are our structural wedge; we lag badly on omnichannel breadth, multilingual depth, and marketplace.

⚡ 2026 state-changes you must know (these invalidate older notes)

1. Intercom / Fin (+ Fin Apex)

Positioning (one line): "The only helpdesk designed for the AI Agent era" — an AI-first CX platform centered on the Fin AI Agent that autonomously resolves support across chat, email, phone, and messaging. (fin.ai↗)

Company / funding / scale: Dublin-founded (2011), CEO Eoghan McCabe. Being acquired by Salesforce for ~$3.6B (June 2026). Total ARR ~$400M (Fin-specific ~$100M, growing ~350%/yr); ~30,000 total customers, ~8,000 using Fin. Salesforce's announcement claims Fin resolves ~76% of support volume end-to-end — but Intercom's own marketing has elsewhere cited ~67% average and ~51% for Fin 2 cohorts, so treat 76% as a best-case headline, not an audited average. (Sacra↗, Salesforce Ben↗, CallSphere↗)

Pricing (one of the few with PUBLIC numbers): Outcome-based — $0.99 per resolution (billed once per conversation; not charged on a no-outcome human handoff; 50-outcome/mo minimum). Fin for Sales: $9.99 per qualified lead. Separate human-helpdesk seats from ~$29/seat/mo across Essential/Advanced/Expert. Add-ons: Copilot $35/user/mo, Pro convo-analysis $99/mo, Proactive Support Plus $99/mo, Fin Voice = custom. No volume discount/cap → cost can be unpredictable at scale. (fin.ai/pricing↗, Intercom Help↗)

16-category feature inventory:

1. Conversational AI/LLM — Now runs primarily on Fin Apex, Intercom's own post-trained vertical model trained on billions of proprietary support interactions; claims 73.1% resolution vs 71.1% GPT-5.4 / Claude Opus 4.5 and 69.6% Sonnet 4.6 on their internal benchmark (vendor benchmark, unaudited). Still "heavy customers of Anthropic" — frontier-model fallback routing not publicly detailed. (Apex blog↗)
2. Knowledge/RAG — Centralized Knowledge Hub ingesting help-center articles, internal/external content, past conversations, URLs, PDFs/snippets; readiness-testing tooling. (Fin 3↗)
3. Omnichannel — Chat (Messenger), email, phone, SMS, WhatsApp, Slack, Discord; Voice shares the same knowledge foundation so improvements propagate cross-channel. Native social breadth beyond these is thinner. (Salesforce Ben↗)
4. Voice — Fin Voice 2 ("Apex Flash" voice model): instant answer, no IVR, multi-step resolution via Procedures + Actions (refunds, order status, subscriptions); noise reduction; 28 voice languages; custom/sales-gated. (fin.ai/voice↗)
5. Avatar/embodiment — None. Text + voice only.
6. Persona/brand — Configurable tone, identity, escalation/office-hours, guardrails; moderate depth, not a headline. (Fin 2↗)
7. Lead/CRM/commerce — Fin for Sales qualifies visitors like an SDR, pushes leads to Salesforce/HubSpot/Attio; commerce actions via Actions framework. (Fin for Sales↗)
8. Analytics/insights — CX Score + reasons, Topics Explorer, content-gap suggestions, Fin performance dashboards, Pro convo-analysis add-on. (Monitor Fin↗)
9. Handoff/HITL/inbox — Intercom Inbox (human workspace) + Copilot agent-assist (claims +31% conversations closed/day) + Procedures with human-approval checkpoints + new Fin Operator (AI that supervises the Fin agent, early access). (Copilot↗, VentureBeat↗)
10. A/B testing — Outbound/Series messaging experiments + batch content-readiness testing; no prominent A/B framework for Fin answers (limited/uncertain).
11. Integrations/API/marketplace — Public App Store + REST API + Canvas Kit + webhooks; native Salesforce/HubSpot/Slack/Stripe/Shopify/Zendesk/Attio. Exact app count not published (historically "300+", unverified for 2026). (App Store↗)
12. Security/compliance — Strong: SOC 2 Type II, ISO 27001:2022, 27018, 27701, ISO 42001, HIPAA, HDS, GDPR, CCPA. No FedRAMP. (trust.intercom.com↗)
13. Proactive/triggers — Full Outbound suite: tours, checklists, tooltips, in-app, push, banners, surveys + Series journey builder. (Outbound↗)
14. Multilingual — 45 languages (chat) with auto-detect + real-time translation; 28 for voice. (Fin 45 languages↗)
15. Deployment/install — Cloud SaaS only. JS Messenger snippet (@intercom/messenger-js-sdk), native iOS/Android/React Native/Cordova SDKs; can deploy standalone on Zendesk/Salesforce.
16. Packaging — Essential/Advanced/Expert seats + Fin outcomes billed separately + add-ons.

Self-hostable? EU residency? Self-host: NO (AWS cloud-only). Regional residency: YES — US (us-east-1), EU (eu-west-1 Dublin), Australia (Sydney) — BUT gated to Advanced/Expert annual plans, and some data still leaves the region: billing/credit-card data processed in US, and inbound/outbound email routes through the US. So "EU hosting" is leaky for strict GDPR buyers. (Regional Data Hosting↗, ClonePartner critique↗)

Top differentiators: (1) Owns the pricing narrative — $0.99/resolution is the industry reference point. (2) Own vertical model (Apex) — cost/latency/control vs reselling frontier APIs. (3) Full-stack suite + scale + Salesforce backing. Notable gaps: (1) No self-host + leaky EU residency. (2) Cost unpredictability (stacked seat+outcome+add-on, no caps). (3) No avatar, no FedRAMP, inconsistent public resolution claims (51/67/76%), + acquisition/roadmap uncertainty under Salesforce.

2. Sierra (Bret Taylor / Clay Bavor)

Positioning (one line): Enterprise platform for building branded, autonomous CX agents ("Agent OS") across chat + voice, sold on outcome-based pricing. (sierra.ai/product↗)

Company / funding / scale: Founders Bret Taylor (OpenAI chairman, ex-Salesforce co-CEO) + Clay Bavor (ex-Google). $950M Series E (May 4, 2026) at $15.8B post-money (Tiger Global + Google GV lead); ~$1.585B raised total; Taylor says "more than $1B to invest." ARR ~$150M+ in eight quarters (higher figures are estimates). Claims >40% of Fortune 50 and 1 in 3 of the world's largest banks; customers: SiriusXM, ADT, Sonos, WeightWatchers, Casper, Prudential, Cigna, BCBS, Rocket Mortgage. (Reputational note: a 2025 Gap-chatbot jailbreak raised guardrail scrutiny.) (TechCrunch↗, CNBC↗, eMarketer↗)

Pricing: Outcome-based (pay per resolved outcome; unresolved/escalated = no charge in most cases), with a consumption-based blend for routing/greeter flows. No public price list, no seats, no self-serve. Third-party estimates ~$1–$2.50/resolution, fully negotiated; analysts warn "resolution" is re-defined at renewal, eroding year-two economics. (Outcome pricing blog↗, OpenNash↗)

16-category feature inventory:

1. Conversational AI/LLM — Model-agnostic "constellation of models" routing every turn to the optimal frontier/open/proprietary model with cross-provider failover; supervisor models above the reasoning layer enforce guardrails + intercept sensitive cases. (Constellation↗, Trust↗)
2. Knowledge/RAG — Retrieval is a separable capability feeding both runtime + test-case generation; deeper RAG internals not public.
3. Omnichannel — Official: voice, chat, email, WhatsApp with shared context. (SMS/ChatGPT cited by secondaries, NOT on official page — unconfirmed.) (sierra.ai/product↗)
4. Voice (core strength) — Production inbound + outbound; interruptions, noise, accents, sentiment; replaces IVR. Voice payments (card + ACH over phone on PCI DSS Level 1). Boosted by Mar 2026 Receptive AI acquisition. (sierra.ai/product/voice↗)
5. Avatar/embodiment — None. Text + voice only.
6. Persona/brand — First-class: policies, tools, guardrails, knowledge, and brand/tone are separable; supervisors+filters act as the "constitution." (Enterprise-grade agents↗)
7. Lead/CRM/commerce — Real backend actions mid-convo: refunds, subscription updates, order/exchange management, balance checks, identity verification, save-the-cancellation, upsell/cross-sell. (Agent SDK↗)
8. Analytics/insights — Insights 2.0 + Explorer + Experience Manager; NL querying over conversation data; CSAT/resolution/business-impact. (Insights↗)
9. Handoff/HITL/inbox — Intelligent escalation w/ auto-generated summaries; Live Assist real-time copilot for human reps. (Live Assist↗)
10. A/B testing — Yes — A/B of conversation designs + behavior variants, backed by a simulation engine (agent/user/judge triads, ~35,000+ simulated tests/day, CI/CD-gated via GitHub Actions). (Simulations↗)
11. Integrations/API/marketplace — Agent SDK + Integration Library. Gap: integrations are custom API projects, NOT marketplace installs; SDK/docs/API gated to contracted customers — no public API/sandbox/GitHub. (eesel↗)
12. Security/compliance — Deepest in the field per their Trust page: SOC 2, HIPAA, GDPR, PCI DSS L1, FedRAMP High, CCPA, CSA STAR, ISO 27001, ISO 42001. (Trust↗)
13. Proactive/triggers — Agent Data Platform gives memory + company context for proactive outreach (renewals, preemptive fixes) + outbound campaigns. (Agent Data Platform↗)
14. Multilingual — 58 languages (official; some pages say 55) with mid-conversation switching. (sierra.ai/product↗)
15. Deployment/install — Managed cloud, enterprise onboarding (weeks–months). Build via Agent Studio (no-code) or Ghostwriter (plain-English agent authoring, ingests SOPs/transcripts/audio/photos); GitHub-style versioning + staging→prod + rollback.
16. Packaging — Outcome-based, fully sales-negotiated, no public tiers.

Self-hostable? EU residency? Self-host: NO (managed SaaS, US-centric). EU residency: uncertain — one secondary claims US/EU/multi-region options; not corroborated on official pages; another says EU specifics undisclosed/case-by-case. Treat as likely-available-but-contract-negotiated and undocumented — verify directly. (Rasa↗, Vetted↗)

Top differentiators: (1) Outcome pricing category-definer. (2) Production voice + voice payments (PCI L1) strengthened by Receptive AI. (3) Best-in-class agent-design + reliability engineering (Studio/Ghostwriter + supervisor constellation + 35k tests/day) and deepest compliance (FedRAMP High + ISO 42001). Notable gaps: (1) No self-host, opaque EU residency, US-centric. (2) Closed/gated ecosystem — no public API/SDK/marketplace, every integration bespoke, weeks–months deploy. (3) Pricing opacity + renewal re-pricing risk; no avatar.

3. Decagon

Positioning (one line): "The AI concierge for every customer" — enterprise conversational-AI support, deliberately repositioned from "AI customer support" to "concierge." (decagon.ai↗)

Company / funding / scale: Founders Jesse Zhang (ex-Lowkey→Niantic) + Ashwin Sreenivas (ex-Palantir, Helia→Scale AI); founded Aug 2023. $250M Series D (Jan 28, 2026) at ~$4.5B (Coatue + Index) — valuation tripled from the $1.5B Series C; ~$481M raised total; reconfirmed at $4.5B in a Mar 2026 tender. 300+ employees, >100 new enterprise customers in 2025, ~$35M ARR (Sacra estimate). Customers: Chime, Duolingo, Notion, Substack, Eventbrite, Bilt, Rippling, Curology, Webflow, Oura, Avis Budget Group, Deutsche Telekom, Affirm, Block. (Brief's "Hertz" is wrong — the confirmed rental logo is Avis.) (Bloomberg↗, Sacra↗, Series D↗)

Pricing: Not public — contact sales. Two outcome/usage structures: per-conversation (most popular; ~$0.99 third-party est.) or per-resolution (~$0.50–1.50 est.). Contract sizes (Vendr) ~$95K–590K/yr, median ~$386–400K — enterprise-only floor. (eesel↗)

16-category feature inventory:

1. Conversational AI/LLM — Multi-model (OpenAI/Anthropic/Cohere + own fine-tunes; in-house models reportedly ~80% of traffic by Mar 2026; selection is Decagon-managed, not a customer toggle). Flagship = Agent Operating Procedures (AOPs) — NL SOPs that compile to executable agent logic, run once across chat/email/voice. Supervisor model + Watchtower QA + Trace View observability. (AOP↗)
2. Knowledge/RAG — Ingests help centers/Drive/SharePoint/tickets/CRM/APIs/Shopify/Stripe into a Unified Knowledge Graph; agentic RAG w/ citations; knowledge-gap detection + auto-draft articles. (Knowledge↗)
3. Omnichannel — chat, email, voice, SMS with cross-channel memory. Gap: no native WhatsApp/Instagram/Messenger/social. (Voiceflow↗)
4. Voice — "Decagon Voice" — Mature: Voice 2.0 GA Sept 2025 (65% faster gen latency, barge-in, inbound+outbound, same AOP brain→real actions); outbound proactive voice + "Missions" campaigns added Spring 2026. Chime runs 1M+ calls/mo at ~70% resolution. (Voice 2↗)
5. Avatar/embodiment — None.
6. Persona/brand — Configurable tone/vocabulary/length/formatting + personality presets; voice has self-serve style/speed/pronunciation. (Tone↗)
7. Lead/CRM/commerce — Real transactional actions (refunds+labels, subscription up/down/cancel, address/payment updates, scheduling, credits) across Zendesk Sunshine/Salesforce/Shopify/Stripe. Gap: support-first, NOT a lead-gen/sales product. (Capabilities↗)
8. Analytics/insights — Admin Dashboard + AOP Copilot/Duet builder + Watchtower always-on QA (scores every AI+human convo on custom rubrics) + Insights/VoC. (Watchtower↗)
9. Handoff/HITL — Routing module escalates to live agents via helpdesks; Agent Assist co-pilot (drafts/summaries/translation/post-call docs). No own inbox (embeds in helpdesk); Agent Assist reportedly Zendesk-only (secondary, unconfirmed). (Agent Assist↗)
10. A/B testing — Genuinely mature — distinct Experiments product (traffic split, universal control, p-value config, ramp/rollback) + Simulations from real failure transcripts + CI/CD + regression replay. (Experiments↗)
11. Integrations/API — Salesforce, Zendesk, Intercom, Kustomer, Confluence, Guru, Slack, Shopify, Stripe, Amazon Connect, RingCentral/SIP, ServiceNow; public self-serve APIs + custom tools + MCP support. No public connector marketplace; docs gated. (Integrations↗)
12. Security/compliance — SOC 2 Type II, HIPAA (BAA), GDPR (DPA); AES-256/TLS 1.2+; zero-day retention / no-train commitment with OpenAI+Anthropic; PII redaction via Google DLP. ISO 27001 + PCI-DSS appear genuinely ABSENT — a real procurement friction. (Security↗)
13. Proactive/triggers — YES but newer (Spring 2026): AI-initiated calls, "Missions" outbound campaigns, lifecycle/event triggers — voice-centric and recent. (Outbound voice↗)
14. Multilingual — Claims "any language" + auto-detect + real-time translation; no fixed number — strongest verified data point is a Rituals case study at 15 languages. "Any language" is marketing; ~15 is the validated real count.
15. Deployment/install — Sales-led only, no trial/signup. Web widget + connectors + API/MCP; white-glove (dedicated Agent Engineers, SOPs→AOPs), ~4–12 weeks. On Google Cloud Marketplace (procurement only).
16. Packaging — Outcome/usage-based, contact-sales, ~$95K–590K/yr.

Self-hostable? EU residency? Self-host: NO — confirmed cloud-only; no on-prem/VPC/single-tenant. Trap: "deploy within your Google Cloud environment" is procurement/billing language (committed-spend drawdown), NOT a real customer-VPC — compute runs in Decagon's managed env. EU residency: claimed-available but undocumented (no named EU region/SKU). (Knowlee↗, GCP Marketplace blog↗, Security↗)

Top differentiators: (1) AOPs — cleanest "non-technical CX team owns the agent" story. (2) Production-grade experimentation/QA loop (Experiments + Simulations + Watchtower + supervisor). (3) Enterprise traction + capital ($4.5B, 100+ blue-chip logos, no-train commitment). Notable gaps: (1) Cloud-only, no self-host/VPC; ISO 27001 + PCI-DSS absent. (2) No social channels; support-only (no lead-gen). (3) No avatar; EU residency undocumented; proactive still young; heavy enterprise-only motion (no trial, $95K+ floor, 4–12wk onboarding).

4. Ada (ada.cx)

Positioning (one line): Enterprise "AI Customer Service Agent" that autonomously resolves inquiries across channels via its Reasoning Engine — branding the "Automated Customer Experience (ACX)" category. (ada.cx/platform↗)

Company / funding / scale: Toronto, founded 2016 by Mike Murchison + David Hariri. $130M Series C (May 2021) at ~$1.2B (Spark Capital + Tiger Global); ~$190–200M total. No newer round confirmed — 2026 valuation unverified. Claims 4B+ automated interactions since 2016; customers Meta, Verizon, Square, Zoom, Shopify, Wealthsimple, AirAsia, YETI, TELUS. ICP gate: their pricing page says best fit is 300,000+ annual conversations — mid-market/enterprise only. (RBCx↗, ada.cx/pricing↗)

Pricing: Not public. Notably, Ada is retreating from per-resolution toward per-conversation pricing and now publicly argues conversation-based gives "stronger alignment." Third-party estimates: ~$1–$3.50/resolution historically; annual ~$30K start / ~$70K median / $300K+ large. (Pricing model blog↗, Vendr↗)

16-category feature inventory:

1. Conversational AI/LLM — Reasoning Engine (unified intelligence layer; intent→knowledge/action→ground→plan→draft→safety-check). Claimed AR (Automated Resolution) >70% baseline, up to ~80–84%. Caveat: "model-agnostic" is marketing-soft — publicly "powered by Microsoft Azure OpenAI Service"; no documented multi-vendor router. (Reasoning Engine↗, Azure OpenAI↗)
2. Knowledge/RAG — Ingests Zendesk/Salesforce/Contentful/Guru/Helpjuice/Paligo etc.; "Coaching" = HITL refinement loop teaching corrections applied automatically. (Coaching↗)
3. Omnichannel — web chat, in-app, email, SMS, WhatsApp, Instagram/social, voice/phone, custom with context carryover. (Messaging↗)
4. Voice — Ada Voice: full phone channel (barge-in, noise suppression, live data, context-transfer handoff); CCaaS integrations Twilio Flex/Genesys/NICE CXone/Amazon Connect/Aircall/Avaya. Voice languages lag text (~7–42 depending on source — unresolved). (Voice↗)
5. Avatar/embodiment — None.
6. Persona/brand — Brand tone/persona config; Coaching refines tone; multi-layer guardrails; Playbooks encode multi-step SOPs.
7. Lead/CRM/commerce — Revenue/upsell use-case; connects order-mgmt/catalog/loyalty/CDP; CRM Actions (Salesforce). Customer-cited revenue impact. (Revenue↗)
8. Analytics/insights — Performance Center (AR tracking, CSAT, transcript review, coaching-opportunity surfacing). (Reports↗)
9. Handoff/HITL — Handoffs route to humans (live chat/call transfer or async ticket) into Zendesk/Salesforce/Genesys. (Handoffs↗)
10. A/B testing — Yes — native A/B of Proactive Campaigns and answer flows across segments. (myAskAI↗)
11. Integrations/API/marketplace — ~22–24 named integrations (Genesys, NICE, Amazon Connect, Salesforce, ServiceNow, Twilio, Kustomer, Gorgias, etc.); Actions + APIs + SDKs + an MCP Server. (Integrations↗)
12. Security/compliance — Trust Center: SOC 2 Type 2, SOC 3, HIPAA (BAA), PCI DSS, GDPR, CCPA/CPRA, PIPEDA, VPAT; TLS 1.2+/AES-256, RBAC/MFA. ISO 27001 uncertain (cited by reviewers, not clearly on Trust Center). (security.ada.cx↗)
13. Proactive/triggers — Proactive Campaigns (outbound/triggered, A/B-testable).
14. Multilingual — Text: 50–60+ languages w/ auto-detect + real-time translation; voice far fewer.
15. Deployment/install — Web/in-app embed + SDKs (incl. Zendesk Messaging SDK); APIs/SDKs/MCP; CCaaS plug-ins for voice.
16. Packaging — Sales-gated, enterprise-only (300K+ convos/yr), conversation/resolution-volume based.

Self-hostable? EU residency? Self-host: NO (AWS cloud-only). EU residency: YES — this corrects the prior assumption. Ada's privacy policy states hosting on AWS in US, Canada, Ireland, AND Germany — so EU (Ireland/Germany) and Canada residency are available. (One reviewer claimed US-only on Enterprise — conflicts with the official policy; the policy wins, but confirm region-pinning in contract.) (Privacy policy↗)

Top differentiators: (1) High AR + enterprise proof (Meta/Verizon/Square/Shopify, 4B+ interactions). (2) Genuinely unified omnichannel + voice with deep CCaaS integration. (3) Coaching + Playbooks + A/B loop for non-engineers. (Bonus: multi-region incl. EU + Canada residency — stronger GDPR posture than most US rivals.) Notable gaps: (1) Opaque, sales-gated, enterprise-only pricing (300K ICP shuts out SMB). (2) No self-host. (3) Voice multilingual lags text; no avatar; "model-agnostic" is marketing-soft (Azure OpenAI in practice).

5. Zendesk AI / Agentic

Positioning (one line): Incumbent omnichannel CX suite repositioned around the "Zendesk Resolution Platform" — an "autonomous service workforce" of AI agents priced per resolved ticket. (CMSWire↗)

Company / scale: Taken private Nov 2022 by Hellman & Friedman + Permira for ~$10.2B; 100,000+ brands; revenue ~$1.9B (private, estimates); AI ARR ~$200M exiting 2025 → projected $500M in 2026. AI build-out via M&A: Klaus (QA), Tymeshift (WFM), Ultimate.ai (agents, 2024), Local Measure (voice/IVR, 2025), HyperArc (GenAI analytics, 2025), and Forethought (self-learning agentic AI — Mar 2026, largest acquisition in ~20 years). (Permira↗, Forethought↗, AI ARR↗)

Pricing (two layers): Per-agent Suite seats: Team ~$55, Growth ~$89, Professional ~$115, Enterprise = contact-sales (billed annually). Advanced AI add-on (historically $50/agent/mo) is being absorbed into base plans (May 2026). Headline shift = outcome-based "automated resolution": from $1.50/resolution (~$2 PAYG) with a three-tier model where only "Verified Resolutions" (AI resolved AND an independent LLM evaluator confirmed) are billed — Assisted Escalation + Contained Resolution are free. Included allowances ~5/10/15 resolutions per agent/mo by tier. (Pricing↗, Outcome pricing↗, premiumplus↗)

16-category feature inventory:

1. Conversational AI/LLM — Zendesk AI Agents (Answer Bot→Ultimate→now Forethought-powered "self-learning") on the Resolution Platform, trained on ~20B ticket interactions, with a Resolution Learning Loop (detects gaps, generates procedures, tests before deploy) + no-code Agent Builder. (CMSWire↗)
2. Knowledge/RAG — Help Center feeds generative replies; content cues + AI-generated procedure drafts (3 auto-created within 24h). (What's new June 2026↗)
3. Omnichannel (breadth leader) — messaging, email, live chat, voice, SMS, WhatsApp, social, communities, review sites, help centers with shared context. (Newsroom↗)
4. Voice — Zendesk Talk + Voice AI Agents (Local Measure acq.); 60+ languages, mid-call switching, ~80% claimed autonomous resolution; Copilot real-time transcription/summaries/suggestions. (Voice↗)
5. Avatar/embodiment — None.
6. Persona/brand — Brand voice/tone + "Enhance writing"; no digital-human persona system. (Copilot↗)
7. Lead/CRM/commerce — Zendesk Sell exists but is a deprioritized/legacy CRM; limited commerce vs dedicated CRMs.
8. Analytics/insights — Explore (BI, being supplanted by HyperArc GenAI analytics) + Klaus (AutoQA/VoiceQA) + Tymeshift (WFM). (HyperArc↗)
9. Handoff/HITL/inbox — Agent Workspace (unified cross-channel ticket view) + Zendesk Copilot (macro/article suggestions, Auto Assist actions, summarization, smart triage). (Copilot↗)
10. A/B testing — No prominent user-facing A/B product; the Learning Loop is internal self-tuning (gap/uncertain).
11. Integrations/API/marketplace (leader) — Marketplace ~1,900+ apps, 1,100+ partners; REST API + ZIS + ZAF apps framework. (Marketplace↗)
12. Security/compliance (deepest cert stack) — SOC 2 Type II; ISO 27001:2022/27017/27018/27701; ISO 42001; FedRAMP LI-SaaS (Low only); HIPAA (BAA); PCI-DSS; GDPR/CCPA (BCRs); Cyber Essentials Plus; CSA STAR AI; HDS. (Trust Center↗)
13. Proactive/triggers — Proactive messages on Web Widget + mobile SDK. (Proactive↗)
14. Multilingual (breadth leader) — AI Agents 80+ languages native, 100+ for generative replies; voice 60+.
15. Deployment/install — Web Widget + iOS/Android Messaging SDKs; no-code agent setup; Employee Service AI Agents for Slack/Teams (early access).
16. Packaging — Per-agent seats + decoupling per-Verified-Resolution AI billing; Advanced AI add-on folding into base.

Self-hostable? EU residency? Self-host: NO (AWS cloud-only — a recurring reason buyers cite for open-source Zammad). EU residency: YES via "Data Center Location" add-on — choose US / EEA (EU) / Australia / Japan; included by default in Enterprise, an add-on for Professional (France gets HDS). (Trust Center↗, eesel↗)

Top differentiators: (1) Incumbent scale + ecosystem (100k brands, ~1,900 marketplace apps, widest omnichannel). (2) Outcome pricing as a CFO-friendly category bet with dual AI+independent-evaluator validation. (3) Aggressive AI consolidation via M&A (Ultimate+Forethought+HyperArc+Klaus+Tymeshift+Local Measure into one platform). Notable gaps: (1) No self-host; EU residency is a paid add-on below Enterprise. (2) Pricing opacity + cost creep (seats+add-ons+overages). (3) No avatar; weak native A/B; stitched-together M&A stack (coherence risk); FedRAMP Low-Impact only.

6. Shorter notes — Forethought + Kustomer

Forethought (forethought.ai) — NOW PART OF ZENDESK

Positioning/ownership: "Enterprise AI Agents for Every Customer Moment" — fully-agentic support automation trained on a company's own ticket history (SupportGPT lineage). CEO Deon Nicholas; $65M Series C (Dec 2021, Steadfast lead), ~$115M total. Acquired by Zendesk (Mar 2026, all-cash, undisclosed) — now inside Zendesk's Resolution Platform. Customers: Upwork, Thumbtack, Carta, Lime, Crunchbase. (Zendesk acq↗, TechCrunch↗) Pricing: Hybrid platform-access fee + outcome component + usage; tiers Team/Professional/Enterprise, all quote-only. (Pricing↗) 16-cat highlights: (1) Agentic multi-agent (Solve/Triage/Assist/Discover) + late-2025 Browser Agent that drives any web UI without APIs; (2) RAG over your own tickets/macros/CRM (not just public docs); (3) chat/email/voice/SMS/API + Slack; (4) native voice automation; (5) no avatar; (6) brand-aligned replies, multibrand (≤20); (7) not a CRM — rides helpdesks; (8) Discover (knowledge gaps) + QA Agent; (9) Triage routing + Assist copilot (browser ext); (10) A/B weak/unclear; (11) helpdesk/CRM integrations + headless API; (12) SOC 2 II, ISO 27001, HIPAA, GDPR/CCPA, no ISO 42001/PCI; (13) workflow automation, not strong outbound; (14) 100+ languages detect / 28 conversational; (15) SaaS over existing helpdesk; (16) quote-only. Self-host/EU: No self-host; no documented EU residency (US-cloud). Post-Zendesk it would inherit Zendesk's higher-tier EU options, not yet documented for Forethought's pipeline. Differentiators: Browser Agent (no-API legacy systems) + trains on your own resolution data. Gaps: acquisition/roadmap absorption into Zendesk; not a system of record; no EU/self-host.

Kustomer (kustomer.com) — INDEPENDENT AGAIN

Positioning/ownership: CRM-centric AI-native CX platform unifying conversations on one customer timeline. Meta acquired for ~$1B (2020), divested back to founders Brad Birnbaum + Jeremy Suriel May 2023 (~$250M); $30M Series B (2025, Norwest lead); backed by Battery/Redpoint/Boldstart. (TechCrunch spin-out↗, Series B↗) Pricing (PUBLIC): Seat — Enterprise $89/seat/mo, Ultimate $139 (8-seat min) OR conversation — $0.35/$0.50 per conversation. AI à la carte: "AI for Customers" $0.60/engaged conversation, "AI for Reps" $40/user/mo. Notably charges per conversation, not per resolution. (Pricing↗, Gorgias guide↗) 16-cat highlights: (1) AI-native; Concierge AI agents + no-code AI Agent Studio (KIQ/KustomerGPT lineage); (2) RAG over KB/FAQ; (3) chat/email/voice/social on one timeline; (4) voice supported (AI less detailed); (5) no avatar; (6) configurable tone/length/brand personality; (7) core strength: full CRM/data platform (custom objects, order/purchase history, system of record); (8) Data Explorer (NL analytics) + Envoy/Signals (frustration/urgency, staffing/coaching); (9) native inbox + Envoy agent-assist; (10) A/B weak/unclear; (11) apps + webhooks + MCP connectivity; (12) SOC 2 II, ISO 27001:2022, HIPAA paid add-on, EU-US Data Privacy Framework (GDPR mechanism, not in-region hosting); (13) proactive outbound + Architect journey builder; (14) "multiple languages/dialects" (no count); (15) SaaS + Mar 2026 standalone "Kustomer AI" layer to modernize existing helpdesks without migration; (16) seat OR per-conversation + AI à la carte. (Platform↗, Trust↗) Self-host/EU: No self-host (no VPC/on-prem/sovereign); no in-region EU residency (GDPR via Data Privacy Framework only). US-cloud. Differentiators: CRM+service unified on one timeline (true system of record); flexible seat-or-conversation pricing + new standalone AI layer. Gaps: AI unbundled + pay-per-conversation (costs stack, weaker outcome alignment); no self-host/EU/sovereign; HIPAA only paid add-on; smaller post-Meta scale.

7. Comparison Table (platforms × 16 categories, 1-line cells)

The two ★ rows are the whole strategic story: every incumbent is "No" on self-host, and EU residency is at best a bolt-on. Nobody has an avatar (row 5 is "None" seven times).

8. What OpenAlice should learn / where we already win

Where we structurally WIN (defensible wedge — protect & lead with it)

1. EU-sovereign self-hosting is a category-wide blind spot. All seven are US-cloud-only SaaS with zero self-host/on-prem/VPC option. EU residency is, at best, a leaky paid add-on (Fin routes email+billing through US; Zendesk gates it below Enterprise; Sierra/Decagon undocumented). Only Ada offers genuine EU+Canada in-region hosting — and even Ada can't be self-hosted. For any EU-regulated, air-gapped, data-sovereignty, or public-sector buyer, the entire incumbent field is disqualified and we are uncontested. This is our single strongest differentiator — lead every enterprise-EU conversation with it.
2. Embodied avatar on a platform — nobody else has one. Row 5 of the table is "None" across all seven. The whole field is text+voice. Our VRM/embodied-avatar layer is a genuine product axis they don't compete on — valuable for brand/persona-heavy, retail, streaming, and "face of the brand" use-cases where a talking avatar beats a text widget.
3. Own-the-stack control. Fin's move to a proprietary vertical model (Apex) validates the thesis that owning the model/runtime beats reselling frontier APIs on cost/latency/control. Our self-hosted-model posture is the same bet, taken further (the customer can own the whole stack, not just us).

Where we LAG (must close to be credible at enterprise CX)

1. Omnichannel breadth. Zendesk/Ada/Fin span chat+email+voice+SMS+WhatsApp+social with cross-channel continuity. We need a credible multi-channel story (at minimum chat+email+WhatsApp+voice with shared context) — this is table-stakes for enterprise CX RFPs, not a nice-to-have.
2. Multilingual depth + auto-detect. Zendesk claims 80+/100+, Sierra 58, Fin 45, Ada 50+ — all with auto-detection. Multilingual breadth with mid-conversation switching is expected. We must quantify and publish our language count.
3. Marketplace / integration ecosystem. Zendesk's ~1,900 apps is a moat; even the startups expose APIs/MCP and CCaaS connectors. We need a published integration catalog + a public API/SDK + MCP support (note: MCP is now standard — Decagon, Ada, Kustomer all support it; we should too).

Tactical lessons worth copying

• Outcome-based pricing is now table-stakes, not a differentiator — Fin ($0.99/res), Zendesk ($1.50/Verified Resolution w/ independent-LLM validation), Sierra, Decagon all do it. We should offer it as an option, but our actual pricing wedge is "you host it, you own the data, predictable infra cost — no per-resolution metering at all." That's a sharper story for cost-conscious EU buyers than competing on $/resolution.
• The "independent evaluator confirms resolution before billing" pattern (Zendesk) + simulation/eval harnesses (Sierra 35k tests/day, Decagon Experiments) are the maturity bar for trustworthy autonomy. We need a real eval/QA + A/B + simulation loop — and notably A/B testing is a genuine gap for Fin, Zendesk, Forethought, Kustomer, so a strong native experimentation surface is a place we can over-deliver vs incumbents, not just catch up.
• "NL SOPs compile to agent logic" (Decagon AOPs, Sierra Ghostwriter, Forethought) is the winning authoring UX — let non-engineer CX teams own the agent. Worth emulating for our persona/agent-config layer.
• Watch the consolidation wave: Salesforce→Fin, Zendesk→Forethought, the M&A roll-up means the independent field is shrinking. The sovereign/self-host/embodied lane is precisely the lane the consolidators structurally cannot enter (their whole model is managed US cloud) — so it's both our wedge and our moat.

Source coverage & honesty notes

Every platform section above is grounded in official product/pricing/security/trust pages plus reputable secondaries (TechCrunch, Bloomberg, CNBC, CMSWire, Forrester, Futurum, Sacra). Soft/uncertain claims explicitly flagged in-text: all per-resolution dollar figures except Fin's $0.99 and Zendesk's ~$1.50 are third-party estimates (vendors gate pricing); vendor resolution-rate % and ARR are unaudited; Sierra/Decagon EU residency is undocumented (verify directly); Ada/Forethought ISO-27001 and Decagon's missing ISO-27001/PCI are notable. The two load-bearing conclusions — (a) none are self-hostable, (b) none have an avatar — are confirmed by absence across all official sources reviewed and are high-confidence.