PolyAI
Best-in-class enterprise voice AI, closed and priced accordingly: no public rate card, budget $150K+/year, and most changes still route through PolyAI's own team.
PRICING
| Enterprise | Custom | Sales-led contract, per-minute usage billed on top; third-party estimates from $150K+/year base |
Per-minute usage-based billing wrapped inside a negotiated annual contract. PolyAI does not publish a rate card, per-minute rate, or offer self-serve signup — every deployment goes through a sales and implementation cycle (reported 4-6 weeks).
checked 2026-08-05 · vendor pricing page
Element scores
Strengths
Raven v2 is a proprietary LLM benchmarked specifically on customer-service dialogue, backed by genuine enterprise scale — 2,000+ live deployments and 100+ enterprise customers across hospitality, banking and insurance (Marriott, Caesars, UniCredit). The ADK gives a real code-first, git-based builder alongside the no-code Agent Studio, and simulation testing runs hundreds of LLM-judged conversations before anything reaches a live phone line.
Honest dings
Pricing is entirely sales-led with no public rate card — third-party estimates put base contracts at $150K+/year — and it's a managed-service model where most configuration changes still route through PolyAI's own team rather than a fully open customer sandbox. Multi-agent orchestration, fine-tuning, small model tiers and interpretability are all absent, since Raven is a single closed model with no customer-facing exposure.
Sources
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- poly.ai/blog/polyai-raises-86m-to-transform-how-ent… — Official: $86M Series D, $200M+ total funding, 100+ enterprise customers, 2,000+ deployments, 45 languages, 25+ countries, Forrester TEI 391% ROI figures (accessed 2026-08-05)
- poly.ai/blog/polyai-raven-v2-large-language-model — Official: Raven v2 proprietary LLM, internal benchmark claims on function calling/instruction following, latency-sensitive reasoning as future work (accessed 2026-08-05)
- docs.poly.ai/home — Official docs: Agent Studio vs ADK architecture, Behavior/Knowledge/Flows/Voice building blocks (accessed 2026-08-05)
- docs.poly.ai/integrations/introduction — Official docs: MCP integration (client-side, auto tool discovery), API/Managed services integration paths (accessed 2026-08-05)
- docs.poly.ai/api-reference/introduction — Official docs: three REST API families (build/run/observe), 99.994% stated production uptime (accessed 2026-08-05)
- polyai.mintlify.app/simulation-testing/introduction — Official docs: simulation testing methodology (scenario + assertions + LLM judge) (accessed 2026-08-05)
- siliconangle.com/2025/12/15/call-center-chatbot-sta… — Independent: Series D $750M valuation, 25x revenue multiple context, Cambridge dialogue-systems origin (accessed 2026-08-05)
- getvocal.ai/blog/polyai-pricing-vs-alternatives — Independent pricing analysis: $150K+/year base contract estimate, per-minute fees, no public rate card (accessed 2026-08-05)
- thoughtly.com/blog/polyai-review — Independent review: G2 rating (5.0/5, only 12 reviews), managed-service limitations, no LLM sandbox (accessed 2026-08-05)
- cxtoday.com/customer-analytics-intelligence/ai-hall… — Independent: CEO interview on hallucination mitigation approach, 50-75% autonomous call resolution claim (accessed 2026-08-05)
- pitchbook.com/profiles/company/227011-60 — Independent: 252 employees, 2017 founding year, London HQ, Series D deal size $85.2M (accessed 2026-08-05)