Cresta
Fine-tunes on your own conversations and coaches human agents in real time like nobody else, but the integration surface is closed — no MCP — and the tuning bill is paid in months of internal effort.
Element scores
Strengths
Real-time human coaching is the most proven surface: G2 reviewers confirm the in-call prompting changes agent behavior while the conversation is live. Fine-tuning on your own conversation data plus 20+ task-routed models is a real accuracy edge over prompt-only rivals. Opera gives ops teams a genuine no-code harness, and Knowledge Agent (03/2026) reads live audio plus on-screen context to serve cited answers unprompted. The CCaaS integration bench is the deepest in the field — cloud (Five9, Genesys, NICE, Amazon Connect) and on-premise (Cisco, Avaya) both supported. $100M ARR (04/2026) with United Airlines, Cox and Marriott in production.
Honest dings
No MCP or A2A anywhere in the stack — every integration is a bespoke connector, which is a closed surface in 2026. G2 (4.2/5 across 43 reviews) flags knowledge-suggestion accuracy misses and transcription inconsistency for accented callers. Tuning the AI to your business takes months of dedicated internal investment before the platform performs. Pricing is undisclosed and enterprise-only. And multi-agent coordination is thin: parallel products on a shared record, not agents working together.
Sources (9) — every claim traceable
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).
- cresta.com/llm-info — Cresta's canonical company facts page (updated 20/05/2026): Palo Alto HQ, founded 2017 from Stanford AI Lab, $280M+ raised, 500+ employees, product portfolio, Ping Wu CEO (accessed 2026-08-01)
- cresta.com/platform-overview — Official platform claims: 20+ LLMs in task-specific systems, fine-tuning on proprietary data, synthetic datasets, LLM-based evals, simulation and regression testing, guardrails (accessed 2026-08-01)
- cresta.com/opera — Opera no-code orchestration engine: GenAI Intents, Workflows, Analyzer — continuous model evaluation and precision tuning without code (accessed 2026-08-01)
- axios.com/pro/enterprise-software-deals/2026/04/30/… — Official: over $100M ARR, told to Axios exclusively by the company (30/04/2026) (accessed 2026-08-01)
- prnewswire.com/news-releases/cresta-launches-knowle… — Knowledge Agent launch (17/03/2026): ambient listening, Context Fields, browser-sidebar UX, cited answers; Ping Wu CEO quote (accessed 2026-08-01)
- cresta.com/blog/the-three-pillars-of-voice-integrat… — Voice integration depth (14/05/2026): SIP/PSTN/CCaaS-native streaming, cloud and on-premise support matrix, AI-to-human handoff architecture (accessed 2026-08-01)
- g2.com/products/cresta/reviews — Independent user signal: 4.2/5 across 43 reviews; recurring complaints on knowledge-assist accuracy and transcription of accented/bilingual callers; praise for real-time assist (accessed 2026-08-01)
- eesel.ai/blog/cresta-review — Independent (competitor-disclosed) review: pricing not publicly disclosed, customer list, 87% of G2 reviewers mid-market/enterprise, Agent Assist strongest capability (accessed 2026-08-01)
- gtmlens.com/cresta-1b-deep-dive — Independent analysis: $125M Series D at ~$1B post-money, positioning as regulated-vertical specialist vs Sierra/Decagon, tuning-effort and setup-time complaints (accessed 2026-08-01)