ooligo

Cresta

ai-customer-experience agent-assist · conversation-intelligence · voice-ai
AI-NATIVE API
Customer Success
8.0 /10

What it is

Cresta now calls itself “the AI operating system for customer experience”, and the product line has grown to match. Four products sit on one set of conversation models and quality rubrics: AI Agent, which handles a voice or chat contact end to end; Agentic Assist (renamed from Agent Assist in 2026), which pushes guidance, knowledge and summaries to a human rep during the live conversation; Conversation Intelligence, which scores every interaction instead of a sample; and Deputy Agent, a long-running agent for back-office case work, launched 13 October 2026 in limited availability. The pitch is unchanged: you are not buying a deflection vendor and an enablement vendor and hoping their definitions of “resolved” agree.

The fact that changes the buying decision: on 2 November 2026 Salesforce signed a definitive agreement to acquire Cresta. Close is expected in Salesforce’s first fiscal quarter of 2028, which starts in February 2027, subject to regulatory approval. Salesforce did not disclose a price. It says Cresta will be combined with Service Cloud and that, after close, Cresta “will continue to support the leading contact center and telephony platforms” its customers use today. Until close, Cresta is an independent company selling under its own contracts.

Cresta was founded in 2017 out of the Stanford AI Lab by Sebastian Thrun, Zayd Enam and Tim Shi, and is run by CEO Ping Wu. It closed a $125M Series D in November 2024, has raised more than $270M in total, and told Axios on 30 April 2026 that it had passed $100M ARR. Named customers skew large and voice-heavy: United Airlines, Alaska Airlines, Marriott, Comcast, Cox, CVS, Brinks Home, Vivint, Propel and Snap Finance.

Why it shows up in CX stacks

  • It runs the human queue and the AI queue under one rubric. Assist, AI Agent and QA share the same scoring, so a contact is judged the same way whether a rep or an agent handled it. If your plan is to run humans and AI side by side for a year, that shared rubric is what you actually need.
  • Knowledge Agent answers from the customer’s live context. Part of Agentic Assist, it pairs live listening with the account on screen and returns a sourced answer for that customer. It targets the time a tier-1 rep loses switching between CRM, knowledge base and billing console.
  • Agent development has a real engineering loop. Conductor (June 2026) is a build-and-test environment that runs agents against Synthetic Customers, personas generated from your own historical conversations. The Cresta CLI (27 August 2026) lets developers and coding agents run regression tests and gated deployments from CI/CD. On 29 October 2026 Cresta opened that stack to customer developers, built on Anthropic’s Claude Agent SDK.
  • Journey Intelligence (12 November 2026) links contacts into journeys. It surfaces repeat-contact drivers and predicts the next contact, then hands that history to reps (Customer Snapshot) and to AI Agent memory.
  • It sits on the CCaaS you already bought. Cresta’s integrations page lists NICE, Genesys, Five9, Amazon Connect, Avaya, Cisco, Twilio and LivePerson, plus Salesforce, ServiceNow, Zendesk, Pega and Microsoft Dynamics on the CRM side.
  • The compliance set fits regulated buyers. The trust centre lists ISO/IEC 42001, 27001 and 27701, SOC 2 Type II, HIPAA, PCI DSS, GDPR, CCPA, TISAX and AIUC-1 for AI Agent.

Pricing reality

Cresta publishes no rate card and has no self-serve price. The only public anchor is its AWS Marketplace listing, re-checked on 17 November 2026 and unchanged: Agent Assist for Chat at $150,000 for 12 months covering up to 125,000 chats, then $1.20 per extra chat, and Agent Assist for Voice at $150,000 for 12 months covering up to 100,000 calls, then $1.50 per extra call. Subscriptions are non-cancellable and non-refundable. The listing still uses the old Agent Assist product names.

Two things follow from that listing. First, the meter is linear: $150,000 over 125,000 chats is exactly $1.20 a chat, so there is no volume discount for committing to more. Second, the $1.50 per voice call buys help on a call that a paid human still handles. AI-agent vendors such as Intercom and Zendesk bill per automated resolution, and that unit removes the labour cost. Normalise the unit before you compare the headline numbers.

Buying both voice and chat assist at list implies about $300,000 a year before overages, AI Agent, Deputy Agent or QA. Cresta describes Deputy Agent pricing as outcome-based and annual, with no published figure. Third-party 2026 guides put whole contracts at roughly $60,000 to $150,000 a year for smaller deployments. Those are estimates, not Cresta quotes.

Best for

Enterprise CX and contact-centre leaders running roughly 300 seats or more, mostly voice, in a regulated or high-value-per-contact operation: telco, airlines, insurance, home security, consumer lending, hospitality. It fits best when the next twelve months are about rep productivity, QA on every interaction and a controlled hand-off of easy contacts to AI, all reported against one quality rubric. Salesforce Service Cloud shops have an extra reason to look now; teams on a competing CRM have an extra reason to check exit terms.

Skip it if your queue is mostly simple chat, if the goal is pure deflection with no assist layer, if you need pricing you can approve without procurement, or if you handle fewer than roughly 50,000 contacts a year, where a $150,000 commitment cannot pay back.

Versus the alternatives

Most installed volume sits with the CCaaS and workforce-engagement suites, NICE CXone, Genesys Cloud and Verint, whose AI is bundled into contracts you already signed. Take the bundled option when your existing QM licences already cover scoring and a second platform cannot clear the finance bar. Sierra, the fastest-growing entrant in the segment, and Decagon are the picks when containment is the goal and no assist layer is planned. Parloa wins when telephony integration and a per-minute meter decide the deal. Observe.AI is the cheaper like-for-like on assist plus QA for mid-market voice teams. Salesforce Agentforce Contact Center is the option to watch if you are all-in on Salesforce, since Cresta is likely to end up inside it. Cresta’s case is narrow: it runs the human queue and the AI queue under one supervisor view and one rubric. If you will not operate both, you are paying for half a platform.

Watch-outs

  • The owner changes mid-contract. A deal signed now will run into Salesforce ownership in 2027, and Salesforce has said nothing yet about standalone pricing, packaging or how existing contracts carry over. Guard: add a change-of-control clause that lets you exit or renegotiate if pricing, CRM neutrality or data-processing terms change, cap renewal increases, and get Salesforce’s continued support for your CCaaS and CRM in writing, not just the press-release line.
  • The commercial model bills seats while the product removes them. Assist is priced on agent seats and assisted conversations, and AI Agent success shrinks that base. Guard: negotiate one blended pool across assist and AI Agent spend, write the conversion ratio into the contract, and refuse a multi-year seat floor that your own automation plan will breach.
  • Every outcome number on the site is vendor-published. Brinks Home’s 30-point NPS gain, Cox’s 20% revenue lift, Aqua Finance’s 61% collections lift and Propel’s 58% containment are Cresta customer stories, unaudited, each with its own metric definition. Guard: run a two-week shadow on your own traffic, score it against your existing FCR and containment definitions, and make that the contract’s acceptance test.
  • Real-time assist is continuous monitoring of employees. Live listening plus screen context means capture on every agent, every shift, and the EU AI Act bans inferring workers’ emotions in the workplace outside narrow exceptions. Guard: complete a DPIA before the pilot, open works-council consultation in DACH and France ahead of procurement, and get written confirmation of which models infer agent emotion versus customer sentiment, with the agent-facing ones off in EU deployments.
  • Much of what you will be shown is new. Conductor dates from June 2026, the CLI from August, Deputy Agent from October (still in beta) and Journey Intelligence from November. Guard: contract only for modules with reference customers at your seat count and in your vertical, and take the rest as a no-cost pilot addendum with a named exit.