Sierra
by Sierra
Outcome-priced autonomous customer agents that act in your systems, not just answer questions
Sierra is an enterprise conversational-AI platform for building, deploying and supervising autonomous customer-facing agents across chat, voice, SMS, WhatsApp, email and ChatGPT. It is aimed at large consumer-facing brands that want agents to complete real transactions in back-office systems, and it charges per resolved outcome rather than per seat or per conversation.
Sierra is an enterprise conversational-AI platform that builds, deploys and supervises autonomous customer-facing agents. It was founded in 2023 by former Salesforce co-CEO Bret Taylor and former Google Labs and VR lead Clay Bavor, and launched publicly in February 2024 with SiriusXM, Sonos and WeightWatchers as reference customers. Rather than shipping a chatbot, Sierra runs what it calls Agent OS: a constellation of specialised models coordinated at runtime, a Supervisor layer that checks generated responses against scope and policy before they reach the customer, an Agent Data Platform that carries context across sessions and channels, a no-code Agent Studio for CX teams, and an Agent SDK for developers who need custom integrations. Agents deploy across chat, phone, email, SMS, messaging apps and ChatGPT, and are designed to execute real transactions in operational systems — upgrading a subscription, rescheduling a furniture delivery — rather than only retrieving answers. Three named products sit on top: Ghostwriter, an agent-building agent that generates multilingual production agents from SOPs, transcripts or plain-English descriptions; Insights, covering conversation analysis, proactive monitors, multivariate experiments and action-level observability; and Horizon, aimed at longer-horizon customer outcomes and proactive engagement. Commercially Sierra is unusual in charging per resolved outcome rather than per seat or per conversation, at roughly $1.50 a resolution on custom enterprise contracts that third-party analysis puts near $150,000 a year to start. The company reported roughly $200M ARR by mid-2026 and raised a $950M Series E in May 2026 at a $15.8B valuation led by GV and Tiger Global.
The VP or SVP of customer experience at a large consumer-facing brand with high ticket volume, who needs agents to complete transactions in order, billing and subscription systems rather than deflect questions, and has the budget for a six-figure managed engagement.
Support volume handled end-to-end by agents you only pay for when they actually resolve the interaction, so cost tracks results rather than headcount or conversation count.
At a Glance
- Category
- AI Agents & Orchestration
- Pricing
- Usage-based, Contact for pricing
- Target Market
- CIOs, CTOs, VP Customer Experience, Heads of Support, Enterprise Developers
- Deployment
- Cloud-only, API-based
- Founded
- 2023
- Customers
- More than 40% of the Fortune 50, with 30+ enterprise brands shown publicly including Rocket Mortgage, SiriusXM, Uber, Vanguard, Wayfair, Sonos, WeightWatchers and Sutter Health
Key Features
- ✓Agent OS runtime
Coordinates a constellation of specialised models per conversation instead of routing everything through one general model, which is what lets a single agent both reason and execute system actions.
- ✓Supervisor guardrail layer
Checks model output against scope and policy in real time before a response is sent, the control that separates a governed enterprise agent from a raw LLM wrapper.
- ✓Ghostwriter agent builder
Generates production-ready multilingual, multichannel agents from existing SOPs, call transcripts or plain-English descriptions, cutting the hand-authoring normally needed to stand an agent up.
- ✓Agent Studio and Agent SDK
A no-code builder for CX teams plus a code path for engineers, so business owners can change behaviour without waiting on a development cycle.
- ✓Insights, monitors and experiments
Conversation analysis, proactive issue detection, multivariate testing and action-level observability, which matter more than usual because billing depends on measured resolutions.
- ✓Omnichannel deployment and simulation
One agent definition runs across chat, phone, email, SMS and messaging apps, with simulation and debugging before release so regressions are caught pre-production.
- ✓Agent Data Platform
Unifies customer context across sessions, channels and connected systems so an agent resuming a conversation on a new channel keeps the prior state.
Capabilities
Use Cases
- •Subscription change and cancellation handling
An agent reads entitlement state, offers a retention alternative, and executes the downgrade or cancellation directly in the billing system without a human touch.
- •Delivery and order exception management
Handles rescheduling, damaged-item claims and partial shipments by reading the order-management system and writing the corrective action back to it.
- •Voice contact-center deflection
Fronts the phone queue, resolves routine calls end to end, and hands off to a human with full conversation context when escalation is warranted.
- •Proactive lifecycle outreach
Horizon-driven agents initiate contact ahead of renewal or churn risk, turning support infrastructure into a revenue-retention channel rather than a cost center.
- •Agent performance experimentation
Run multivariate experiments across agent variants and measure resolution rate directly, because the commercial model makes resolution the billed unit.
Ideal For
Best For
- ✓High-volume consumer support where agents must take actions in order-management, billing or subscription systems, not just answer FAQs
- ✓Voice and chat deflection programmes at brands already running large outsourced contact-center operations
- ✓Retention and save flows where an agent can offer a downgrade or credit instead of processing a cancellation
- ✓Omnichannel CX teams that need one agent definition running across chat, phone, email, SMS and ChatGPT
- ✓Enterprises that want vendor incentives tied to resolution rate rather than seats deployed
Not Ideal For
- ✗Teams that want to own and iterate on agent logic themselves — reviewers repeatedly report that prompt and logic changes route back through Sierra's team, slowing iteration
- ✗Small or mid-market support teams: contracts are custom-quoted with no self-serve tier and third-party analysis puts entry points near $150,000 a year
- ✗Organisations expecting fast time-to-value — implementation is measured in weeks to months, not the minutes a self-serve chatbot takes
- ✗Buyers who need out-of-the-box marketplace connectors for Zendesk, Intercom, Freshdesk or Salesforce, since integrations are built per customer through the Agent SDK
- ✗Latency-critical voice deployments where multi-model verification adds round-trip time to every turn
Integrations
Deployment
Market & Ratings
More than 40% of the Fortune 50, with 30+ enterprise brands shown publicly including Rocket Mortgage, SiriusXM, Uber, Vanguard, Wayfair, Sonos, WeightWatchers and Sutter Health
Market Analysis
Pros
- ✓Pricing is tied to measured resolutions, so an underperforming deployment costs less rather than more — a genuinely rare vendor incentive in enterprise CX
- ✓Agents execute real transactions in operational systems rather than deflecting to a knowledge base, which is the capability most chatbot projects fail to reach
- ✓One agent definition spans chat, phone, email, SMS, messaging apps and ChatGPT, avoiding per-channel rebuilds
- ✓Reported traction is unusually strong for the category: roughly $200M ARR by mid-2026, from launch in February 2024
Cons
- ✗No published pricing at all — buyers must complete a full sales cycle just to get a quote, a complaint that recurs across public reviews
- ✗Limited self-service control after go-live: changing agent logic or prompts frequently means going back to Sierra's team, which slows iteration
- ✗No native marketplace connectors for Zendesk, Intercom, Freshdesk or Salesforce; every integration is custom work through the Agent SDK
- ✗Multi-model verification adds latency, and in live voice a sub-second delay is enough to create awkward pauses
- ✗Brand-safety risk is real and demonstrated: in late 2025 a coordinated jailbreak campaign got Gap's Sierra-powered chatbot to respond on prohibited topics because Gap's guardrails had been misconfigured, and Sierra's CEO publicly apologised
- ✗Onboarding is a managed-service engagement measured in weeks to months, with a learning curve reviewers describe as steep
Pricing
Outcome-based enterprise contract
Contact for pricing
- ✓Charged per resolved interaction rather than per seat or per conversation
- ✓Custom-negotiated annual commitment
- ✓Managed implementation and ongoing agent tuning
- ✓Omnichannel deployment across chat, voice, email, SMS and ChatGPT
- ✓Insights, monitors and experiments included
Sierra publishes no rate card and has no pricing page, self-serve tier or public calculator — every contract is custom-quoted through a direct sales cycle, which reviewers name as a recurring frustration. The model is outcome-based: you pay when an agent achieves a defined result such as a resolution, a saved cancellation or an upsell, and generally not when a conversation goes unresolved. Independent analysis puts the rate near $1.50 per resolved interaction with enterprise contracts typically starting around $150,000 a year. Because billing is entirely usage-and-outcome driven rather than seat-locked, spend swings with call volume, and the definition of 'resolved' is negotiated rather than standardised.
Security & Compliance
Sources
This page was written from 6 sources, 4 on domains other than sierra.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
xpander.ai
Vendor-neutral enterprise AI agent platform with a universal agent harness and Omni, an AI forward-deployed engineer
WSO2 Agent Manager
Open-source control plane for governing AI agents across any framework, model or cloud
Moveworks
Agentic AI assistant and enterprise search for employee IT, HR and finance requests
Convey
Digital teammates that own back-office operations, trained by screen-share not code