What it is
Maven AGI is an enterprise CX agent platform — AI agents that resolve customer requests across chat, email, voice, and web by taking multi-step actions in the systems the request is actually about, not by answering from a help-center article. It runs as an overlay: it sits on the help desk you already own and requires no data migration, so ticketing, routing, and reporting stay in Zendesk or Salesforce while the agent layer changes.
The company was founded in early 2023 by Jonathan Corbin (previously Global VP of Customer Success and Strategy at HubSpot), Sami Shalabi (ex-Google, led engineering for Google News, 50-plus patents), and Eugene Mann (ex-Stripe, led the Applied ML team). It has raised $78M in total, including a $50M Series B announced 18 June 2025 led by Dell Technologies Capital with Cisco Investments and SE Ventures alongside returning investors Lux Capital, M13, and E14. At that raise Maven reported going from 0 to 40 enterprise customers over the prior year, roughly $7M in 2024 revenue, and a 100% renewal rate. It is independent, names 90-plus companies as customers — Tripadvisor, Clio, Rho, ClickUp, Hard Rock Digital, SS&C, Thumbtack — and was listed in the CX Agents category of The Agentic List 2026.
The named product pieces are the AI Agent Designer (tune behavior and watch performance without engineering), the AI Voice Agent (speech-to-speech, not a chat bot bolted to a phone line), the Graph of Record (the customer data layer the agent reasons over), and Data and Insights.
Why it shows up in CX and support stacks
- The deployment model is what you are buying. Maven staffs forward-deployed engineers who embed with your team to wire the integrations, encode your policies and tone, tune edge cases as they surface, and validate production readiness. Published onboarding lengths are specific: K1x went from zero to 80% resolution in one week, Mastermind reached 93% in six weeks, and Clio ran a six-week onboarding to over 80% autonomous chat resolution and 60% more tickets solved than its previous chatbot. That is the account-led pole of this market, and it is the reason Maven wins deals against self-serve builders.
- The compliance surface is the widest in its peer set. Nine certifications: SOC 2 Type 2, ISO 27001, ISO 27017, ISO 27018, ISO 27701, ISO 42001 for AI management systems, PCI-DSS 4.0 Level 1, plus HIPAA, GDPR, and CCPA alignment. ISO 42001 and PCI Level 1 together are what carry an agent through a regulated procurement review that stalls lighter vendors.
- Retrieval is version-aware. Maven’s knowledge layer targets the specific failure where an agent answers confidently from a superseded document revision — the error class that costs you a policy dispute rather than a CSAT point.
- It connects to what you already run. Zendesk, Salesforce, Freshdesk, Front, Genesys, HubSpot, Intercom, and ServiceNow on the help-desk side; Confluence, Notion, Google Drive, and GitHub for knowledge; Snowflake, BigQuery, and S3 for data.
Pricing reality
Nothing is published. The pricing page is a demo-booking form with no tiers, no units, and no numbers. Vendr’s marketplace data puts the median Maven AGI contract at $100,000 per year, with observed purchases spanning $42,000 to $200,988 — a median Vendr scores 17.48% above Ada at $72,000. Deals are negotiated on ticket volume, channel count, and integration complexity rather than seats. Treat $42K as the practical floor and assume implementation and volume minimums land on top of the license line.
Best for
Enterprise and upper-mid-market CX leaders in regulated or payment-handling industries who run an existing help desk they will not migrate, and who want a vendor team embedded through deployment rather than a builder UI and a support portal. The scoped case where it beats everything else: a support org that must clear PCI and ISO 42001 review, has a policy corpus with real versioning, and cannot spare the internal engineers to integrate the agent itself.
Skip it if your annual support volume cannot pay back a $42K floor, if your team wants to configure the agent itself and would rather not manage a vendor engagement, or if you need to point Claude or ChatGPT at your CX analytics — see the last watch-out.
Versus the alternatives
The incumbents hold the installed base: Zendesk and Intercom Fin ship agents attached to the help desk you already pay for, at $1.50 and $0.99 per resolution respectively. Take one of those when their automation clears your bar, because adding a vendor is real work. The fastest-growing entrant is Decagon — a $250M Series D in January 2026 at a $4.5B valuation — and it wins when you want the steepest deflection curve and will accept outcome-based billing. Ada is the pick when you want the invoice decoupled from agent performance, since it bills per conversation rather than per resolution. Sierra competes for transaction-executing agents, though Maven’s PCI-DSS 4.0 Level 1 status closes that specific gap. Forethought is the cheaper multi-agent option when triage and agent assist matter more than autonomous resolution. Maven’s argument against all of them is the same: it brings engineers who stay until the deployment works, and it costs accordingly.
Watch-outs
- It is the most expensive median in the shortlist and publishes nothing. A $100,000 median against Ada’s $72,000, with a top observed contract just over $200K, and no public unit price to anchor against. Guard: put the Vendr band on the table in the first pricing call, get the billing unit in writing — per conversation, per resolution, or per action, they price differently as automation rises — and model three years at your projected volume before agreeing to a multi-year term.
- The forward-deployed engineering has no published commercial terms. Maven does not state whether embedded engineers are included in the contract, billed separately, or capped by hours. “They stay until it works” is a promise, not an SLA. Guard: write the FDE allocation into the order form as named weeks or hours with a defined handover point, and name the resolution rate that counts as done.
- The headline rates are vendor ceilings from favorable deployments. “Up to 93% resolved autonomously” and “80% in one week” are Maven’s own numbers from its own accounts, not a baseline you inherit. Guard: pilot one high-volume journey, measure containment and reopen rate against your current tooling for a full quarter, and set the contract’s volume commitment off your measured rate rather than the case study.
- Maven is an MCP client, not an MCP server. You can paste a third-party MCP server URL into Maven’s UI to give the agent new actions, which is useful. What you cannot do is point Claude or any other MCP client at Maven to query resolution drivers or CSAT trends conversationally — Ada ships that and Maven does not. Guard: if agent-side reporting is part of your plan, confirm the REST API exposes the metrics you need before you assume the workflow exists.