Demandbase

Across CRM, marketing automation, sales engagement and social, Demandbase offers pre-built plug-and-play integrations for Salesforce, HubSpot, Marketo, Outreach and LinkedIn. A Demandbase Marketplace lists third-party connectors after they pass certification, so the catalog extends beyond what Demandbase builds.

An Account Connector API exports account data to arbitrary third-party platforms, bidirectional HubSpot CRM sync runs over REST with a configurable read cron and a daily writeback window, a Salesforce Data Integrity integration supports unlimited enrichment workflows, and a Demandbase connector is published inside Salesforce Data 360.

The Sendoso integration chains that partner's MCP to the Demandbase MCP so one prompt spans account signal, contact selection, Salesforce context and a gift catalog, which is authenticated action across four systems in a single flow.

Sourcedeveloper.demandbase.com, support.demandbase.com integrations documentation and demandbase.com partner ecosystem postread 2026-09-02

Agentbase is a system of connected agents rather than one assistant, with unattended execution documented. Demandbase describes it as a system of connected AI agents that surface high-value accounts and buying groups, orchestrate journeys and pinpoint actions, built on open architecture and, for the Filter and Action agents, alongside the AWS GenAI Innovation Center on Amazon Bedrock.

The clearest case of agents acting rather than advising is the Campaign Outcomes Agent, which optimizes bidding strategies inside Demandbase's own B2B demand-side platform toward a customer-chosen outcome, with reported results of 40 percent higher click-through and 25 percent greater lift in page visits against standard campaigns.

The April 2026 release adds a conversational orchestration layer that Demandbase describes as coordinating cross-channel activations and continuously adapting, and Plays are orchestrated as a first-class object across the agents.

Sourcedemandbase.com press releases March 2025, May 2025 and April 2026 and support.demandbase.comread 2026-09-02

The agents are grounded on a maintained structure that joins Demandbase's own data estate to the customer's records. Context Intelligence, introduced with the April 2026 release, is a proprietary layer that applies each company's own GTM context to analyze account signals and patterns against that company's pipeline goals, so the customer's material shapes what the agents conclude.

Underneath sits company, contact, technographic and intent data, account identification, buying group detection and intent scoring built on billions of signals, joined in the Demandbase Database to the customer's synced CRM and marketing automation records and read as the standing basis for what the agents say. Demandbase's AI principles page separates agentic AI from the predictive and algorithmic AI that powers identification and scoring.

Sourcedemandbase.com/ai-principles, April 2026 release and support.demandbase.comread 2026-09-02

Demandbase gives admins real control over whether an agent runs and thinner control over what it does once running. Admins can enable or disable each agent at tenant level, switching off a specific AI agent while leaving the others on, and a global toggle disables all current and future AI capabilities, betas included, for organizations with strict internal generative AI policies. Permission sets grant Full, Limited, View Only or No Access per platform area, and the documented Sendoso MCP flow includes an approve-before-send step in which the agent hands the user a proposal to approve.

Demandbase also documents what disabling does: it removes AI chat interfaces and insight panels and leaves previously generated personas and buying groups intact, which makes the control usable rather than nominal. The approval step is documented in one partner integration flow rather than as a platform-wide gate, and no approval routing or review queue is documented for the agents that act directly, notably the Campaign Outcomes Agent as it optimizes live bidding.

Sourcesupport.demandbase.com Agentbase settings and permission sets documentationread 2026-09-02

Both attestation and controls are documented on the trust center. SOC 2 Type II against the Security criteria and ISO/IEC 27001:2022 were both recertified December 2025 and audited by a named firm, Schellman and Company, with independent third party penetration testing at least annually and continuous internal application and infrastructure testing.

Granular permission sets grant Full, Limited, View Only or No Access per platform area, and audit logging is listed as a product security feature alongside access monitoring, data access controls, password security, disk encryption and centralized log retention with anomaly alerting.

On the legal side there is a published subprocessor list, a customer DPA incorporating standard contractual clauses with security measures in Annex 2 and a security audit clause, and a data transfer impact assessment.

The trust center publishes a CAIQ and a SIG Lite under self-assessments and lists a SOC 2 Type 2 report, pentest reports for 2024 and 2025, a vulnerability assessment report, the ISO 27001 certificate, an Agentbase AI FAQ and a data diligence FAQ behind a request-access step.

Sourcetrust.demandbase.com and demandbase.com/security-policyread 2026-10-01

Audit logging exists as a product feature, but there is no agent trail. The trust center lists Audit Logging under product security alongside data security and integrations, and the help center documents the customer-facing surface behind it: an admin change audit trail where permission set edits are reviewable under Settings with a 90-day retention window, plus data sync status monitorable through system notifications.

Access monitoring and centralized log retention with anomaly alerting also appear, but the security policy describes those as Demandbase's own monitoring rather than a surface exposed to the buyer, and the trust center lists them under corporate and network security rather than product security. Nothing separates agent action from user action, with no AI-specific audit log, no per-run record of what an agent decided, and no way to reconstruct why the Campaign Outcomes Agent changed a bid or why the Account Engagement Agent summarized as it did. The AI agents FAQ claims full transparency into how decisions are made, with no artifact behind it.

Sourcetrust.demandbase.com and support.demandbase.com permission sets documentationread 2026-09-02

Durable account records are kept, but agent memory is not. What persists is account and engagement history in the Demandbase Database, synced CRM and marketing automation records, and agent outputs saved as records: the help center states that when Agentbase is disabled "Previously generated data (for example, personas and buying groups) remains unchanged." Those are business records the agents write and read. Nothing is exposed as an agent memory, no per-agent state across sessions is documented, and there is no retention or forgetting control at the agent level.

Sourcesupport.demandbase.com and demandbase.com/ai-principlesread 2026-10-01

No customer control over where anything runs. The trust center at trust.demandbase.com covers compliance, documents, risk profile, product security, reports, self-assessments, data security, app security, legal, privacy, access control, infrastructure, endpoint security, network security, corporate security, policies and security grades. Infrastructure lists Amazon Web Services, anti-DDoS and business continuity; privacy lists cookies, cross-border data transfers and a data protection officer.

Nowhere does it offer region selection, an EU hosting option, a single-tenant or private VPC deployment or a residency commitment. The one entry that touches location, cross-border data transfers, is legal cover rather than a deployment choice, and it sits alongside a data transfer impact assessment and standard contractual clauses, the set a vendor produces when personal data moves to the United States. Demandbase is US incorporated, so CLOUD Act reach applies regardless.

Sourcetrust.demandbase.com and demandbase.com/security-policyread 2026-09-02

Demandbase offers a browsable set of named agents a buyer turns on individually. Administrators manage Agentbase at tenant level, locating an agent, clicking Edit and toggling it on or off, with a separate global toggle for all of them, and a companion article lists Agentbase features against their AI credit usage, so each agent is enumerated in the product, separately adoptable and costed per feature.

The named set covers distinct jobs: Campaign Outcomes for bid optimization inside Demandbase's own demand-side platform, Account Engagement for engagement summarization, Filter and Action built with the AWS GenAI Innovation Center, and a Site Customization Agent for conversational landing page creation added April 2026, with further agents in beta with 35 members of an AI design partner program. Plays sit alongside as a first-class orchestrated object.

Sourcesupport.demandbase.com Agentbase settings article and demandbase.com press releasesread 2026-09-02

Both intake and activation are wide. Work starts from intent scoring built on billions of signals, account identification, buying group detection, web visit and IP-based company identification through the IP API v3, and CRM and marketing automation events pulled through bidirectional sync, so it begins because the market moved rather than because someone scheduled it.

Output reaches advertising through the owned B2B demand-side platform, website through the Site Customization Agent, email and nurture through Marketo and HubSpot, sales engagement through Outreach, and social through LinkedIn, a list the integration documentation confirms. Demandbase frames the platform as turning signals into cross-channel activation, and the April 2026 release adds an orchestration layer coordinating those activations.

Sourcedeveloper.demandbase.com, support.demandbase.com and demandbase.com April 2026 releaseread 2026-09-02

No customer-facing model choice. Agentbase is built on AWS with the AWS GenAI Innovation Center on Amazon Bedrock, and Demandbase's AI principles page says only that agents may leverage multiple AI techniques including large language models and retrieval systems, naming no specific model and offering no selection, routing, bring-your-own-model or bring-your-own-key.

The ChatGPT, Claude, Copilot and Gemini names attached to Demandbase run the other way: they are MCP clients consuming Demandbase data, so the customer picks the assistant that reads Demandbase rather than the model that powers its agents. The only model-related control a buyer has is the tenant switch that turns generative AI off, per agent or entirely, so the provider behind the agents stays undisclosed.

Sourcedemandbase.com/ai-principles and developer.demandbase.comread 2026-09-02

Demandbase runs a dedicated developer site with three extension surfaces and a certification path. The Demandbase API Suite covers account intelligence, engagement insights, user management, data enrichment and analytics. The Demandbase MCP is a first-party Model Context Protocol server described as a secure gateway giving natural language access to both first-party tenant data and third-party B2B intelligence, with ChatGPT, Claude and Visual Studio Code named as supported clients.

The Partner Tag is an authenticated JavaScript module letting partners activate Demandbase advertising and data capabilities, and the IP API v3 resolves a visitor IP or cookie to company, firmographic and registry data. Third-party vendors can build connectors, pass certification and then be listed in the Demandbase Marketplace, and the Sendoso integration chains that partner's MCP to Demandbase's so a single prompt spans account signal, contact selection, Salesforce context and gift catalog.

Sourcedeveloper.demandbase.com and demandbase.com partner ecosystem documentationread 2026-09-02

Demandbase measures what the agents produce but offers nothing that tests the agents.

Pipeline Influence capabilities shipped in April 2026 specifically for proving contribution to pipeline, campaign performance is measured inside the demand-side platform, and lift figures are reported on agent-optimized bidding, so the business result of agent work is readable and comparable.

No surface evaluates the agent itself, with no sandbox, evaluation harness, regression suite, prompt or output scoring, or published methodology for assessing agent quality. The one adjacent control, disabling an individual agent, is a switch rather than a test.

Sourcedemandbase.com April 2026 release and support.demandbase.comread 2026-09-02

No browser control, computer use, screen operation or robotic process automation of any kind is documented. Demandbase agents act through APIs, native integrations, the MCP interface and Demandbase's own advertising platform, which suits account-based go-to-market work, since everything they touch publishes a programmatic route. Nothing across the product pages, the developer portal, the help center or Demandbase's announcements describes an agent operating software through an interface.

Sourcedeveloper.demandbase.com, support.demandbase.com and demandbase.com product pagesread 2026-09-02