Planned

The agent acts on the vendor's own rails rather than the customer's systems: the AI reaches out to suppliers across a marketplace the vendor puts at 300,000 hotels and venues and 60,000 vendors, collects their quotes, processes invoices, can sign a contract on the customer's behalf when the customer chooses, and pays every supplier as vendor of record so the customer receives one invoice.

No route into the customer's own estate is documented. No system of record, expense, ERP, HRIS or calendar integration is described and there is no connector catalog or tool framework; the Cronofy calendar integration entry on the subprocessor list shows one such integration exists in the product without documenting it.

SourcePlanned, planned.com home, ai-approach/ai-release and security.planned.com subprocessor listread 2026-09-07

The homepage states the AI sources, negotiates, reads contracts, processes invoices and enforces policies so a planner moves from request to booking without waiting on anyone, and that from collecting quotes to updating rooming lists the AI completes entire workflows unprompted; the AI release page lays out the stages, destination recommendation, RFP pre-processing and supplier outreach, live proposal comparison as responses arrive, and contract summary through to signature, with the vendor then paying as vendor of record. Temporal on the subprocessor list for automated workflows points to a workflow runtime behind this.

SourcePlanned, planned.com home and ai-approach/ai-releaseread 2026-09-07

Supplier Intelligence on the AI release page states the AI taps into years of supplier conversations to pre-process each RFP, removing the questions it already holds answers to, parking, Wi-Fi, room specs, so the customer never asks the same question twice; that is a retained store of prior answers consulted per request rather than context assembled fresh, and the subprocessor list's Weaviate entry for semantic search and database vectorisation shows the store is indexed.

Alongside it the customer's own policies, approval flows and supplier standards are set once and applied to every event, and Where to Meet grounds destination recommendations in travel time, seasonal weather and local event data. The AI being trained on years of event planning data is model training rather than grounding.

SourcePlanned, planned.com/ai-approach/ai-release and security.planned.com subprocessor listread 2026-09-07

Contract Summaries on the AI release page has the planner review the AI's summary of cancellation, attrition and payment terms and then request edits or approve the contract for signature, with signing on the customer's behalf only if the customer chooses it; the homepage's Establish step has the customer set approval flows and policies that the AI then enforces on every event across every user, team and region.

The approve before sign step is one the customer configures and operates. The dedicated sourcing specialist who reviews AI outputs and owns negotiation, and the experts who step in for VIP events or whenever a human is wanted, are vendor staff inside the engagement and separate from that step.

SourcePlanned, planned.com/ai-approach/ai-release and planned.com homeread 2026-09-07

Attestation and access control are both on the vendor's own security center. The compliance article states the last SOC 2 audit was in May 2024 with no unusual findings and the report is available on request to the customer's implementation manager, a completed audit the vendor asserts in its own name.

For customer facing control, the same article answers that the product features role based access control to manage and restrict user access by role and responsibility, and the subprocessor list names WorkOS as the platform integrating authentication providers, which points to an SSO path without documenting it. Encryption at rest and in transit and password hashing are stated alongside. The same page documents audit logging, and the policy and approval flow enforcement on the homepage is program governance rather than an access model.

SourcePlanned, security.planned.com compliance and data security article and subprocessor listread 2026-09-07

Activity logging is documented and agent level tracing is not.

The security center states that Planned enables logging to track activity and configuration changes and retains account activity logs for at least a year, and the product surfaces are visible in the vendor's own words: supplier responses populate a live comparison table shared with stakeholders, every supplier, contract and payment flows through one vendor of record with one invoice and full spend visibility, and the program tracks every dollar and minute saved. That records what happened and what it cost.

Nothing shows a customer the run of an agent, which suppliers it wrote to, what it answered on the customer's behalf or why it made a recommendation, and the AI observability tools on the subprocessor list, Braintrust and LangSmith, are the vendor's own build dependencies whose output no customer is documented as receiving.

SourcePlanned, security.planned.com compliance article, planned.com home and ai-approach/ai-releaseread 2026-09-07

The homepage states that every event feeds program intelligence so the program compounds, with the AI learning from every guest reply, negotiation and booking decision; that is absorbed learning and program analytics rather than a memory layer. The subprocessor list notes that AI subprocessors do not train on client data, so the improvement is not model adaptation on the customer's data either. No memory store, per user or per program recall surface or deletion path is documented.

SourcePlanned, planned.com home and security.planned.com subprocessor listread 2026-09-07

The subprocessor list, edited 30 April 2026, names Google Cloud Platform as application hosting and cloud provider in the USA and Neon on AWS as the Postgres database platform in the USA, with every data subprocessor located in the USA except the Weaviate vector store in the Netherlands, and no customer selection of region, tenant or environment is documented anywhere. The Netherlands entry is a vendor infrastructure location, not a residency option.

SourcePlanned, security.planned.com subprocessor listread 2026-09-07

No catalog of agents a customer browses and selects among is documented.

The AI release page presents four named capabilities, Where to Meet, Supplier Intelligence, Live Proposal Comparison and Contract Summaries, as features of one sourcing workflow rather than units a buyer adopts separately, and the homepage's Establish step, where the customer sets policies, approval flows and supplier standards, is configuration of that one workflow.

Standardized event briefs, policy and approval templates and the curated supplier marketplace are a form, a rule set and inventory, and none of them does work when selected.

SourcePlanned, planned.com home and ai-approach/ai-releaseread 2026-09-07

Work reaches the agent by a planner's request in the application, after which the run continues on inbound supplier responses: the AI release page has responses from suppliers automatically populating the comparison table as they arrive, and the homepage describes an always on, proactive AI that completes entire workflows unprompted and enforces policy on every event automatically. No trigger class a customer configures, no schedule, no event from a connected system and no channel other than the application itself, such as email, chat or a ticketing queue, is documented, and the messaging, email and calendar entries on the subprocessor list are the vendor's infrastructure rather than documented intake channels.

SourcePlanned, planned.com home and ai-approach/ai-releaseread 2026-09-07

OpenAI, Google Gemini and Anthropic are named as generative AI subprocessors on the subprocessor list, edited 30 April 2026, with OpenRouter as an AI gateway, which is the vendor routing across three foundation model providers behind its own product; the note that AI features are opt in and that AI subprocessors do not train on client data is a customer switch on AI use, not a choice of model. The homepage's line about AI trained on years of event planning data sits on these third-party models. Nothing on the homepage or AI release page lets a customer pick or configure a model.

SourcePlanned, security.planned.com subprocessor listread 2026-09-07

No API, SDK or MCP server for calling Planned from outside is documented. The site has no developer section, and the security center, which is the one technical surface the vendor publishes, describes controls and subprocessors rather than an interface. The marketplace's Become a partner route is supplier onboarding, not a programmatic surface.

SourcePlanned, planned.com home and ai-approach/ai-release, security.planned.comread 2026-09-07

The AI release page has the program track every dollar and minute saved and publishes a policy compliance rate and time to first quote as program results, and the homepage describes program intelligence that compounds as costs come down; those are measures of the operation. No customer facing harness, test set, comparison of a change against a baseline or controlled optimization loop is documented.

The vendor's own evaluation and observability tooling, Latitude for AI evaluation and Braintrust and LangSmith for AI observability on the subprocessor list, shows the vendor evaluates its agents internally, but no customer can point an account at it or receive its output. The vendor's reported blind judgment benchmarking of recommendation quality is the vendor testing its own model, and specialist review of outputs is staff QA.

SourcePlanned, planned.com/ai-approach/ai-release and security.planned.com subprocessor listread 2026-09-07

The agent works through the vendor's own marketplace, messaging and payment rails: it sends pre-processed RFPs to suppliers, receives their responses into the comparison table and summarizes contracts the suppliers return. No browser, desktop or remote computer control is documented.

SourcePlanned, planned.com/ai-approach/ai-releaseread 2026-09-07