Traza

A named integration catalog covers eighteen systems (Jaggaer, SAP, SAP Ariba, Oracle, Coupa, Ivalua, Kinaxis, Blue Yonder, Oracle NetSuite, SharePoint, Confluence, Procore, DocuSign, Gmail, Outlook, Teams, Slack, Microsoft Dynamics). AI workers read, write and operate on top of the ERP, communication systems and systems of record, and a worked process ends in a worker step that updates SAP after a human approval.

SourceTraza, traza.ai homepageread 2026-09-08

The customer maps a multi-step process once and workers follow it every run. A worked four-step flow runs from the trigger to a worker verifying certifications, a human approving a substitution and a worker updating SAP. Skills are reusable units of work composed across the workforce, steps can be routed to different teams, and policy changes propagate to every active worker without redeployment. That is multi-step, multi-participant execution across a workforce of named workers on Traza's own runtime.

SourceTraza, traza.ai homepageread 2026-09-08

Skills take named sources (a Supplier DB and an ISO Cert DB in the worked example), workers read SharePoint and Confluence through the integration catalog, and the Spend Intelligence worker classifies every transaction. That is retrieval over the customer's connected systems, assembled per run. There is no maintained index, graph or embeddings layer that persists and stays queryable, and the supplier database is not presented as a maintained structure Traza owns.

SourceTraza, traza.ai homepageread 2026-09-08

The customer maps the process, sets the guardrails and pins a human wherever judgment is required. A worked process shows an Approve substitution step assigned to a human between two worker steps, and a policy example auto-approves substitutions at or below a stated threshold and applies to all active workers. Every action is tagged worker or human, and the team can step in at any point, with the worker resuming where it left off. That is a threshold governed approval gate in Traza's own product.

SourceTraza, traza.ai homepageread 2026-09-08

Traza shows SOC 2 Type II and ISO 27001 badges with GDPR under an Enterprise-ready heading reading Built for procurement at scale, audited to prove it, an unhedged claim. SSO and directory sync use SAML 2.0 and SCIM provisioning to sync users and roles from the customer's identity provider, with role-based access and granular permissions per team, per worker and per integration. The DPA adds MFA, passkeys, encryption at rest and access controls. The auditor, report date and type sit in the trust center at trust.delve.co/traza, so the attestation as shown rests on the homepage badges.

SourceTraza, traza.ai homepage and /dpa sections 4 and 9, trust portal refusedread 2026-09-08

Every worker action lands in an immutable audit log, timestamped, signed and export-ready for compliance reviews and external audits, and every run is visible with every action tagged worker or human. That signed, per-action, exportable record is Traza's own audit trail, not one inherited from the customer's ERP, and run-level visibility of each step lets a team reconstruct a run. Retention is stated only generically in the DPA, and no screenshot or documentation of the log surface itself is published.

SourceTraza, traza.ai homepage and /dpa section 7read 2026-09-08

Workers retain supplier decisions, benchmarks and policy state between tasks, with no repeated lookups and no lost context from previous runs. Traza shows the store as named keys (last supplier swap, quarterly price benchmark, escalation policy, preferred panel, last PO updated, negotiation outcome, compliance flags) under the heading context that carries across every run.

The scope is the worker and the lifetime is across runs, and the retained items are the worker's derived state, not the ERP record, so deleting them would not delete the purchase order. This is explicit state retention, not a model improving. There is no memory management surface, retention control or per-customer deletion, and the keys appear in a marketing panel, not a reference.

SourceTraza, traza.ai homepageread 2026-09-08

Customer Personal Data is primarily processed in the United States, with potential processing where subprocessors operate, and transfers are covered by SCCs. Cloud infrastructure providers handle storage and hosting, but none is named. There is no region list, no customer environment option (VPC, on-premises, customer tenant) and no way to select one. Zero access to ERP data outside worker execution is a posture, not a deployment choice. The processing location is disclosed, with nothing to choose.

SourceTraza, traza.ai/dpa Appendix A.4 and section 6read 2026-09-08

A Workforce panel names six prebuilt workers (Intake Agent, Sourcing and RFx, Supplier Onboarding, Contracts and Renewals, Invoice Matching, Spend Intelligence), a carousel walks through a worker for every job, and the CPO worker is described individually. Intake, sourcing, onboarding, contracts, invoice matching and spend classification are distinct jobs and none depends on another, so the panel is a catalog the customer browses. Policies are attachable rules (ISO 9001 required, preferred panel only), which support the skills mechanism and are not a template pack.

SourceTraza, traza.ai homepageread 2026-09-08

Work reaches the workers through a named trigger list of inbound email, SAP change order, webhook, scheduled and manual dispatch, and workers can start from any system. In a worked process a received change order is the trigger step, and a Procurement Intake worker routes and classifies incoming requests. That covers events, schedules, webhooks and an inbound channel, and email, Slack and Outlook appear in the integration catalog.

SourceTraza, traza.ai homepageread 2026-09-08

Traza's Data Processing Agreement (effective 1 January 2025, section 5.1) names OpenAI and Anthropic as subprocessors for AI processing. That is disclosed use of two providers on Traza's side. There is no customer or admin model selection surface or routing control, and the DPA does not state how work is routed between the two.

SourceTraza, traza.ai/dpa section 5.1read 2026-09-08

A webhook is one of five ways a worker starts, alongside inbound email, an SAP change order, a schedule and manual dispatch, and an inbound webhook lets an outside caller drive the platform. There is no API reference, developer portal, SDK, MCP server or agent card. The webhook is a single word in a trigger list, with no addressing, payload or authentication given, and the login at platform.traza.ai is a product login, not a developer surface. Skills written in plain language are the customer building inside the platform.

SourceTraza, traza.ai homepageread 2026-09-08

There is no evaluation harness, scored test cases, release gate or controlled post-deployment experiment measuring the workers. Running side by side with existing operations before cutover is part of Traza's onboarding (we connect to your ERP, configure the workers and run side by side), an engagement practice, not a platform feature. The visible-run and step-in surface is oversight and audit, and the 68 percent and 94 percent figures are outcome marketing.

SourceTraza, traza.ai homepageread 2026-09-08

Workers read, write and operate on top of the ERP, communication systems and systems of record through a named integration catalog, so execution runs through integrations. There is no browser, desktop or remote computer control.

SourceTraza, traza.ai homepageread 2026-09-08