# AI for Propel PLM: Using Salesforce Data Access Properly, With Models You Control > Source: https://netray.co/erp-ai/propel-plm-ai-integration > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: AI for ERP ## Short answer Propel PLM runs on the Salesforce platform, so AI for Propel reads item, revision, BOM and change records through Salesforce APIs under the asking user's sharing and field-level security. A privately hosted model then answers questions and drafts change text with citations. It is useful where Salesforce's own AI features do not reach product data or where you need to control hosting. ## Kind platform ## Erp Family mfg-systems ## Erp Products - Propel PLM ## Industries - Medical Devices - Consumer Products - Electronics - Life Sciences ## Region Global ## Persona Product Operations Manager ## Eyebrow Propel PLM + private AI ## Intro Propel PLM is built as a managed application on Salesforce, so items, revisions, BOMs and change orders are Salesforce records, governed by the same profiles, permission sets, sharing rules and field-level security as the rest of your org. For companies already living in Salesforce for sales and service, that puts product data next to customer data, which is its appeal. That architecture changes what AI integration looks like. Data access is via the Salesforce REST API, SOQL and Bulk API with OAuth through a connected app, and automation can run in Apex or Flow. It also means Salesforce's own AI offerings are a natural first look, and an honest assessment starts by asking whether they cover your needs. They often cover CRM-adjacent tasks well. Where they fall short for product teams is in cross-system questions that combine PLM, ERP and quality data, in regulated settings that restrict where inference runs, and in customer-specific change-control rules that require predictable, citation-based behavior. This page covers a practical private design for Propel: how to read the data safely, how to keep sharing rules in force, where AI helps change, quality and launch teams, and where Salesforce-native tools or standard reports are enough. ## Pain Points ### Product data and ERP data live apart **Detail:** Item and BOM data sits in Salesforce while costs, stock and supplier orders sit in the ERP. Cross-questions need exports and manual joins. ### Change orders are written inconsistently **Detail:** Change justification and impact text vary by author, making review slower and audit trails harder to follow for regulated products. ### Report building is limited by skills **Detail:** Salesforce reports and dashboards are powerful, but questions across item revisions, BOM lines and changes need someone who knows the object relationships. ### Governor limits and data volume **Detail:** Large BOM explosions and bulk reads run into API and processing limits, so naive integrations are slow or fail. ### Security model must stay intact **Detail:** Sharing rules, permission sets and field-level security protect sensitive costs and supplier terms. Extracting data with an over-privileged integration user undermines them. ## Use Cases ### Ask-your-PLM for items and BOMs **Description:** Answer where-used, BOM revision and lifecycle questions in plain English, returning item records with links back into Propel. **Erp Objects:** Propel item, item revision and BOM records, lifecycle phase fields **Outcome:** Routine structure lookups no longer need a report builder or an administrator. ### Change order drafting **Description:** Draft reason for change, affected item lists and impact statements from selected records and past similar changes, for author review. **Erp Objects:** Propel change orders, affected items, attachments **Outcome:** More consistent change records and quicker reviews. ### Quality event linkage **Description:** Connect complaints, nonconformances or CAPA records to affected items and changes so recurring failure modes surface. **Erp Objects:** Propel quality records, item revisions, related changes, Salesforce case data where used **Outcome:** Trends across product, change and service data become visible with less manual correlation. ### BOM to ERP comparison **Description:** Compare released Propel BOMs against ERP manufacturing BOMs and list quantity, unit and revision differences. **Erp Objects:** Propel BOM records, ERP BOM and item master, integration logs **Outcome:** Mismatches found at release, not during production planning. ### New product launch readiness **Description:** Check item completeness, compliance attributes, supplier approvals and open changes against your launch checklist. **Erp Objects:** Item fields, supplier records, compliance and document attachments, task records **Outcome:** Fewer gaps at launch gates; the checklist remains yours. ### Document retrieval **Description:** Answer from controlled documents and specification attachments, citing the file and version. **Erp Objects:** Content versions, attached files, document records **Outcome:** Quicker answers during audits and customer reviews. ### Cross-system executive questions **Description:** Combine PLM revision status, ERP inventory and supplier lead times to answer questions such as which new revisions are blocked by missing components. **Erp Objects:** Propel items and changes, ERP on-hand, open POs, vendor lead times **Outcome:** Leaders get cross-system answers without a manual data-gathering round. ## Architecture **Overview:** A connector authenticates to Salesforce through a connected app, queries Propel objects with SOQL and the REST API as the asking user, and hands retrieved records to a privately hosted model. Drafts are created as records for human review. **Layers:** #### Authentication **Detail:** OAuth through a connected app with user-delegated flows for questions and a tightly scoped integration user for indexing. #### Data access **Detail:** SOQL via REST API for interactive questions; Bulk API for scheduled indexing; field-level security and sharing evaluated by Salesforce. #### Retrieval layer **Detail:** Object-aware structured lookup for items, revisions and BOMs, plus a vector index over attachments, preserving record ids for citation. #### Model layer **Detail:** Open-weight models running on your GPUs or private cloud, with no calls to public inference APIs. #### Governance layer **Detail:** Per-user logs, limit monitoring, answer evaluation on known cases, and draft-only writes. ## Deployment Options ### Private cloud inference **Fit:** Most Propel customers, who already run Salesforce in the vendor cloud **Detail:** The retrieval and model services run in your own cloud tenancy and call Salesforce APIs over private or restricted network paths. ### On-premises GPU **Fit:** Companies with strict data handling for supplier terms or regulated product data **Detail:** Extracted data and models are kept on-site. Salesforce stays vendor hosted, so the design covers only data after extraction. ### Salesforce-native first **Fit:** Teams whose needs are CRM-adjacent and non-sensitive **Detail:** Start with Salesforce's own AI features and add a private layer only for the gaps. This can be the cheapest path and should be evaluated honestly. ## Compliance ### Change and design records are read only. Draft text is labeled and must be approved through Propel's own workflow, so the controlled record is unchanged. **Framework:** FDA 21 CFR Part 820 (QMSR) and ISO 13485 ### Electronic signatures remain in Propel. The AI layer never signs or approves anything. **Framework:** 21 CFR Part 11 ### Item and document references are cited exactly. Regulatory submissions still rely on your regulatory team's review. **Framework:** EU MDR and UDI traceability ### Because data is copied out of Salesforce, document where it lands, who can reach it, retention and encryption as part of your own control environment. **Framework:** SOC 2 and customer security reviews ## Integration Notes - Use a connected app with scoped OAuth permissions; avoid a system administrator profile for the integration user. - Check Propel's object and field names against your installed package version; managed package namespaces and field sets differ across releases. - Use Bulk API for scheduled indexing and REST with SOQL for interactive queries, and monitor daily API request limits. - Evaluate sharing as the asking user; do not cache query results across users without keeping their permissions attached. - Handle large BOMs by paging and by querying levels rather than whole explosions in one call. - Read ERP data through its own API for comparisons so the check does not depend on the synchronization under review. ## Product Fit ### Custom build **Why:** A Salesforce-aware connector, change-drafting agent and BOM comparison tailored to your Propel configuration. ### ERPray **Why:** Covers the ERP side for cross-system questions with read-only queries and visible SQL. ### DataRay **Why:** Useful when attachments, shared drives and ERP tables must sit behind one private chat. ## Engagement ### Discovery **Duration:** 1 to 2 weeks **Deliverables:** - Review of Propel configuration and object model - Comparison against Salesforce-native AI options - Security and data-handling decision - Question inventory ### Pilot build **Duration:** 3 to 5 weeks **Deliverables:** - Connected app and delegated access - Item, BOM and change retrieval - Cited answers - Evaluation set ### Workflow agents **Duration:** 3 to 4 weeks **Deliverables:** - Change drafting - ERP BOM comparison - Launch readiness checks - Quality linkage report ### Handover **Duration:** 2 weeks **Deliverables:** - Runbook and limit monitoring - Credential rotation procedure - Model update plan - User training ## Buyer Questions - Have you compared this with Salesforce's own AI features for our use cases? - Do queries respect sharing rules and field-level security for each user? - What does the integration user have access to, and how is it scoped? - Where does extracted data sit, and for how long? - How do you stay within Salesforce API limits at our volume? - How do you handle package upgrades that rename or add fields? - What is written back to Propel, and who approves it? - How do you test answers on our actual BOMs? ## Faq ### Should we just use Salesforce's own AI? **Answer:** Evaluate it first. For CRM-adjacent summaries and simple record questions it may be enough. A private layer makes sense when you need your own model hosting, cross-system answers with ERP and quality data, or tightly controlled, citation-based change drafting. ### Does the AI respect Salesforce sharing and field-level security? **Answer:** It should, by querying as the asking user through delegated OAuth so Salesforce evaluates access. An integration user with broad access answering for everyone is a design to avoid, particularly for cost and supplier fields. ### Can it modify Propel records? **Answer:** By default, no. It reads and drafts. Changes are created and approved by users in Propel, retaining the audit trail and signatures required for regulated products. ### Will Salesforce API limits be a problem? **Answer:** They can be, for large BOMs or frequent syncs. Incremental indexing, Bulk API for batches, caching of non-sensitive structure and paged queries keep usage within limits, and monitoring flags trouble early. ### Is on-prem possible given Salesforce is cloud? **Answer:** The AI stack can be on-prem or in your private cloud, pulling data over APIs. Propel and Salesforce remain vendor hosted, so the privacy benefit applies to inference and the copied dataset, not to the source system. ### What about our ERP? **Answer:** ERP data is read separately through its own API or read-only views. This lets the AI compare Propel BOMs against ERP BOMs and answer questions mixing revision status with stock and purchasing. ### How long does a pilot take? **Answer:** Often four to eight weeks from access to a working read-only pilot, depending on how customized the Propel configuration is and how quickly connected app and security approvals are granted. ## Search terms propel plm ai, propel plm salesforce ai, propel plm integration, propel plm llm, propel plm change order automation, salesforce plm ai, plm ai medical devices, propel plm erp integration, private ai salesforce plm, propel plm bom chatbot, propel plm api, propel plm quality ai --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json