# How to add an AI chatbot to Global Shop Solutions > Source: https://netray.co/integrations/global-shop-ai-chatbot > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer An AI chatbot for Global Shop Solutions answers questions about jobs, orders, inventory and procedures by retrieving the facts first and then phrasing them, so it cites records instead of guessing. Run it as an internal assistant for staff first, add a customer-facing version later, and host the model on-prem if your data requires it. ## System Global Shop Solutions ## Capability ai-chatbot ## Intro Most of the questions that interrupt people in a job shop are repetitive: where is this job, did that material arrive, what is the process for rework, who approves this. Answers exist in the ERP and in procedure documents, but finding them takes screens and tribal knowledge. A chatbot grounded in those sources removes that friction. The word grounded is doing the work. A bot that answers from the model's memory will invent plausible shop facts. A bot that retrieves a job record or a document paragraph and quotes it will not. Global Shop Solutions customers also have a large installed base of customers who want portal-style status. A tightly scoped external bot can deliver that, provided it only ever sees that customer's records. ## Use Cases ### Internal job status assistant **Description:** Supervisors ask about a job or a work center queue in plain language and get current operation, hours remaining and blockers. ### Procedure and policy answers **Description:** Staff ask how to handle a nonconformance or a return, and the bot answers from the approved quality procedure, citing the document and revision. ### Customer order status bot **Description:** Customers identify themselves, then ask about their own POs, shipments and invoices without calling the office. ### New-hire onboarding helper **Description:** New planners and buyers ask what a screen or term means and are pointed to internal training notes and the right workflow. ### Sales desk support **Description:** Quotes, past pricing and lead-time answers for existing parts, with the source job or quote number attached. ## System Surface - Structured data: database views or vendor web services for jobs, sales orders, POs, inventory and shipments, read with a restricted account. - Documents: quality manuals, work instructions and customer requirements held in shared folders or a document control system, indexed for retrieval. - Identity: for staff, connect your directory (for example Entra ID or Active Directory) and map groups to data scopes. For customers, map a portal login to a customer number. - Existing portal or web front end: if you already expose a customer portal, the bot can sit next to it. Confirm what your version includes. - Notifications: email or messaging hooks for handing a conversation to a human. ## How It Works ### Choose the audience **Detail:** Start with internal staff. The data risk is lower, feedback is fast, and you learn which questions matter before opening to customers. ### Build the structured lookup **Detail:** Create a few lookups (job, PO, inventory, shipment) that return small, labelled JSON. The bot calls these rather than searching free text. ### Index the documents **Detail:** Split approved procedures into sections with document ID and revision metadata, embed them and store in a local vector index. Exclude drafts and superseded revisions. ### Compose the answer with citations **Detail:** Instruct the model to answer only from retrieved material, quote the record or section, and say it does not know otherwise. ### Apply identity and scope **Detail:** Filter every lookup by the signed-in user's role or customer number before the model sees anything. ### Escalate and learn **Detail:** Provide a handoff to a person, record unanswered questions weekly, and fix content gaps rather than prompts. ## Example Flows ### Supervisor asks about a stalled job **Trigger:** Why hasn't job 20931 moved since Tuesday? **Steps:** - The bot looks up the job and its last labor posting. - It checks for open material shortages and outside-process lines. - It answers that operation 30 is waiting on a heat-treat vendor and cites the PO line and promise date. **Result:** The supervisor knows who to call within a minute. ### Customer asks about a shipment **Trigger:** A customer signs into the portal chat and asks about their PO. **Steps:** - The bot resolves the customer number from the login. - It retrieves only that customer's open orders and shipment records. - It replies with ship date, carrier and tracking where recorded. **Result:** Fewer status calls to the office and a log of what customers ask. ## On Prem A chatbot is where on-prem pays off, because the conversation itself contains the sensitive information. Host an open-weight chat model and an embedding model on a GPU server in your facility, keep the vector index and conversation logs on local storage, and serve the interface through your intranet or a reverse proxy for customers. Nothing about orders or procedures goes to a third-party AI service. For small shops, a single workstation-class GPU can serve a handful of simultaneous staff users, but size by measuring concurrent load. Keep a plan for model updates and for re-indexing when procedures change. ## Security - Enforce per-user and per-customer filters in code, not in the prompt. - Keep customers strictly separated: a customer-facing bot must be unable to retrieve another customer's records, even if asked. - Index only approved, current documents and record revision metadata. - Log conversations with retention rules and restrict who can read them. - Add rate limits and abuse protection on any public-facing endpoint. ## Options ### Vendor portal or built-in search first **Fit:** Good if your version already gives customers and staff the status views they need. **Effort:** Configuration only. ### Build with an open-source RAG stack **Fit:** Good for internal assistants where you have a developer to maintain it. **Effort:** One to three months including evaluation. ### Partner deployment such as Netray **Fit:** Good for on-prem hosting, role scoping and a customer-facing version with proper separation. **Effort:** A pilot of a few weeks for an internal bot. ## Pitfalls - Letting the model answer without retrieval, which yields confident invented job facts. - Indexing every document including obsolete revisions, so the bot cites outdated procedures. - Filtering by prompt instruction instead of by query, which leaks data under clever questions. - Launching to customers before staff have tested it for a few weeks. - No owner for content gaps, so the same unanswered questions repeat. ## Faq ### Can the chatbot run entirely inside our network? **Answer:** Yes. An open-weight chat model and an embedding model can run on a local GPU server with a local vector index. The ERP read access also stays on the LAN. Internet access is only needed if you choose to expose a customer-facing interface. ### How do we stop it from making things up? **Answer:** Force retrieval before answering, require citations to a record or document section, and instruct it to say it does not know when nothing is retrieved. Then test with questions where the right answer is none. ### Is a customer-facing bot a good first project? **Answer:** Usually not. Start internal, fix the content and scoping issues, then expose a narrow customer scope with strong separation and a human escalation path. ### What about our existing portal? **Answer:** Check what your Global Shop Solutions version provides. If the portal already answers the common questions, a bot may add less than expected. It earns its place with unstructured questions and documents. ### Can it take actions, like creating an RMA? **Answer:** It can draft them. Keep a human confirmation step and post through a limited service account. Do not allow customer-facing bots to write directly to the ERP. ## Search terms global shop solutions chatbot, ai chatbot for global shop, erp chatbot job shop, global shop customer portal chatbot, on-prem erp chatbot, global shop solutions ai assistant, chat with global shop data, rag chatbot manufacturing erp --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json