# How to use ChatGPT with Global Shop Solutions > Source: https://netray.co/integrations/global-shop-chatgpt > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer You can use ChatGPT with Global Shop Solutions in three ways: paste or upload exports for one-off analysis, build a custom GPT with actions that call a small read-only API in front of your data, or move to a private model when the data is controlled. Start with exports, because it teaches you which questions are worth automating. ## System Global Shop Solutions ## Capability chatgpt ## Intro Plenty of Global Shop Solutions users already use ChatGPT informally: pasting a quote description to tidy the wording, asking it to explain a report, or uploading a spreadsheet export to find patterns. That works, and it is also where data-handling mistakes start. The more valuable step is giving ChatGPT a governed way to look things up, so a salesperson can ask about a job status without opening four screens. That means putting a narrow interface between ChatGPT and the ERP rather than handing the model a database login. This page walks through the realistic options and where each one stops being appropriate, especially for shops with customer flow-down terms about where technical data may go. ## Use Cases ### Spreadsheet analysis of exported data **Description:** Export closed jobs with estimated and actual hours, upload to ChatGPT, and ask which part families consistently overrun. Good for a one-time estimating review. ### Quote and email wording **Description:** Draft customer-facing quote clarifications and delivery updates from a few bullet facts you type in, then edit before sending. ### Report explanation **Description:** Paste a variance report layout and ask for a plain-language reading for a supervisor who does not live in the ERP. ### Custom GPT for order status **Description:** A custom GPT with an action calling your own status endpoint, so customer service can ask where a customer PO stands. ### SOP and work instruction drafting **Description:** Turn shop-floor notes into clean work instructions, then store the approved version in your document control system rather than the chat. ## System Surface - Exports: reports and data exports from Global Shop Solutions (including spreadsheet-friendly outputs) are the lowest-risk way to hand data to ChatGPT. - Database access: often ODBC to an Actian Zen style database in on-prem installs. Do not point ChatGPT at it directly. Version and hosting differ, so confirm yours. - Vendor web services: where your version provides documented APIs, a thin wrapper can expose read-only endpoints for a GPT action. - Your own wrapper: a small HTTPS service that returns job, PO and inventory facts in clean JSON, with an OpenAPI description that a custom GPT action can import. - Identity: the ERP user model does not map automatically to ChatGPT users, so your wrapper has to decide what each caller may see. ## How It Works ### Classify the data **Detail:** Decide which fields are safe for a hosted service: item numbers and generic status are often fine, customer drawings and controlled technical data often are not. ### Start with governed exports **Detail:** Produce a sanitised export (remove customer names or part numbers if needed) and use ChatGPT Team or Enterprise rather than a personal account. ### Build the thin API **Detail:** Write a read-only service with two or three endpoints, such as job status by job number and open PO lines by vendor. Authenticate callers and rate-limit. ### Describe it in OpenAPI **Detail:** Give each endpoint a precise description and parameter schema. The quality of those descriptions decides whether the GPT calls the right one. ### Create the custom GPT with an action **Detail:** Import the schema, configure authentication, and write instructions telling the GPT to cite the job number and never guess when a lookup fails. ### Review logs and tighten **Detail:** Check which questions people ask, which return nothing, and which attempt things the GPT should refuse. Adjust scope accordingly. ## Example Flows ### Estimating overrun review **Trigger:** An estimator wants to find where quotes are consistently low. **Steps:** - Export closed jobs from the last two years with quoted and actual hours by operation. - Remove customer identifiers and upload to an enterprise ChatGPT workspace. - Ask for overrun patterns by part family and operation. - Verify the top three findings against the ERP before changing standards. **Result:** A short list of estimating adjustments, backed by numbers the estimator rechecked. ### Order status through a custom GPT **Trigger:** Customer service asks, where is job 10482? **Steps:** - The GPT calls the status action with the job number. - Your API returns current operation, percent complete and planned ship date. - The GPT answers in plain language and states the data timestamp. **Result:** A fast answer without a screen hunt, and a log entry of what was asked. ## On Prem ChatGPT is a hosted service, so by definition the prompt and any attached data leave your network, even with enterprise controls such as no training on business data. For shops handling export-controlled drawings, defense flow-downs or customer NDAs that restrict third-party processing, that may be a deal-breaker. The practical alternative is a private open-weight model on your own GPU server, fronted by a similar chat interface and the same thin API. You keep the ChatGPT-style experience for non-sensitive tasks and send controlled work to the private model. Decide the boundary by data class, and write it down. ## Security - Use a business plan with admin controls rather than personal accounts, and confirm the data-use terms for your plan. - Never paste credentials, database connection strings or customer-restricted drawings into a chat. - Keep the action API read-only, authenticated, and logged, with per-user scoping where possible. - Return only the fields needed for the question, not entire records. - Review customer contracts for clauses on third-party data processing before sending any of their data. ## Options ### Exports into ChatGPT Team or Enterprise **Fit:** Good for estimating reviews, wording help and report explanation with non-sensitive data. **Effort:** Hours to set up a repeatable export routine. ### Custom GPT with an action over your own API **Fit:** Good when staff want conversational lookups and your data policy allows hosted processing. **Effort:** A few weeks to build and secure the wrapper. ### Private model with a partner such as Netray **Fit:** Good when data is controlled or contracts forbid hosted AI. **Effort:** A pilot of a few weeks, hardware dependent. ## Pitfalls - Believing a pasted export is private because the chat feels private. Check the plan terms and your customer contracts. - Letting ChatGPT do arithmetic on large exports without checking. Verify totals in the ERP or a spreadsheet. - Vague action descriptions that cause the GPT to call the wrong endpoint or invent parameters. - Exposing an action that returns entire job records including costs to every user. - Treating the answer as authoritative when the data is a stale export. Always show the as-of time. ## Faq ### Can ChatGPT connect directly to Global Shop Solutions? **Answer:** Not out of the box. You connect through an exported file or a small API you control. Direct database connections from a hosted AI service are not recommended, and your ERP agreement may restrict them as well. ### Is it safe to upload job data to ChatGPT? **Answer:** It depends on the data and the plan. Business plans offer stronger controls, but controlled technical data and contract-restricted information may not be permitted at all. Classify your data first and ask your compliance contact when unsure. ### What is a custom GPT action? **Answer:** It is a defined call from the GPT to an external HTTPS API, described with an OpenAPI schema. It lets the GPT fetch live facts such as a job status instead of relying on pasted text. ### Can it write back to the ERP? **Answer:** Technically possible through your API, but we recommend read-only to start. If you add writes, require human approval and use a service account with narrow permissions. ### When should we pick a private model instead? **Answer:** When the data includes export-controlled or customer-restricted material, when contracts bar third-party processing, or when you want predictable cost for heavy usage. Many shops run both, split by data class. ## Search terms global shop solutions chatgpt, chatgpt for global shop, global shop erp chatgpt integration, custom gpt for erp, chatgpt job shop erp, global shop solutions ai, chatgpt erp data privacy, use chatgpt with on-prem erp --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json