# How to add an AI chatbot to Infor M3 > Source: https://netray.co/integrations/infor-m3-ai-chatbot > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer An AI chatbot for Infor M3 answers staff or customer questions by retrieving data through M3 API calls and, for policies and procedures, from an indexed document set, then responding in plain language with sources. It can sit in a web portal, Teams, or a customer site and can run on open models inside your network. ## System Infor M3 ## Capability ai-chatbot ## Intro Most M3 users do not live in M3. Sales reps, shop supervisors, buyers outside the core team and customers ask the same handful of questions again and again: where is my order, do we have this item, when will it ship, what is the lead time. Each question lands on a planner or customer service desk. A chatbot removes the routine ones by looking up the answer itself. The key requirement is grounding. It must fetch the actual order status or balance from M3, quote what it found, and say when it cannot find something, rather than answering from general knowledge. This page treats the chatbot as a product with two audiences, internal staff and external customers, because the authorisation model and tone differ. ## Use Cases ### Customer order status portal **Description:** Customers log in and ask about their own orders; the bot looks up only that customer's lines, dates and shipment tracking. ### Internal sales assistant **Description:** Reps ask about stock, lead times and recent order history for an account from a Teams chat or web panel. ### Shop floor Q and A **Description:** Supervisors ask for the status of a manufacturing order or the next operation, in simple language on a tablet. ### Procedure helper **Description:** Staff ask how to perform a task in M3, answered from your own work instructions and training notes. ### Supplier self-service **Description:** Suppliers confirm PO line dates and ask what is outstanding, with replies recorded as proposals for buyers. ### Escalation with context **Description:** When the bot cannot answer, it hands over to a person with the conversation and looked-up data attached. ## System Surface - M3 API (MI programs): real-time lookups for customer orders, items, balances, POs and manufacturing orders. Program and transaction names vary by version. - ION API Gateway for cloud tenants, using OAuth 2.0 service accounts, or the on-prem M3 API access method configured by your administrators. - Infor Document Management or file shares: source for procedures, product documents and certificates the bot can retrieve and cite. - Identity: your portal or Entra ID user maps to an M3 user or a customer number, which determines what data the bot may query. - Infor OS and M3 Mashups: an existing embedding point for UI components inside the Infor user experience, if you prefer the bot inside the current interface. ## How It Works ### Define audience and scope **Detail:** Separate internal and external bots. Write the top twenty questions each should handle and what it must refuse. ### Build the lookup tools **Detail:** Wrap M3 API calls as functions with strict filters, such as customer number locked to the signed-in user. ### Index documents **Detail:** Chunk and index work instructions and product sheets in a private vector store, keeping source references and access labels. ### Choose the interface **Detail:** Embed a chat widget in your portal, add a Teams bot for staff, or surface it inside Infor OS where supported. ### Ground and cite **Detail:** Require answers to include the order or document reference they came from, and a fallback message when no data is found. ### Add handoff **Detail:** Create a ticket or email to the right team with the transcript when confidence is low or the user asks. ### Test with transcripts **Detail:** Use last quarter's support emails as test questions and measure correct, partial and wrong answers before launch. ## Example Flows ### Customer asks about a delayed order **Trigger:** A signed-in customer types: Where is order 100245? **Steps:** - The bot checks the order belongs to that customer number. - It calls the order status lookup through the M3 API. - It reads line statuses, planned delivery dates and any carrier reference. - It replies with the current date and offers to notify customer service. **Result:** The customer gets an immediate answer without calling anyone. ### Supervisor checks a work order **Trigger:** A supervisor asks for the status of manufacturing order 4001234 on a tablet. **Steps:** - The bot fetches order header and operations from the M3 API. - It identifies the current operation and any material shortages. - It replies with progress and the next step. - If a shortage exists, it offers to message the planner. **Result:** The supervisor avoids walking to a terminal and the planner is notified with context. ## On Prem A chatbot is a good fit for on-prem because the sensitive part, ERP data, never has to leave. Host the chat service, the retrieval index and an open-weight model on servers in your network using vLLM or Ollama. For an internal bot, users reach it over the corporate network or VPN. For a customer-facing bot, place the web tier in a DMZ and have it call an internal service that holds the model and the M3 credentials, so the internet-facing component never sees the ERP directly. For modest question volumes a single GPU is often enough, since most answers are short. ## Security - Bind every lookup to the authenticated identity, for example customer number from the session, not from the chat text. - Keep M3 credentials in the backend only, with read-only roles for the bot. - Filter retrieved documents by access label so customers never see internal procedures. - Rate limit and monitor the external bot to prevent data scraping through repeated questions. - Log conversations with retention suited to your privacy obligations, and tell users it is an AI assistant. ## Options ### Vendor-provided assistant or digital assistant in the Infor stack **Fit:** Teams wanting standard help inside the Infor interface and accepting its scope. **Effort:** Low to medium; limited customisation to your data and flows. ### Build a bot on a framework with your own tools **Fit:** Teams with developers who can maintain retrieval, tools and testing. **Effort:** Four to eight weeks for a first audience. ### Partner deployment with private models, such as Netray **Fit:** On-prem, regulated or customer-facing bots needing tight grounding. **Effort:** Scoped pilot then expanded question coverage. ## Pitfalls - Letting the model answer from general knowledge when lookup fails, producing confident wrong dates. - Taking the customer number from the typed message instead of the login. - Indexing internal documents without access labels and then exposing them externally. - Launching without a handoff path, so stuck users abandon the bot. - Measuring only deflection, not whether answers were correct. ## Faq ### Can the chatbot run entirely inside our network? **Answer:** Yes. Use an open-weight model served locally, a private retrieval index, and backend tools calling the M3 API internally. A customer-facing version needs a DMZ web layer, but the model, index and M3 credentials can stay on internal servers. ### How do we stop the bot giving wrong order dates? **Answer:** Make it answer only from tool results, require a reference to the looked-up order, and return a clear not-found message otherwise. Test with real past questions, and track wrong answers as a metric alongside resolution rate. ### Should customers and staff use the same bot? **Answer:** Usually no. They need different tools, permissions and tone. A shared engine is fine, but separate configurations, document sets and identity rules reduce the risk of exposing internal data to customers. ### Does it need ION? **Answer:** Not necessarily. Lookups can use the M3 API directly. ION helps if you want event-driven notifications, such as proactively messaging a customer when a delivery date changes, or if your cloud tenant routes API access through the ION API Gateway. ### Can the chatbot create orders? **Answer:** It can, technically, but start with lookups and handoffs. Creating or changing orders needs validation, pricing rules and approval, so many teams have the bot collect details and create a draft for a person to confirm. ## Search terms infor m3 chatbot, ai chatbot for infor m3, infor m3 virtual assistant, infor m3 customer portal chatbot, chatbot on infor m3 data, infor m3 order status chatbot, on-prem chatbot infor m3, infor m3 conversational ai, add chatbot to infor m3 --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json