How to add an AI chatbot to SYSPRO
Short answer
An AI chatbot for SYSPRO answers questions from live ERP data through fixed tools over e.NET and a read-only SQL Server login, respects the asking user's role, and shows where each answer came from. It can sit in a web portal, Teams or a customer site, and can run fully on-prem with an open-weight model.
SYSPRO users ask the same questions all day. Where is this order? Do we have stock of this part? What is the price for this customer? When is the PO due? Each answer is a few screens deep, and the person asking is often a salesperson, a customer or a shop-floor lead who does not live in SYSPRO.
A chatbot reduces that load by letting people ask in plain language. It matters that it is grounded: the bot must call SYSPRO data for facts rather than guess, and it must be honest when it cannot find something. A chatbot that sounds confident and is wrong costs more than no chatbot.
There are two audiences with different risk profiles. An internal bot for staff can see broad data under a role. A customer-facing bot should see only that customer's orders, shipments and invoices, with a hard filter applied in code, not in the prompt.
What teams use it for
Customer order and shipment status
Customers log into a portal and ask where an order is, and the bot answers from sales order lines and dispatch data limited to their account.
Sales desk availability and lead time
A salesperson asks if an item can ship by a date, and the bot checks stock, open POs and typical lead time.
Shop-floor how-do-I
Operators ask about job status, current operation and material locations, and the bot answers with the right job data and links to procedures.
Supplier self-service
Suppliers ask which of their POs are open and confirm dates, with replies recorded for a buyer to accept.
Policy and procedure Q&A
Staff ask about returns, credit holds or inspection steps, and the bot answers from controlled documents with citations.
New-hire SYSPRO help
New users ask which program to use for a task and get step-by-step guidance based on your own training notes.
SYSPRO integration surface
- e.NET query objects: provide business-logic-aware reads for orders, stock and purchase orders. Object names and parameters depend on version, so check the documentation for your install.
- SQL Server read-only views: create purpose-built views over order, shipment and invoice tables so the chatbot's access is narrow and stable across SYSPRO updates.
- SYSPRO customer and supplier masters: tables like ArCustomer and ApSupplier hold the identifiers needed to scope portal users to their own data.
- SYSPRO Web UI and portals: where you embed a chat widget, you will integrate with your existing identity provider rather than SYSPRO operators for external users.
- Document sources: SOPs, work instructions and price terms in file shares or SharePoint can feed a retrieval index for policy questions.
How it works
- 1
Choose audience and scope
Decide whether the first release is internal, supplier or customer facing, and list the 10 most common questions.
- 2
Build the data tools
Create functions for order status, availability, PO status and shipment tracking that return JSON, with the user's account scope applied in code.
- 3
Add document retrieval
Index approved SOPs and terms and return passages with source names, kept separate from transactional data.
- 4
Select the model
Use an open-weight model on your own server for on-prem, or a hosted model where policy allows, with tool calling and strong instruction following.
- 5
Write system rules
State that factual answers require a tool result, that the bot should say when no data is found, and that it must not discuss prices unless authorised.
- 6
Embed in the channel
Add the chat to a portal, Teams or intranet page, passing authenticated user identity to the backend.
- 7
Review transcripts weekly
Sample conversations, fix wrong answers by improving tools or documents, and add a human handoff for unresolved questions.
Example flows
Customer asks where their order is
Trigger: A logged-in customer types, 'Where is my order for the brackets?'
- The backend maps the login to a SYSPRO customer code.
- It searches that customer's open orders for matching lines.
- It reads shipped quantity and dispatch date for the matches.
- The bot replies with status, carrier reference if present and next steps.
Result: The customer gets an answer without calling the sales desk, and cannot see other customers' data.
Handoff for a pricing dispute
Trigger: A user asks the bot to explain why their price changed.
- The bot finds no authorised tool for price history.
- It says it cannot explain pricing and offers a handoff.
- It creates a note with the question for the account manager.
Result: A clear boundary and a logged request instead of a guess.
Keeping data on-prem
Chatbots over SYSPRO can be fully on-prem: the web widget talks to a backend inside your network or DMZ, which calls e.NET and SQL Server and sends prompts to an open-weight model on your own GPU server. No customer or order data goes to a third-party cloud. For external users, place only the chat front end and authentication in the DMZ, and keep the model and ERP tools internal behind a firewall rule that allows one narrow path. Plan GPU capacity for peak concurrent users, since interactive latency matters.
Security checklist
- Apply customer or supplier scoping in code, using the authenticated identity, never by asking the model to filter.
- Use read-only credentials and narrow views, and keep bank, cost and margin fields out of tool output.
- Rate-limit and monitor public-facing endpoints, and block prompt content that tries to extract system instructions.
- Log conversations and tool calls with retention rules that match privacy law and customer contracts.
- Provide a clear human escalation route so the bot is never the only path to an answer.
Ways to get it done
Vendor or marketplace chatbot
Effort: Weeks, mainly configuration.
Check SYSPRO and partner listings for a bot that covers your portal and version.
Build in-house on an open model
Effort: 6 to 12 weeks including testing.
Fine for an internal bot with a small tool set and a developer who knows the SYSPRO data.
Partner-delivered private chatbot
Effort: A pilot in weeks, then channels added.
Good for customer-facing or regulated use where scoping, evaluation and on-prem hosting must be right.
Pitfalls
- !Letting the bot answer from memory when a tool returns nothing produces confident wrong answers, so require a source for every fact.
- !Applying customer scope in the prompt rather than the query risks cross-customer leakage.
- !Indexing outdated SOPs gives employees obsolete instructions, so index only controlled, versioned documents.
- !Underestimating GPU needs makes the bot slow at peak, which kills adoption.
- !Launching without transcript review lets small tool errors persist for months.
Frequently asked questions
Can a chatbot answer questions from live SYSPRO data?
Yes, if it calls tools that query SYSPRO through e.NET or a read-only SQL login each time. Without tools, a chatbot only knows what is in its prompt or documents. Tool-based answers should cite the order, item or PO they used.
Can customers use it without seeing other customers' data?
Yes, if the backend maps the authenticated user to one customer code and applies that filter inside every query. Do not rely on instructions to the model. Test with accounts that have similar order numbers to confirm isolation.
Do we need a GPU for an on-prem chatbot?
For acceptable response times with open-weight models, usually yes. Small models run on a single modest GPU for a few users, while many concurrent users need more capacity. Size it from your peak concurrency and test with representative questions.
Is SYSPRO's own assistant enough?
It may be, if its features cover your questions and users and you are on a version that includes it. A custom bot is worth considering for customer portals, supplier access, mixed document and data questions, or strict on-prem requirements.
How do we stop wrong answers?
Require tool results for facts, return clear no-data messages, review sampled transcripts and keep a test set of questions with known answers. Fix problems by improving tools and documents rather than only changing the prompt.
Want AI chatbot working on your SYSPRO?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment