The conversation may be incomplete
A missing customer name, address, time, or price is often absent from the source—not a fact AI should invent.
AI can reduce retyping and expose missing details, but it should not invent customer intent. DMNexa creates a private working draft that the seller verifies before confirmation.
Address and delivery time are missing
A model can recognize common order language, but it cannot guarantee that every date, negative instruction, or changed choice was interpreted correctly.
A missing customer name, address, time, or price is often absent from the source—not a fact AI should invent.
Customers revise options, use shorthand, or mention several possibilities before choosing one.
An editable field can have a fallback value while the review warning still says verification is required.
The AI step is part of a larger validation pipeline controlled by the seller.
DMNexa uses the recent stored messages available for the conversation the seller selected.
The model looks for customer, items, quantity, customizations, fulfillment, address, date, price, and notes.
Visible AI Brain rules can guide which details deserve attention for that seller.
DMNexa stores a reviewable draft with missing fields, warnings, and editable sections.
No customer receives the AI result until the seller corrects it and chooses the approval action.
The product is designed around reviewability rather than pretending the model is always correct.
Organize plain-language requests into predictable order fields.
Keep removals such as “no onions” or “no nuts” visible in the item details.
Surface gaps instead of silently inventing a reliable customer fact.
Use AI Brain instructions to highlight the checks that matter to the seller.
Correct customer, items, address, fulfillment, dates, pricing, and notes.
Treat AI analysis as preparation, never acceptance.
Clear limits make the draft safer and easier to evaluate.
| Task | AI can assist | Seller remains responsible |
|---|---|---|
| Find stated item details | Extract likely values | Verify exact choice and quantity |
| Identify missing information | List gaps and warnings | Ask the customer and fill them |
| Apply review rules | Flag relevant language | Decide whether the order is acceptable |
| Handle dates and defaults | Map explicit wording | Confirm the real due date and time |
| Send the final order | Prepare the draft | Approve and choose when to send |
AI drafts are especially helpful when requests are detailed enough to require retyping but still need human judgment.
These answers describe the current supported workflow and its boundaries.
The product may use a configured AI provider behind DMNexa’s validation and order workflow. The quality of a result also depends on source context, seller rules, defaults, matching, and validation—not only the provider.
No. Confidence is an aid, not a guarantee. Dates, quantities, addresses, price, changed choices, and negative instructions always deserve seller review.
No. Creating a draft does not send it. The seller reviews the order and chooses when to approve and send an eligible customer link.
Yes. The review screen provides editable customer, item, fulfillment, address, date, pricing, and note sections.
Start with a private draft, keep uncertainty visible, and let the seller own the final order.