How to automate email in IFS Cloud
Short answer
Automating email for IFS Cloud means reading a shared mailbox through Microsoft Graph or IMAP, classifying and extracting each message with a model, and creating staged records in IFS through projections or IFS Connect. A human reviews low-confidence or high-impact items before anything posts.
Even on a modern IFS Cloud estate, a surprising amount of work arrives by email. Customers send purchase orders as PDFs and change requests in prose. Suppliers confirm, push out or partially ship by replying to a PO email. Quality receives certificates and nonconformance notes as attachments. Someone then retypes it into Aurena.
IFS can send email well, through its mail and notification features, and has some inbound handling options. What it does not do on its own is understand a messy human email and turn it into the right customer order line or promised date. That is a classification and extraction problem sitting in front of a transaction.
The design that works keeps IFS as the system of record, uses staging so wrong guesses are cheap, and treats the mailbox as an untrusted input. The notes below cover mailbox access, extraction, matching to IFS keys and write-back.
What teams use it for
Customer purchase orders to customer order drafts
Read the PO PDF and email body, match the customer and parts, and create a draft customer order for the sales administrator to check.
Supplier order acknowledgements
Match replies to open purchase order lines and propose updates to promised delivery dates and quantities.
RFQ intake
Detect requests for quote, pull part numbers and quantities, and open a quotation draft with the original email attached.
Change and cancellation requests
Spot an email asking to move or cancel an order, link it to the order, and raise a task for the owner rather than changing anything.
Shipping and delay notices
Extract carrier, tracking and revised dates from supplier notices and attach them to the relevant purchase order.
Certificates and quality emails
File certificates of conformance against receipts and route nonconformance reports to quality with a draft record.
IFS Cloud integration surface
- Outbound mail: IFS Cloud can send notifications and documents by email through its mail configuration, which is useful for confirmations back to the sender.
- IFS Connect: provides inbound and outbound messaging, and in some setups mail-based connectors. Inbound mail features and names vary by release, so confirm what yours offers before relying on them.
- Projections: the supported OData path to look up customers, parts, open orders and to create or update draft records.
- Document Management and attachments: store the original email and attachments against the created record using document reference objects.
- Master data lookups: customer, supplier, part catalog and customer-specific part numbers are the keys that matching depends on.
How it works
- 1
Provision the mailbox access
Use a shared mailbox with Microsoft Graph application access limited to that mailbox, or IMAP with an app password where Graph is not an option.
- 2
Triage and classify
Classify each message as PO, acknowledgement, RFQ, change request, invoice or other, and drop marketing and out-of-office noise early.
- 3
Extract structure
Pull header and line data from the body and attachments into a defined schema, with a confidence score and a source quote for each field.
- 4
Match to IFS keys
Resolve customer and supplier ids, map customer part numbers to your part numbers and find the open order the email refers to.
- 5
Stage, do not post
Write the result to a staging table or a draft record with the email attached. Show it in a review screen next to the original.
- 6
Post with a gate
After approval or when confidence and rules pass for a trusted sender, call the projection under a service identity and record the outcome.
- 7
Reply and learn
Send an acknowledgement from IFS or the mailbox, and capture corrections to improve matching rules.
Example flows
Customer PO PDF to draft order
Trigger: An email with a PO attachment arrives in the orders mailbox
- Classifier marks it as a customer PO
- Extraction reads PO number, ship-to, lines, dates and prices
- Matcher resolves the customer and customer part numbers
- A draft customer order is created through the projection with the email attached
- Sales admin reviews differences and releases it
Result: Data entry shrinks to review, and mismatches against price lists are flagged before booking.
Supplier delay reply
Trigger: A supplier answers a PO email saying lines will ship late
- Reply is matched by PO number and thread
- Extraction captures new dates per line
- A proposal is staged for the buyer showing old and new dates
- Buyer accepts and the projection updates the promised dates
Result: MRP sees real supplier dates sooner, without the buyer retyping.
RFQ to quotation draft
Trigger: A customer emails a request for quote
- Classifier marks it as RFQ
- Parts and quantities are extracted and matched to the catalog
- Unknown items are listed for engineering
- A quotation draft is created and assigned to a salesperson
Result: Quote turnaround starts with a structured draft instead of an inbox item.
Keeping data on-prem
Mailbox contents often include pricing, drawings and program details, so keep the pipeline in your environment. Run the extraction model as an open-weight model on your own GPUs, served through vLLM or Ollama, and keep staging in a database you control. Microsoft 365 mailboxes can still be read through Graph from an on-prem service, or you can use an on-prem mail server over IMAP. For IFS Cloud in a customer-managed deployment, the service reaches projections over the internal network, so neither email text nor attachments go to an external AI API.
Security checklist
- Treat email content as untrusted: strip active content, scan attachments and never let message text act as an instruction to the system.
- Use a service identity with permissions to create drafts, not release or approve.
- Restrict mailbox access to the specific shared mailbox with application access policies.
- Keep the original email as evidence and log every extraction and approval.
- Block auto-posting for new senders, changed bank details and unusual amounts.
Ways to get it done
Use IFS mail and Connect features with rules
Effort: Weeks, mainly configuration
Structured, predictable messages from a few partners
Build a mailbox pipeline with an open-weight model
Effort: Roughly two to three months for a first document type
Varied PO and acknowledgement formats, on-prem data
Partner delivery with Netray and your IFS partner
Effort: Phased, usually starting with one mailbox
Multiple mailboxes, review UI and matching to IFS masters
Pitfalls
- !Auto-posting orders from unverified senders, which invites fraud and spoofing.
- !Matching parts by description when customer part numbers are the real key.
- !Ignoring email threads, so a later correction creates a duplicate order.
- !Dropping the original attachment, which leaves reviewers unable to verify the extraction.
- !Measuring accuracy by field instead of by posted-transaction correctness.
Frequently asked questions
Can IFS Cloud read inbound email by itself?
IFS offers messaging and mail-related features through IFS Connect, but behavior and names depend on release and configuration. They suit structured messages. For free-form emails with PDFs, you generally add a classification and extraction layer in front of IFS.
Is it safe to let AI create customer orders automatically?
Start with drafts and review. Allow auto-posting only for known senders and templates, with price and quantity checks and a clear rollback. Many teams never auto-post, because reviewing a good draft already removes most of the effort.
What do we need for Microsoft 365 mailbox access?
An Entra app registration with application permissions to read mail, restricted by an application access policy to the one shared mailbox. This avoids broad access to every mailbox in the tenant, and avoids storing a user's password.
How do we reduce wrong matches?
Use deterministic keys first: PO numbers, customer part numbers and thread ids. Then use a model for fuzzy cases, with a confidence threshold. Anything below the threshold goes to a person with candidate matches listed.
Can this run without sending emails to a public AI service?
Yes. An open-weight model on your own servers can classify and extract from email text, and OCR can read attachments locally. Quality depends on model size and document variety, so test on a few hundred of your real emails before committing.
Want Email automation working on your IFS Cloud?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment