For an importer working with two or three dozen suppliers, email isn't a "communication channel" — it's a production line. Every day it fills up with price lists, invoices, order confirmations, packing lists, freight updates, and certificates. And almost everywhere, that line is worked by hand: a manager opens the email, downloads the attachment, guesses what it is and which brand it belongs to, files it into a folder (or doesn't), copies the invoice amount into a spreadsheet, and sets a payment reminder for themselves. Supplier email automation targets exactly this layer of routine work — and in this article we'll walk through how it works under the hood, using HORUVIA as the example.
Why supplier email is an importer's bottleneck
The problem isn't the volume of email itself — it's that every message demands a micro-decision: what is this, where does it go, what needs to happen, and by when. If you run import operations, these manual-sorting failures will sound familiar:
- The invoice was spotted too late. It arrived Friday evening, the manager opened it on Tuesday — the payment deadline is already closing in, and the shipment along with it.
- The price list was saved to the wrong place. A file called "price_new_final2.xlsx" sits in one employee's downloads folder, and a month later nobody knows which price list is current.
- An offer with a deadline landed in "Other." A supplier sent a pre-season deal with an early-order bonus — the email looked like just another newsletter, and the deadline passed in silence.
- An amount was copied with a typo. Someone retypes figures from a PDF into a payment spreadsheet by hand — sooner or later a digit or a currency goes wrong.
Each of these failures looks minor on its own. Together they add up to a systematic loss of money and time — you can estimate the scale for your own company with our loss calculator.
What AI email processing actually does — and what it doesn't
An honest framing matters more than marketing. AI is good at the parts of supplier email that come down to classification and routing: recognizing that an attachment is a price list rather than an invoice, that a message belongs to brand X rather than brand Y, that the text contains an offer with a deadline. These are recognition tasks, and language models handle them reliably.
Where AI must not be used is money and customs. The invoice amount, currency, and payment deadline should never be "guessed" by a neural network: the model can get it wrong, and the cost of a mistake is a real payment. That's why HORUVIA enforces a strict separation:
- AI classifies emails and attachments, routes them by brand and document type, and spots offers and important events in the text;
- deterministic code — plain parsers with no neural networks — extracts amounts, currencies, and deadlines from invoices. Either the figure is read exactly, or the document is flagged for manual review. Hallucinations in payment data are impossible by design.
This is the core principle we cover in more detail in our article on AI in auto parts import operations: automation in this field only works when the boundaries of AI responsibility are drawn in advance.
Step 1: Connect via IMAP and sort attachments by brand and type
Technically, everything starts with connecting to the mailbox over IMAP — the standard protocol supported by any business email. Nothing needs to be migrated: the system reads the same emails your manager does, it just does so continuously rather than whenever someone gets around to it.
From there, every incoming email with attachments goes through two levels of sorting:
- By brand / supplier. The system identifies who the document is from — by sender, domain, and the content of both the email and the file itself. An auto parts importer may deal with dozens of suppliers, each with its own folder, its own history, its own context.
- By document type. Price list, invoice, order confirmation, shipping document, stock report — each type flows into its own processing pipeline. A price list updates the pricing database, an invoice kicks off the payment workflow, a shipping document feeds cargo tracking.
The result: instead of a pile of emails in the inbox, a structured archive where every supplier's documents are visible by type and date. Filing price lists stops being somebody's chore — it simply happens.
Step 2: Invoice processing — amount, currency, deadline, no hallucinations
Once the system recognizes an attachment as an invoice, deterministic processing takes over. Three critical fields are extracted from the document:
- the amount due;
- the payment currency;
- the payment deadline — an explicit due date or deferred-payment terms.
The key point: the extraction is done by code, not by a language model. The parser either finds the field and reads it character by character, or honestly reports that it couldn't — and the document goes to a person for review. There is no "the model thinks the amount is roughly this." For invoice processing, where one wrong digit means a bad payment or trouble at customs, no other approach is acceptable.
The payment draft: AI prepares, a human approves
From the extracted data, the system automatically creates a payment draft: supplier, amount, currency, deadline, and a link to the original invoice. Crucially, it is exactly that — a draft. HORUVIA doesn't execute payments and has no access to your money. The final call always belongs to a human: they open the draft, check it against the document, and approve.
What this changes in practice:
- a payment no longer stalls because someone forgot to log the invoice in a spreadsheet — the draft appears right after the email arrives;
- people spend their time on checking and deciding, not on retyping figures;
- every payment has a trail: which email the invoice came from, when, and what was in it.
This "AI prepares — a human decides" split is the standard pattern for anything that touches money, and it underpins the other scenarios we describe in our overview of the AI assistant for business.
Telegram alerts: an invoice arrived, a deadline is looming, a certificate is expiring
Sorted email is useless if the important events still reach you late. So on top of the processing sits a Telegram notification layer — right where the owner and the purchasing manager actually spend their day. The system sends alerts on events:
- an invoice arrived — with the amount, currency, and deadline, before anyone has even opened the mailbox;
- an offer with a deadline was detected — a pre-season deal, a bonus promotion, a time-limited term;
- a certificate is about to expire — the system tracks certificate validity for every supplier and warns you in advance, not at the moment the document is needed at customs;
- "went down → back up" — if any processing step fails, you get a warning, and when it recovers, a mandatory paired all-clear. No alarms left hanging without a resolution.
Deadline offers that used to slip away
There is one category of email that pays for the whole automation on its own: commercial offers with a limited window. Auto parts suppliers regularly send out pre-season terms, volume bonuses, early-order discounts. The trouble is that these emails look exactly like ordinary newsletters, so manual sorting routinely dumps them into "Other" — the deadline passes, and the company ends up ordering the same items on standard terms.
AI treats this as exactly what it is — a recognition task: the offer detector analyzes the email text, picks out the markers of a deal — terms, deadline, bonus — and immediately sends a Telegram alert with the gist and the time limit. The decision, again, stays with a human: take the offer or pass. But now it's a conscious decision, not a silent loss.
A memory for every supplier: documents, prices, certificates
Email processing is also about building a structured history. For every supplier, the system keeps a memory: every document received, the history of price lists and prices, certificates with their validity dates. HORUVIA's working database today holds prices for 266,000 items from 27 brands — collected from incoming documents, not entered by hand.
This memory isn't an archive for its own sake. Practical features run on top of it:
- order recommendations are calculated by a set procedure that accounts for seasonality, ABC classification, and current stock — based on real prices and delivery history;
- Company Pulse shows shipments by stage and highlights what needs attention right now;
- any dispute with a supplier is settled with facts: here's the email, here's the document, here's the date.
How to roll out email automation without breaking your processes
The good news: this kind of automation doesn't require changing your mail server, retraining your suppliers, or migrating to a new system. A sensible rollout looks like this:
- Connect the mailbox over IMAP in read-only mode — the system starts processing incoming mail alongside your people, deleting and moving nothing in the original mailbox.
- Check the sorting quality on your own flow: are brands and document types identified correctly, and is every invoice being recognized?
- Turn on Telegram alerts — from that moment, your team learns about invoices and offers the minute they arrive.
- Enable payment drafts and plug them into your existing payment approval process.
At every step, a human stays in the loop: the system takes over the mechanics, people keep the decisions. That's exactly why this format sticks in places where "fully autonomous" solutions meet justified skepticism.
If you're dealing with a steady flow of supplier emails and want to see how this works on your own documents, request a free demo on the HORUVIA homepage. And to estimate what manual email sorting and missed offers are costing your company, try the loss calculator — it takes a couple of minutes.