Skip to content

An emailed order becomes a ready-to-review ERP draft.

The sales team reviews a short exception queue instead of retyping the whole order. Every held line shows the possible matches and the reason it stopped.

Email orders into ERPDemonstration systemTest one order from your inbox

Measured on test data

The system read an unstructured order. Catalog, unit, and substitution rules held an uncertain line for the operator, while the history retained the input, decision, and result.

Recorded run
Measured on test data
Measurement scope
Email orders into ERP
Verified
synthetic data

Problem, solution, and result

  1. The problem

    B2B orders arrive in emails, PDFs, and spreadsheets. Sales staff retype lines into the ERP, interpret units, and search for product codes. Mistakes often surface during picking or delivery.

  2. How the system works

    The system reads the order and checks every line against the product catalogue. Confirmed products enter the ERP draft. An unclear code, unit, or substitute is held for the operator with possible matches.

  3. What we tested

    The system read an unstructured order. Catalog, unit, and substitution rules held an uncertain line for the operator, while the history retained the input, decision, and result.

For whom

This is a good automation candidate when orders arrive by email, staff retype them into an ERP, and product-code errors appear during fulfillment.

Email → catalogue → ERP draft

  1. 01The system reads each line and checks the catalogue
  2. 02Confirmed lines form the order draft
  3. 03Exceptions wait for the operator
Business type
B2B wholesalers and distributors working in an ERP
Input
Email and files with orders
Boundary
A line without catalogue confirmation waits for the operator's decision
Cost
About PLN 0.0027 for the recorded read of one order.
Security
An uncertain line never reaches the ERP. The operator sees possible matches and makes the decision.
Speed
We measure the full order-entry time in a pilot at the client’s actual volume.
Trace
The history shows extracted lines, rule decisions, and the operator’s approval.
Similar system build
from €3,500 net (approx. $3,800) · 2–6 weeks

Where automation stops

When the operator decides

A line goes to the operator when the catalogue cannot confirm it, the unit is unusual, or a substitute remains ambiguous. The screen shows why it stopped and which matches were rejected. The operator resolves it before the draft is approved.

Boundary
A line without catalogue confirmation waits for the operator's decision
Cost
About PLN 0.0027 for the recorded read of one order.
Security
An uncertain line never reaches the ERP. The operator sees possible matches and makes the decision.
Speed
We measure the full order-entry time in a pilot at the client’s actual volume.

Estimated impact

Calculate the impact at your volume

This estimate uses the stated volume. Enter your own numbers to assess the possible impact in your company. The result is confirmed only during a pilot.

  1. Today

    144 h

  2. With the system

    56 h

  3. Time or cost saved

    70-105 h recovered/mo, base 88 h

Volume
480 orders/mo
Formula
480 x 11 min / 60
Calculation status
medium

Data in the screenshots. The names, amounts, and documents shown in the screenshots are synthetic. Client data remains private. Measurements describe the demonstrator and are kept separate from production outcomes.

Working surfaces

The operator resolves exceptions and approves the draft. They no longer retype the entire order.

Routine orders appear as drafts. Sales staff open only the exception queue and approve the completed document. Handling time and ERP integration accuracy are measured during a pilot on client data.

Three views for order intake

Order register

Every incoming order, its status, and value in one place.

Exception queue

Held lines with the reason and possible catalogue matches.

ERP draft

Confirmed lines ready for approval and transfer through an integration.

System screens

See how the system works in practice

These desktop and mobile screenshots come from the running application. They show the process and the points where a person makes the decision.

Screens
10
px
1440 · 390
011440×1100
A register of 480 test orders with the status of each case.
390390×844
  1. 021440×3393
    An exception queue with candidate matches and a reason for each hold.
  2. 031440×1100
    The same queue in a mobile layout without horizontal scrolling.
Open the remaining screen archive (7)
  1. 041440×1100
    Lines to check
  2. 051440×1100
    Item file and stock
  3. 061440×1100
    Sales orders to raise
  4. 071440×1100
    Restart checkpoint and decision history
  5. 081440×1100
    A single test record
  6. 091440×1100
    Intake rules
  7. 101440×1100
    Rule measurement results

Technology stack

A fixed workflow built from testable rules

gpt-5-mini reads the order, n8n controls the sequence, and code performs catalogue matching and enforces the write boundary. The trace therefore covers the model response and every later rule decision.

n8n
runs the flow: the webhook entry, routing, and step order
FastAPI (Python)
validates the input contract and executes the typed matching decisions
PostgreSQL 17
holds orders, lines, decisions, and the audit chain
Next.js
the inflow register and the operator's review station
Docker Compose
runs the whole stack locally: postgres, api, n8n, web

The client receives the rules, n8n workflow, test data, and documentation. The system can run in their infrastructure. This version does not connect to a production ERP.

Technical details and measurements

The working loop

From document to draft, one line at a time

Every line passes through the same checks. A confirmed result enters the draft. An unconfirmed result becomes an operator task with the relevant context.

01

Email or order file

02

Product and unit recognition

03

Confirmed lines → ERP draft

Exceptions → operator with an explanation

04

Decision record and draft approval

System architecture

Four parts: intake, rules, draft, and decision record

n8n runs the workflow, code checks the catalogue and units, PostgreSQL records each step, and the operator screen shows exceptions.

  1. 01

    Input

    One entry point for email and files.

    An order from email or a file arrives through a single webhook entry, and n8n runs the routing and the order of steps. Intake immediately opens a run record with a snapshot of what came in, so what arrived and when stays visible. The call ends with the decision returned, so the sender knows what the system did.

  2. 02

    Matching

    SKUs, units, and substitutes are computed by code.

    SKU matching, unit and packaging conversion, and substitute assessment are deterministic rules with a typed result. The same input always produces the same decision, and that decision can be tested.

  3. 03

    Write

    Only a confident result enters the draft.

    A line with catalogue confirmation becomes part of the reference implementation draft. A line without confirmation has no route forward; instead, an operator task is created with the full context. The ERP adapter is not part of this reproduction.

  4. 04

    Trace

    Runs, decisions, and tools in the audit.

    Every run records its steps, decisions, and tool calls in an audit chain, so any line can be replayed after the fact: what the system saw, what it chose, and why.

Why there is no agent here

Every line follows a fixed pipeline: recognize, convert, match, write or escalate. The next step is decided by process rules you can read and test. The documented test run showed that no decision in this pipeline needed a model call. The three judgment nodes stay open for a model if a client's catalogue turns out to be harder than the rules, and we would measure whether it improves the decisions first.

  • The next step is chosen by process rules you can read
  • The language model performs no action in this process
  • A line without catalogue confirmation always reaches the operator
  • Every decision leaves an entry in the audit chain

Want to assess a similar process in your company?

  • A 30-minute call with the engineer who would lead the work.
  • A review of the processes that cost you the most time and money.
  • A written summary of what to automate first and the likely cost range.
€030 minutes · written takeaway within 2 business days
Book a free process scan (30 min)

Times are shown in your own time zone. We work with clients across time zones.

The scan is free and creates no obligation. If automation is unlikely to pay off, the written recommendation will say so.