Real projects, real outcomes

Each of these started as a manual process consuming hours of work every week. Below is what the automation replaced and what the result was.

AI Workflow · Email Automation

AI-Assisted Email Processing Workflow

An n8n-based system that triages, classifies and generates draft responses for inbound business emails — reducing manual handling time significantly across supplier and customer communications.

The Problem

An e-commerce operation receiving 60–80 inbound emails daily across supplier queries, customer order issues, return requests and logistics notifications. Each one was read, categorised and responded to manually — a task that occupied 1–2 hours every working day without adding any business value.

The challenge wasn't just volume. Different email types required different handling: some needed immediate replies, others required escalation, and many were purely informational. There was no systematic way to tell them apart without reading each one.

The Solution

An n8n workflow triggered on new email arrival. Each inbound message is passed to an AI model which classifies it by type — order query, return request, supplier communication, logistics update, or other — and extracts the key details: order reference, customer name, issue description.

For types with standard responses, the AI generates a context-aware draft. A human reviews the draft in a simple approval interface before it's sent. Non-standard emails are flagged and routed to a review queue. The entire classification and draft step takes under 30 seconds.

The Result
~4 hours saved per week

Roughly 70% of inbound emails now route through the automated draft-and-approve flow. The remaining 30% are still handled manually, but the triage step alone eliminates most of the reading-and-sorting overhead.

Response time on standard queries dropped from several hours to under 20 minutes. The human-in-the-loop approval step was kept deliberately — speed matters less than accuracy for the messages that still require personal attention.

Stack used
n8n · OpenAI API · Gmail API · PostgreSQL
Business type
Multi-channel e-commerce (Amazon, eBay, website)
Build time
~2 weeks from brief to live
Discuss a similar project

Data Pipeline · Accounting Integration

Marketplace CSV to Accounting Automation

A fully automated pipeline that takes Amazon FBA VAT transaction CSV exports, processes them and creates correctly attributed sales invoices in Sage Business Cloud — with zero manual input.

The Problem

Every week, Amazon generates a VAT transaction report as a CSV export. The file contains hundreds of line items — sales, refunds, fees, FBA charges — each needing correct VAT treatment depending on the buyer's country and transaction type.

The existing process involved opening the file in Excel, manually mapping rows to customer records in Sage, calculating VAT for each transaction type and creating invoices one by one. This took 3–4 hours every settlement period and was prone to VAT classification errors.

The Solution

An n8n workflow triggered on a file drop watches for new settlement exports. The CSV is parsed, and each row is evaluated against a VAT logic tree — GB standard rate, zero-rated international, EU distance selling — and grouped by transaction type.

For each unique customer, the system checks Sage for an existing contact record. If none exists, a new contact is created using data extracted from the Amazon invoice PDFs. Finally, sales invoices are posted to Sage via API with the correct line items, VAT codes and ledger mappings.

The Result
Zero manual entry

The entire weekly settlement process now runs unattended. A notification is sent when the workflow completes, with a summary of invoices created and any records flagged for review.

VAT classification accuracy improved immediately — the logic is consistent and auditable. The 3–4 hours previously spent on data entry each week were fully eliminated, freeing up that time for work that actually requires a human.

Stack used
n8n · Sage API v3.1 · Node.js · PostgreSQL
Business type
Amazon FBA seller — UK and EU markets
Build time
~3 weeks including VAT logic
Discuss a similar project

PDF Processing · Multi-language

Invoice PDF Data Extraction System

An automated system that extracts structured data from supplier and platform invoices arriving as PDFs — in multiple languages — and posts the transactions to accounting with correct exchange rate handling.

The Problem

An e-commerce business sourcing from European suppliers and using Amazon's EU fulfilment network receives invoices from companies in France, the Netherlands, Italy and Germany — each formatted differently and in the local language.

Each invoice had to be opened, the key figures identified (total net, VAT amount, exchange rate if applicable, invoice date, supplier reference), manually converted if in EUR, and entered into the accounting system. For a business processing 30–50 supplier invoices per month, this added up to a meaningful amount of time.

The Solution

A workflow monitors an inbox folder for new PDF attachments. Each PDF is processed through an extraction layer that identifies invoice fields regardless of language or layout — using a combination of structured parsing and AI-assisted extraction for documents that don't follow standard formats.

For EUR invoices, the system fetches the relevant exchange rate from the invoice date and applies the correct conversion. The extracted data is then validated against expected fields and posted to Sage as a purchase transaction, with the supplier matched to an existing contact or a new one created.

The Result
Multi-language, zero re-entry

The system currently handles invoices in English, French, Dutch and Italian reliably. German is supported with slightly lower accuracy on non-standard formats and includes a review flag for those.

Monthly invoice processing time dropped from approximately 4–5 hours to under 30 minutes of review and exception handling. The accuracy rate on extracted fields is consistently above 95% across languages, with lower-confidence extractions flagged automatically for human check.

Stack used
n8n · OpenAI API · Sage API v3.1 · Exchange Rate API
Languages supported
English · French · Dutch · Italian
Build time
~2 weeks from sample invoices to live
Discuss a similar project

Have a similar process you'd like to automate?

Bring the process, the data format and the tools you're already using. We'll work out what's realistic and what it would take to build.