Every logistics business we've spoken to has the same problem: invoices. Dozens arriving daily by email in different formats from different suppliers, each needing to be checked against purchase orders, coded to the right cost centres, approved, and entered into the accounting system. It's not complex work, but it's slow, error-prone, and it keeps smart people chained to a spreadsheet instead of doing something that matters.
This is the story of how one UK-based freight and logistics business solved it — in under two weeks, at a fixed price of £2,497.
Note on client confidentiality: The company name has been changed, but the situation, process, numbers, and outcomes are real and representative of this type of project. We've called them "Meridian Freight Solutions" — a fictional name for a real business based in the East Midlands.
The Situation at Meridian Freight
Meridian Freight Solutions is a regional freight forwarding and logistics business with around 45 employees. They manage around 200–300 shipments per month for clients across manufacturing, retail, and distribution. Their operations team of 12 handles everything from booking to delivery confirmation to supplier invoicing.
The invoicing problem had grown slowly. When the business was smaller, two people could handle the supplier invoices manually without it being a significant burden. But as volumes grew, the same two people were spending a combined 40+ hours per month just on invoice processing — receiving invoices by email, extracting the relevant data, checking it against purchase orders in their freight management system, coding it to the right cost centre and client matter, and entering it into Xero.
The errors were the bigger issue. Invoices arrived in different formats: some as PDFs, some as Excel attachments, some as body text in emails. Line items were labelled differently by different suppliers. VAT treatment varied. When the operations team were tired or under pressure — which in logistics is frequently — mistakes happened. Wrong amounts, wrong cost centre codes, invoices matched to the wrong purchase order. Each error took 20–45 minutes to investigate and correct, and some didn't get caught until month-end reconciliation.
What they had tried before
Meridian had looked at two options before coming to LoopStack. The first was an off-the-shelf accounts payable automation tool — expensive (£800+/month), required extensive setup, and was designed for enterprise finance teams with complex approval hierarchies, not an SME freight company. The second was asking their accounting software provider about their automation features — they were told it would require a "custom implementation project" with a quoted price of over £15,000.
Neither option made sense. They needed something that worked with their existing tools (Xero, their freight management system, Gmail, and a shared Google Sheets tracker) and cost a fraction of what they'd been quoted.
The Scoping Session
Our first conversation with Meridian's Operations Director, Claire, lasted 45 minutes. In that time we mapped the exact current process:
- Supplier invoice arrives in a shared Gmail inbox (invoices@meridianfreight.co.uk)
- Operations team member opens the email, downloads any attachment
- Extracts: supplier name, invoice number, invoice date, due date, line items (description + amount + VAT), total amount
- Opens the freight management system, finds the corresponding purchase order
- Checks the invoice amounts against PO amounts — flags discrepancies
- Assigns the correct cost centre code from a lookup table
- Enters everything manually into Xero as a bill
- Updates the shared Google Sheets tracker with invoice status
- Files the original email
End-to-end: 8–20 minutes per invoice depending on complexity. At 200–300 invoices per month, that's 27–100 hours. Their actual measured time was around 42 hours per month average.
We identified three main variables that drove complexity: the format of the invoice (PDF vs Excel vs email body text), the number of line items (single-line vs multi-line), and whether the PO reference was included in the invoice (it wasn't, about 40% of the time).
The key insight from scoping: The AI component would handle format variance and data extraction. The PO-matching would use fuzzy matching logic (supplier name + amount + date proximity). The cost centre assignment would use a rules table we'd build with Claire during the project. Human review would only be required for genuinely ambiguous cases — projected at under 5% of invoices.
The Solution We Built
The automation runs as a Make.com scenario triggered every 15 minutes. Here's how it works:
Step 1: Email monitoring and document extraction
Make.com watches the invoices@ Gmail inbox. When a new email arrives, it captures the email body and any attachments (PDF or spreadsheet). For PDF attachments, it routes them to a Claude AI extraction step. For spreadsheet attachments, it uses Make's native data parsing. For invoices embedded in the email body, it uses Claude to parse the plain text.
Step 2: AI data extraction via Claude
We wrote a prompt for Anthropic's Claude that instructs it to extract structured invoice data from whatever format it receives. The prompt is specific: extract supplier name, invoice number, date, due date, each line item with its description and amount, subtotal, VAT amount, and total. Output as structured JSON. Flag any fields that couldn't be extracted with confidence.
Claude handles format variance naturally — whether it's a clean professional PDF from a major carrier or a scrappy plain-text invoice from a small owner-driver courier, the extraction accuracy is consistently above 97%. The remaining 3% get flagged for human review automatically.
Step 3: PO matching
Make.com queries the freight management system's API for open purchase orders. It attempts to match the extracted invoice to an open PO using three criteria: supplier name (fuzzy match, >80% similarity), invoice amount (within 5% of PO value), and date (invoice date within 30 days of PO date). Matches above a confidence threshold are auto-approved; lower-confidence matches are flagged for human review.
Step 4: Cost centre assignment
A lookup table maps each supplier name to their standard cost centre code. Claire built this table with us during the project — about 60 rows for their regular suppliers. For unrecognised suppliers (typically one-off or new suppliers), the system flags for manual cost centre assignment.
Step 5: Xero bill creation
Make.com creates the bill in Xero via the Xero API — populating supplier, invoice number, date, due date, line items, amounts, and cost centre codes. The bill is created in "Awaiting Approval" status for review and payment approval to remain with the finance team. The system doesn't approve or pay bills — it handles the data entry and matching, not the authorisation.
Step 6: Tracker update and filing
The Google Sheets tracker is updated with the invoice status (auto-processed, needs review, or unmatched). The original email is labelled and archived in Gmail. A summary Slack message posts to the operations channel each evening showing the day's invoice processing stats.
Implementation: Scoping to Go-Live in 13 Days
The timeline from first scoping call to a fully operational automation:
- Day 1: Scoping call. Mapped current process. Agreed scope and deliverables. Project confirmed.
- Day 2: First working version deployed. Basic Gmail → Claude extraction → Xero draft bill flow, tested on 10 sample invoices provided by Claire.
- Days 3–6: PO matching logic built and tested. Cost centre lookup table built with Claire's input.
- Days 7–9: Edge case handling — multi-currency invoices (they occasionally receive invoices in EUR and USD), credit notes, and consolidated invoices covering multiple POs.
- Days 10–11: Parallel run — the automation processed real invoices alongside the manual process so Claire could compare outputs and catch any issues.
- Day 12: Minor adjustments based on parallel run findings (three supplier name variants that weren't matching, one edge case in the VAT calculation for a specific supplier).
- Day 13: Handover — full documentation, Make.com scenario walk-through, Xero integration explanation, video recording, and live Q&A with the operations team.
Results: Six Weeks Post Go-Live
Six weeks after go-live, Claire sent us an update:
"We process about 250 invoices a month. Of those, the system handles 240 automatically — 96% hit the Awaiting Approval queue in Xero with everything filled in correctly. The other 10 or so need a quick human check, usually new suppliers or invoices with missing references. Total time spent on invoices by the team is now about 3–4 hours a month instead of 40. The error rate is essentially zero — we haven't had a single duplicated payment or incorrectly coded invoice since we went live."
The financial impact: Meridian's operations team fully-loaded cost is approximately £38,000/year, working 230 days/year, or about £165/day. Forty hours per month represents roughly £3,300 in annual labour cost redirected from data entry to actual operations work. Add the error-correction time that's been eliminated and the real-world saving is closer to £4,000–£5,000/year.
Project cost: £2,497 one-time. Running costs: approximately £45/month (Make.com subscription + Claude API usage). Payback period: approximately 7 weeks.
What Happened Next
Three months after the invoice automation went live, Meridian came back for a second project: automating their customer shipment status notifications. When a shipment moved through certain milestones in their freight management system, clients were being manually emailed by the operations team. A second Make.com automation now handles that automatically, generating personalised status emails from a template and sending them via their business email. That project was a Starter (£997), took 5 days, and eliminated another 8 hours per week of manual communication work.
This pattern is typical. The first automation solves the most painful problem and builds trust. Once the operations team sees it working reliably, they start identifying the next thing — and the next.
Could This Work for Your Business?
Invoice processing automation is one of the highest-ROI automations available to UK SMEs because the problem is universal, the data is structured (even when it doesn't look it), and the AI extraction technology has matured to the point where format variance is handled reliably.
The pattern applies beyond logistics. We've built similar automations for professional services firms (processing contractor invoices), property management companies (utility bills and maintenance invoices), and manufacturing businesses (materials supplier invoices). The core architecture is the same — the specific integrations and rules change.
If your business processes more than 50 supplier invoices per month and more than one person is involved in that process, the economics almost certainly stack up. The question is whether the specific tools in your stack (accounting software, email provider, management system) have the API access needed to make it work. In most cases, they do.
Tools needed for this type of automation: An email provider with API access (Gmail, Outlook 365), an accounting platform with an API (Xero, QuickBooks, Sage, FreeAgent — all have excellent APIs), and optionally a project management or tracking tool for status updates. Total setup time: 1–3 weeks depending on complexity.