Why purchase invoices are so difficult to handle in wholesale
The core of the problem is variation. Supplier A sends a clean PDF with a fixed table structure. Supplier B emails a scan of a paper receipt. Supplier C delivers an Excel file, supplier D an email with invoice details buried in the body text. Every document has different column names, a different position for the total amount, a different way of stating VAT. Your accountant or purchasing administrator has to recognise all those variations and manually convert them into lines in the ERP. That works fine until things get busy, or until someone calls in sick. Then invoices pile up, payment deadlines are missed, and suppliers start calling.
What are the biggest bottlenecks in purchase invoice processing?
Three bottlenecks come up in virtually every wholesale business. The first is the missing PO reference: a supplier forgets to include the purchase order number, or puts it in a field that does not match what you expect. As a result, the invoice cannot be automatically linked to the order and ends up on a pile for manual follow-up. The second bottleneck is three-way matching: you want to verify whether the invoiced quantity and price match the purchase order and the goods receipt. Manually, that means switching between three screens or three systems. The third bottleneck is the approval workflow: who is authorised to approve which invoice, what is the limit, and how do you make sure the invoice gets back to the accountant in time so the payment does not run late?
What does automating purchase invoice processing concretely solve?
A system like dottle reads the incoming invoice, recognises the relevant fields (supplier, invoice date, invoice number, line items with description, quantity and price, VAT amount, total) and places that structured data directly into your ERP, accounting package, or purchasing module. Matches with open purchase orders are proposed automatically. Discrepancies, a price difference beyond a set margin or a missing PO number, are flagged for human review. The employee no longer sees a blind stack of PDFs, but a list of exceptions that genuinely need attention. Invoices that check out go straight through. Invoices that do not are surfaced. That saves tens of minutes per day and significantly reduces the risk of payment errors.
When does automating purchase invoices not make sense?
Being honest about the limits is at least as important as highlighting the benefits. If you process fifty invoices a month from ten regular suppliers who all reliably include a PO number, setting up an automation project takes more effort than it saves. The breakeven is roughly a few hundred invoices per month, or the point where current processing time is structurally at odds with available capacity. Also, if your source data is structurally poor, think suppliers who send handwritten receipts or use formats that change every month, you need to fix those quality issues first. Automation makes bad data visible faster, but it does not fix it.
How does the initial setup per supplier work?
A common follow-up question is: how much work does it take to configure all supplier formats? That is a fair concern. For each supplier, you need to establish what the document looks like, where the relevant fields are, and what variations exist (some suppliers send credit notes in a different format than regular invoices). That takes time upfront. dottle works without requiring you to manually build templates per sender, but a review step is still needed for new suppliers or significantly deviating formats. That review becomes lighter as volume grows: suppliers who send you twenty invoices a year matter less than the five major suppliers who account for ninety percent of your volume. Start with the largest groups and expand from there.
Human-in-the-loop: why people are never redundant when it comes to exceptions
The model behind dottle is that a person approves, not the software. That sounds like a limitation, but it is a deliberate choice. An invoice that matches the purchase order exactly and shows no discrepancies barely needs to cross the employee's desk. But an invoice with a three percent price difference, an unrecognised cost line, or a supplier who has never sent anything before deserves human attention. Automation exists to remove the dull, repetitive work. Exceptions stay with the human by design, because a system does not know the business context. Did the buyer make a pricing arrangement outside the system? Does the extra freight line make sense because it was an urgent delivery? The employee knows. The system does not.