What double data entry costs
A supplier invoice arrives by email, as a PDF or as a photo. Someone opens it, reads it, re-enters its content in the accounting tool, then files it. Multiply by the number of documents in a month: at an importer, that is thousands of supplier invoices every month. This re-entry costs time, and it is where discrepancies are born.
Before automating anything, put a number on the problem: the time lost, the errors that follow, or their equivalent in francs. Without that starting point, there is no way to know, months later, whether the project actually paid off. It is the “Measure” stage of the Devlab method, and it comes before any decision to automate.
None of what you need to gather first is technical: invoices you can actually read (a box of paper gets digitised before AI comes anywhere near it), someone on the business side able to settle borderline cases, and that opening measurement.
How AI reads an invoice
The principle comes down to three steps. Documents are first organised by client file, whatever channel they arrive through: email, PDF or photo. AI then extracts the data line by line, not just the invoice header: that detail is what the work and its review depend on. Finally, the data is filed and exported to accounting tools.
That is what Kura does, the document processing software Devlab publishes for accounting firms and their clients: whatever channel a document arrives through, it joins its client file and comes out with data ready to check. The result: nothing left but to approve it before it hits the books.
Automatic document processing is part of Devlab’s artificial intelligence practice, alongside agents and tailor-made automations. Our SaaS products, built in Tahiti, are all in production.
Review before you book
AI can be confidently wrong. Hence a strict rule: nothing moves into the books until it has been checked. In Kura, the source document and the values to verify sit side by side on screen, and every extracted figure keeps a link back to the document that backs it up. Accountants keep control of discrepancies and approve, instead of re-entering everything.
The split of responsibilities is set task by task: what the tool handles alone, what it hands to a person, what it never touches. That gets decided before the tool goes live, not after the first problem shows up, and every action stays reviewable afterwards.
That trail then lets you trace back a correction, an approval or a note. An automated invoice is an invoice whose every figure stays explainable.
Source document and extracted data, shown side by side.
Discrepancies and approval: the accountant decides, the AI prepares.
What runs on its own, what waits for sign-off, what stays off-limits: the rule changes by task.
History: corrections and approvals stay available for review.
Connecting accounting and the ERP
An invoice that is read but then re-entered in the accounting tool has automated nothing. The last step is therefore filing and exporting to accounting tools, or directly to the ERP when there is one. Plugged into an ERP, the agent touches sales, stock and invoices directly; without one, work starts with reading documents alone, and the rest connects later.
Devlab connects its agents to your management systems, whatever your ERP, Odoo, Sage or another used in Polynesia, and accounts for the local reality: Pacific francs, local payroll and accounting. Devlab also publishes Vaka, an ERP developed in Tahiti that brings together quotes, orders, invoices, inventory, banking and accounting.
Once the ERP is in place and connected, automatic reading of supplier invoices becomes one building block among others: file preparation, answers to incoming requests, without double entry.
Beyond invoices: purchase orders, customs files
The same logic, read a document, prepare the data, have a human review it, then export, applies to other documents that circulate in a Polynesian company.
Purchase orders: on paper or by email, a purchase request tends to get lost along the way, and tracking spend becomes hard. Awa, published by Devlab, gives these orders a path again: approval levels by role, allocation by department and a history for each order.
Customs files: an import shipment means many lines to re-key by hand. Ara turns supplier invoices into customs data to review and suggests a tariff classification; the import team checks and corrects before anything is sent on.
Where to start with Devlab
A free first call, 30 minutes, is where we pin down the real pain point and check that automating actually changes something. Bring the documents concerned, the people who touch them along the way, your current tools and the step that eats your time: the demo matches your situation, not a feature catalogue.
The exact scope is then set before anything is deployed: history migration, permissions per person, configuration, staff training, links to the tools already in place. Go-live moves forward step by step, on real invoice flows, tracking results as it goes; where the data travels is documented from day one.
Keeping pace with AI without committing a large sum upfront is the appeal of the subscription option, the one we recommend most often; buying the solution outright stays available for those who would rather own it, at a considerably heavier upfront cost. And if AI is not the right answer to your problem, we say so plainly.
A representative month of invoices: emails, PDFs, photos.
The time currently spent entering them, even as an estimate.
The destination accounting tool or ERP, and who administers it.
The person who will settle ambiguous cases.