Nona Automates Supplier Invoice Processing with Koncile's OCR

Nona builds management software designed for collective catering: school canteens, medico-social facilities such as nursing homes, central kitchens and local authorities. More than 650 establishments now use it to design menus that meet EGalim sustainability targets, track stock, manage budgets and digitize their food-safety plan (HACCP). Much of that work rests on one invisible building block: a reliable product database able to link every ingredient purchased to its nutritional and regulatory equivalents. That database is built from unglamorous but essential material: the invoices and purchase documents each kitchen receives from its suppliers, which Nona now has read automatically by Koncile's data extraction platform.

Onboarding a kitchen starts with a year's worth of invoices

Before a kitchen can start reconciling its accounts in Nona, it first has to feed everything it already buys into the system. In practice, that means retrieving roughly a year of supplier invoice history to identify the products actually in use (meat, vegetables, dry goods, dairy) and make the software usable from its very first menu. Historically, this was done by hand: every invoice line was re-read and retyped, whether the document was a clean PDF or a handwritten invoice from a small local supplier. A slow, repetitive process, inherently prone to human error, imposed on kitchens whose job is cooking, not data entry.

It's this step, invisible to the cook using the finished software but structurally important for everything else in it, that Nona set out to automate first.

Why a collective-kitchen invoice is unlike any other

What makes this hard isn't the volume of documents, it's how different they are from one another. A collective kitchen doesn't receive one standard invoice type: it receives invoices from food wholesalers, multi-column product catalogs, unit price schedules from public tenders (BPU), and sometimes restaurant menus that need to be re-exported while keeping their column-by-dish layout. Each of these documents has its own layout, its own conventions, and its own way of naming the same product.

A generic food database to feed line by line

To support nutritional calculations and EGalim ratio tracking, every invoice line has to be linked to a generic food item (beef, tomato, tuna, bread...) drawn from a database of roughly 3,300 values at Nona. The catch is that the same food item is named differently depending on the supplier, the packaging or the unit of sale: an invoice line never explicitly says "this is beef." The system therefore has to match a supplier's own label, sometimes cryptic, to the right entry in the generic database, without manual work on every new invoice.

The meat invoice case: when the right code depends on a breed mentioned at the bottom of the page

The clearest example concerns invoices from certain specialized meat suppliers. The product code to extract can't always be read from the line alone: with a given supplier, two different products can share the exact same reference number, distinguished only by the breed of the animal, noted in small print at the bottom of the page. Extracting the right line therefore means cross-referencing several pieces of information scattered across the document, in a reading order that isn't obvious to a standard OCR engine.

What's great about Koncile is that there's AI built in, which lets it adapt to different types of invoices.
^{Timothée Berthault, Supplier Relations & Integration Manager, Nona}

The technical solution built with Koncile

To handle this diversity, Nona doesn't rely on a single extraction model but on several dedicated ones, one per document type: standard invoice, unit price schedule, wholesaler catalog, restaurant menu. Each model applies instructions written together with the Koncile team, down to very specific rules, such as the one that looks up the meat's breed at the bottom of the page to disambiguate a product code. The extracted data comes out structured, ready to use without reformatting, through Koncile's OCR and its data extraction building blocks.

Two other features address Nona's constraints directly. Smart splitting automatically breaks up a PDF batch containing several invoices, based on rules defined with the client (a new invoice date or a new supplier marks the start of a new document). Table match automatically links each invoiced product to a value in Nona's generic food database, based on an exact or close match between the extracted label and the database entries. A move toward a dynamic match, connected continuously to an external database rather than a static import, is currently in progress.

The software processes the invoices on its own and hands us back an Excel file with data that's easy to work with.
^{Timothée Berthault, Supplier Relations & Integration Manager, Nona}

Who does what between Nona and Koncile

Nona remains the sole owner of its business rules: Timothée Berthault's team defines the structure of its generic food database, the splitting rules specific to its invoice batches, and the expected export format for each document type. Koncile, on its side, configures and evolves the extraction models and their prompts, tunes the reading engine (particularly for low-resolution scanned documents), and handles one-off requests, such as adjusting a model when a supplier introduces a new invoice layout.

This split of responsibilities was built iteratively: as more kitchens were onboarded, new edge cases were reported to Koncile, which turned them into reusable rules for the next kitchens rather than one-off fixes.

Kitchens up and running faster

The clearest benefit is speed of rollout. Setting up a new account, which used to depend on weeks of manual re-typing, now happens at the scale of an entire kitchen, by dropping its invoice history directly onto the Koncile platform rather than retyping it document by document. Cooks get access faster to software already populated with their everyday products, which in turn speeds up day-to-day adoption of the tool.

We were able to send all of a kitchen's invoices in one go, and quickly see good results. Koncile's adaptability was a real plus.
^{Timothée Berthault, Supplier Relations & Integration Manager, Nona}

What's next: a product match that keeps up with the supplier database

The next step in the collaboration is moving from a static match (the food database is imported at a given point in time) to a dynamic one, connected continuously to Nona's database: every new reference added on Nona's side would immediately become available for automatic matching, with no manual re-import. Other players in collective catering are working on a related challenge: that's the case for Grainz, which automated the processing of its invoices and delivery notes for the collective kitchens it serves.

"What Koncile offers is genuinely great, and we're very happy with it," is how Timothée Berthault sums up this long-running collaboration.

Frequently asked questions