Automate invoice coding and workflow routing with self-learning AI in SAP

It learns from your own invoices only. There is no shared or global model, and no data moves between customers. Every prediction it makes comes from how your organisation has coded invoices in the past.

Typically 6 to 12 months, together with the coding and workflow data that goes with it, so the model can see who normally reviews and approves what. If you have less than that, tell us and we will look at it, but 6 months is where the results start to get good.

You can enter the change in the admin tool and the AI adapts to it immediately. If nobody enters it, Snowfox usually picks up the new cost centre by itself after 3 to 5 invoices.

Yes, and those are the most valuable ones. Every corrected or rebooked invoice goes back in as training material, which is why accuracy keeps climbing after go live rather than flattening out.

Each model is trained from scratch for one customer and is never shared. The extraction can be done by your own team or by your SAP implementation partner, and we provide the documentation for it.

Most customers land between 85 and 95 % within the first six months. Automatic posting is something you switch on per dimension once you are comfortable, using a confidence threshold you set yourself.

In practice, almost never. The supplier specific patterns and the odd exceptions are already in your invoice history, and that is exactly what the AI reads.

Yes. If different parts of the group code invoices in genuinely different ways, we train a separate model for each. Models are never shared across tenants.

We are a separate add-on. Snowfox is an automation layer that sits between your invoice sources and SAP, enriches the invoice, and hands it over.

In SAP, where they already are. We do not touch them. SAP automates what its own rules can handle, and Snowfox takes the invoices that would otherwise end up on somebody's desk. As your rules improve, our share goes down, which is how it should be.

All of them. Peppol, other e-invoice formats, and scanned images that have been through an OCR solution. If it reaches your invoice process today, we can read it.

No changes to core SAP. Snowfox runs as an external layer, which is also why SAP updates are a non event for us.

In SAP, exactly as they do today. Approvals, corrections and final posting all happen in the system they already know. Most people only notice that the invoice arrives already coded.

Three points. Invoices come in to Snowfox from your invoice source or e-invoicing operator, the coded invoice goes from Snowfox to SAP, and the finished coding and approval data comes back from SAP to us so the model keeps learning.

Coding explainability is close to release. Once it is out, you can open a prediction and see what drove it. Today you get a confidence value, which is the AI telling you how sure it is about that particular prediction on a scale of 0 to 100 %.

Yes. Snowfox can consolidate those lines before the invoice goes to SAP, which keeps the posting a lot tidier than the original invoice often is.

Snowfox can either complement your existing precoding or predict the work order and project codes itself. It does well whenever there is something consistent to go on, a delivery address for example.

We are launching an invoice validation service that covers incorrect VAT, missing entity data and the other discrepancies that normally get found far too late in the process.

That stays in SAP as standard. Our focus is the coding and routing of invoices that arrive without a purchase order, and we do not get involved in arrival registration.

Mostly we do not need to. SAP matches and codes those natively and does it well. The manual coding sits with the invoices that arrive without a purchase order, which is where we put our effort.

Not at the moment. It is on the list of things we are looking into, but I would rather say so plainly than promise it.

A fixed monthly fee based on how many purchase invoices you have to code in a year, which in practice means the ones that arrive without a purchase order. There are no per transaction charges, so a busy month does not cost you more.

As a rule of thumb, from 1,000 purchase invoices a month that somebody codes by hand today. Below that the saving is real but modest, and I will say so rather than sell you something that does not pay back.

Both, in the sense that the fee is fixed monthly and the level of it is set by your total annual volume. Nothing varies per transaction.

No. The number of companies in the group makes no difference. Pricing looks only at how many purchase invoices the whole organisation has to code in a year.

A manually coded invoice costs most organisations somewhere between 4 and 8 euros once you count the time it takes in AP and with the business managers approving it. With Snowfox the same invoice costs well under a euro. Take your own annual volume and the comparison takes about a minute.

A few days of working effort on the technical side. In calendar time, including testing and go live, 3 to 6 weeks is the normal range.

Less than you would expect, and this is not an IT project. Somebody has to extract the invoice history and somebody has to be available for the integration, and that is close to the whole list.

No. You update approvers and reviewers in the admin tool and carry on. There is no retraining project waiting for you.

Immediately, if you make the change in the admin tool.

Very little. The configuration data is mirrored from SAP, so there is no second dataset to keep in step. In practice the ongoing work is reading the analytics now and then.