My client receives strange predictions – what affects the quality of predictions?

In FabricAI, there is always an AI in the background predicting at least the account and VAT code, and if necessary, cost centers and invoice date. AI predictions are based on the client's old purchase invoices. However, several factors affect how good the predictions for the client's purchase invoices will be:

• The number of the client's old purchase invoices (in the accounting software currently being used)
  • Basically, the more invoices there are, the better for the success of the AI. A few examples:
    • No invoices yet -> predictions are likely to be quite strange at first
    • Invoices from 1-3 months -> predictions are likely to be at least moderate
    • Invoices from 3-12 months -> predictions are likely to be good
    • Invoices over a year -> the most optimal situation
  • Additionally, please note that the accounting periods of old purchase invoices should be closed if possible! For example, in Procountor, only purchase invoices from closed periods are used as AI training material.
• The quality of the client's old purchase invoices
  • The more purchase invoices come as E-invoices, the better the AI succeeds in its predictions. The AI also makes predictions for scanned and self-added invoices, but they may not be as good as for E-invoices.
• The consistency of the client's old accounting
  • If the client's posting practices or chart of accounts have changed significantly, then the AI training material is contradictory, and the AI predictions are not optimal either. If your client is in such a situation, please contact tuki@fabricai.fi, so we can tailor the client's training material to start only from a certain date.

Did you find it helpful? Yes No