AI workflow: Plausibility check

MANUAL

time required

Approximately 90 minutes
per incident

AI

time required

∼ 20 min
per incident

AI workflow: Automatically detect plausibilities

Introduction

In accounting, errors rarely go unnoticed in large cases—they arise in mass business: duplicate receipts, incorrect tax codes, unusual account assignments, period errors. Traditional controls are tedious and often performed too late. In invoicing processes, duplicate payments are a well-known problem; 0.1–0.5% of invoices are estimated to be paid twice.

use case

TAXPOINT creates a review queue for anomalies so that humans only have to deal with genuine exceptions:

  • Duplicate detection (similar document numbers, same amount, same supplier, variations in spelling)

  • validity checks

    • Tax amount does not match tax rate

    • Supplier suddenly treated differently (VAT logic/tax code)

    • unusual fluctuations (amount, frequency)

    • Recurring entries on "new" accounts

  • Explanatory notes ("Why flagged?" + reference documents/history)

  • Checklist per client/period: "Top anomalies" instead of 100% review

Why it works better with TAXPOINT DMS

  • Document + posting in the same context: Duplicates and patterns are best recognized when the document pool, history, and filing are neatly aligned.

  • Semantic search & metadata: Duplicates are often "almost identical" – DMS context + AI similarity search is more powerful here than pure rules.

  • Audit trail & versioning: Important for law firms: who checked which anomalies, what was corrected, which documents were used as a basis.

  • BMD/DATEV connection: Anomalies can be sent directly to the workflow as "cases to be clarified" instead of via voice commands or Excel lists.

  • Private AI: Verification logic accesses a lot of sensitive document data—this must run in a secure setup.

procedure

  1. Documents/records end up in TAXPOINT (upload/interface/sync)

  2. AI checks for duplicates and plausibility and creates an exception queue

  3. Accountant works through the queue (confirm, correct, trigger query)

  4. Result is included in posting/revision (BMD/DATEV)

time saving

  • Manual: 30–90 minutes per client/month (random checks, searching for "similar" documents, follow-up research, correction loops) – significantly more for problematic clients.

  • With TAXPOINT: 10–30 minutes of targeted review of the marked exceptions
    Savings: 20–60 minutes per client/month plus less costly rework (because errors become visible earlier).

You might also like this

There are no more contributions