A 50-person customer's monthly close used to take 28 hours of human work. Today it takes 6. No magic — 12 of 14 steps are now automated, and 2 deliberately are not. Here's what the machine does and what we left in the bookkeeper's hands.
The 14-step checklist
- Receivables aging. All open customer debt bucketed 0–30 / 31–60 / 61–90 / 90+ days. Automated.
- Payables aging. Same on the vendor side. Automated.
- Inventory reconciliation. Physical vs. ERP stock per warehouse, variance report. Automated (the new mobile scanner app feeds the inventory data to the Logistics module).
- Bank statement import. SFTP pull from each bank, MNB daily FX rates. Automated.
- Bank → accounting entry matching. Convention-based matcher (amount + memo + known counterparty). 95% auto-matched, 5% manual. Partially automated.
- FX revaluation. Foreign-currency balances revalued at month-end MNB mid-rate. Automated.
- Accruals/deferrals. Recurring contracts (rent, insurance) auto-deferred. Automated.
- Cost-center allocation. Overheads (utilities, IT) allocated per the configured key. Automated.
- Intercompany elimination. Both sides of intercompany entries auto-matched. Automated.
- VAT computation. VAT totals per code, NAV Online Számla 3.0 error filtering. Automated.
- VAT return prep. VAT-return XML generated, error list produced. Automated, but submission is manual.
- Closing snapshot. Every GL account balance frozen into the
period_close_snapshottable. Automated. - Final journal-entry sign-off. Stops here. See below.
- Tax-position review. Stops here too. See below.
The two deliberately manual steps
13. Final journal-entry sign-off. The bookkeeper reviews the snapshot from step 12 and either signs it off or kicks it back. The ERP offers no automatic approval flow for this. Not disorganization, deliberate. At the close of each month a responsible human pair of eyes is required to declare: "this is okay, we can close." If the machine did it, and three months later someone asked who took responsibility, "the system" is not an acceptable answer.
14. Tax-position review. Each month-end during the year the bookkeeper (or tax advisor) checks whether anything booked in the period crosses the company into a different tax bracket (e.g. KIVA → TAO threshold, or KATA limit). The system shows the data, computes thresholds, but the strategic decision (e.g. "delay this invoice to avoid the crossover") is made by a human. This is also part of the value an accounting firm provides — automating it would push in the wrong direction.
What's required to automate
Not every tenant can start tomorrow. Three preconditions:
- Clean master data. Customers, vendors, products, cost centers consistently coded. If the tenant records the same vendor 17 different ways, no automation will work over the top.
- Bank SFTP access. Most Hungarian banks provide it (OTP, K&H, Erste, Raiffeisen, MKB, Gránit). UniCredit has been hard for a long time. Anyone working from manual PDF statements stays manual here.
- NAV Online Számla 3.0 hookup. Essential for VAT; outgoing and incoming invoices flow in real time.
What we measured
At the same 50-person tenant, the 28 → 6 hour drop is a six-month measurement. Month 1 was still 18 hours, month 3 was 9. Accuracy (auditor reject rate) also improved, from 1.8% to 0.4%. The two intentionally manual steps still take about 2 hours, unchanged — which wasn't the goal.
Close automation isn't an end in itself. The freed-up time gets spent by the tenant's accounting team on tax planning and management reporting, work where human expertise creates more value.