No announcements, no marketing, and nothing written to fill a calendar. This is how NG One works and why it works that way: why dimensional posting has to exist before the first posted line, how an OCR that refuses to guess gets measured, what Serbian regulation demands of a system, and where the international ERP stops. Where a number appears, so does where it came from.
Three pieces about decisions taken before the signature and paid for after it: what in an ERP choice actually decides the next ten years, why projects fail even when the software is sound, and why Excel survives every rollout.
An ERP is not an invoicing program with more options. It is one record of the business from which goods, money, regulation and labour all live at once — and that is how you recognise a real one.
No serious vendor will quote a price before seeing the scope, and that is not evasion. But the structure of the cost can be explained up front — here it is, including the items most often missed.
A module list describes what a system does today; the decade is decided by five choices made on day one that cannot be retrofitted — and that you verify from a terminal, not a slide.
Ordered by theme, and by date within a theme. There is no archive by year: a piece on effective-dated parameters does not expire because May has passed — it expires when the rule changes, and then it gets rewritten.
Moving to a new system does not have to be an event with a date and a knot in the stomach. It can be a rehearsal repeated until it runs clean — and only then does it become the move.
A demo is a performance the vendor prepares. These questions cannot be prepared in advance — which is why they separate systems better than any feature list.
An accounting firm does not measure a system by the depth of one flow but by what the thirtieth client costs. Everything done manually per client multiplies by the number of clients — and that is the whole calculation.
The chaos around transformation is not in the change but in the big-bang cutover, and in handing the work to IT as if it were a server swap. A phased approach spreads the risk instead of piling it onto one weekend.
Sixty domains is a fact about the architecture; fifty-seven equal menu items is a decision — and a poor one.
The decision rarely turns on price, and almost never on technology. It turns on who in the company has to be responsible when something stops at three in the morning.
Analysis by project, cost centre and channel is not a report you add when someone asks for it — it is the shape of the journal line, chosen before the first entry is posted.
Isolation between customers is a solved problem; isolation between platform staff and customer data almost never is — and it is built with the same tool and one test.
A matrix of roles and checkboxes answers one of the six questions ERP authorisation has to settle — the other five usually end up in a spreadsheet, a policy nobody reads, and the head of the chief accountant.
A manufacturer rarely fails to know how much it produced. It almost always fails to know what that cost — and that is not a reporting problem but a question of where consumption is recorded.
In wholesale, margin is rarely lost on price. It is lost on a discount nobody tracks, a rebate calculated later, and a landed cost that never reaches the item.
Every firm that bought an ERP still runs on spreadsheets — not because users are undisciplined, but because Excel fills the exact gaps the system left open.
Everyone who has been somewhere long enough carries the company's map in their head — private, out of date, impossible to query; the Atlas moves that map into the system, with state on every handoff.
The value of a WMS in the core is not a longer feature list but a single stock figure: sales, procurement, the floor and finance read the same number at the same moment, with no nightly syncs and no two truths to reconcile.
A data culture is not a BI tool you buy but a habit — that a number in the system is trusted enough that nobody re-checks it in a spreadsheet. It is built slowly and broken by a single disagreement.
Inventory accounting answers how much you have; a WMS answers where it sits and who touches it next — and those two questions demand different data models.
A goods receipt dated in the past changes the cost of stock that has already been shipped, invoiced and posted — the hardest problem in inventory accounting.
In retail, regulation dictates the architecture. The register must issue a receipt even when the internet is down, and stock must be right even while three locations sell at once.
Invoicing, VAT records, electronic delivery notes and the invoice registry now share one portal and one certificate — yet they remain four services on four clocks, which is why an ERP that merges them into one module breaks on the first rule change.
A VAT rate changes once a decade — which is exactly why most systems store it somewhere that makes it impossible, after a change, to prove what was applied to a document from a closed period.
Payroll and local documents are not features missing from foreign ERPs; they are a boundary data crosses by hand every month, which is precisely why an add-on does not close them.
Almost every ERP claims automatic statement matching. The difference is what it does with a payment that matches no invoice exactly — and those are the majority.
Automation is not switched on but designed: an event, a condition, a named action, and a fuse. When to reach for a deterministic rule, when for an AI proposal, and what to automate first.
Automation is invisible while it works and visible only when it fails — which is why the only honest proof of its value is a counted event no human had to touch.
Every ERP now “has AI” — the real question is whether it sits in the core or is bolted on the side, because that single architectural choice decides what the AI may do, in whose name, and whether it can be proven later.
A copilot that sits next to the document, works inside your permissions and hands back a draft with evidence — four everyday tasks from the warehouse, procurement and finance, and the line it will not cross.
A single accuracy percentage hides the only thing that costs money: which field was missed, and whether the system stayed silent or filled in its best guess.
Autonomy levels, evidence behind every conclusion and golden-set evals in CI — why the line between a proposal and a posted entry belongs in the architecture, not the prompt.
If something here made you go and check how that stands at your company, the next step is an hour about your documents rather than ours.