Data & Context

One record everything works from.

Accounting, sales and service see the same debtor, and an agent reads the same data under the same permissions. What a view shows is not a snapshot: a view is a stored query, and it is recomputed every time it is opened.

PF-04-DE
Queue 284
Sent today 0
Exposure 48.2 M €
Krause Textil GmbH €37,150 2nd dunning
Wagner & Söhne GmbH €4,200 1st dunning
Mustermann AG €12,600 1st dunning
Wagner & Söhne GmbH €4,200 Open
Mustermann AG €1,150 In review
Schmidt Logistics GmbH €780 Resolved

Search or jump …

Views

All documents View · 12,418 documents
Reminders due today View · 63 reminders

Customers

Becker Industries SE Customer 48213 · balance €1,204,000
Wagner & Söhne GmbH Customer 51077 · balance €4,200

Open documents

RE-8832-2001 Becker Industries SE · €445,420 open

Clarifications

Price dispute RE-2044-118 Wagner & Söhne GmbH · €4,200 open

⏎ opens the record

We use cookies to run the site and, with your consent, for embedded content, analytics and marketing. More in our cookie policy and privacy policy.

Ingest

Connected once, checked from then on.

Your ERP or CRM delivers debtors, documents and payments over the REST API, the batch interface or as a file. The import runs in seconds, webhooks keep the record in sync, and integrity is checked around the clock — not first thing when someone opens a report.

How the connection is built is the Integration Layer's subject. This page starts one step later: at the record that comes out of it, and at what everyone involved sees in it.

Source Import Checks One record
Source
ERP, CRM or file — REST API, batch interface or file handover.
Import
Debtors, documents, payments and master data land in the record in seconds.
Checks
Integrity checks run around the clock, automatically, not on request.
One record
From here on every surface and every agent works on the same data.
Objects

Everything hangs off the debtor, and that is more than a dozen things.

It is not only documents that hang off a customer. Limits, securities, risk classes, credit reports, insurance cover, balance-sheet analyses, clarifications, letters and their history all sit in the same record under the same permissions — there is no second place where part of it is maintained separately.

  • Customers

    Master data, balance, reminder level, customer group.

  • Open documents

    Amount, due date, currency, company code.

  • Reminders

    Level, reminder date, open volume, dunning stop.

  • Promises to pay

    Date, amount, whether it was kept.

  • Customer letters

    Occasion, recipient, delivery channel.

  • Clarifications

    Reason, owner, effect on the reminder.

  • Credit limits
  • Limit suggestions
  • Securities
  • Risk classes
  • Credit reports
  • Insurance cover
  • Credit policies
  • Balance-sheet analyses
  • Risk alerts
  • Payments & clearing
  • Deductions
  • Sales orders
  • Addresses & contacts
  • Customer groups
  • Company codes
  • Dunning stops
  • Comments
  • Documents & files

The routes between the objects are part of the explanation: from a customer to its documents, from a document back to its customer. Navigating is never an action — a change of view changes nothing.

Of those, analysable by an agent

Six of these objects an agent can analyse itself — customers, open documents, reminders, promises to pay, customer letters and clarifications. That is the agent's boundary, not the record's: everything else is there in the interface, in the reports and over the API.

Same record, same customer

  • Credit limits
  • Limit suggestions
  • Securities
  • Risk classes
  • Credit reports
  • Insurance cover
  • Credit policies
  • Balance-sheet analyses
  • Risk alerts
  • Payments & clearing
  • Deductions
  • Sales orders
  • Addresses & contacts
  • Customer groups
  • Company codes
  • Dunning stops
  • Comments
  • Documents & files
Views

A view is a query, never a snapshot.

What is stored is the question, not the answer. That is why a view shows the current state the second time it is opened, and why no analysis can work with figures nobody can find again.

Recomputed on opening
Reopening, reloading and "show more" all take the same path. Stale rows cannot occur.
Exact totals, per currency
Totals are counted, not estimated, and reported per currency — a total across currencies would be wrong.
Refining creates a new query
A query never changes after it is created. Refining produces a second one; the first stays readable.
Two inputs

Type, don't search — two places, two questions.

Two input fields that differ on purpose: one asks "where to?", the other "what with what is loaded here?". Both run the paths the mouse already takes; neither is a second execution path.

⌘K

Command palette

12 reminders selected

What to do?
  • Target
  • Selection (12) checked
  • Actions
  • Send reminder
  • Set dunning stop customer, document
  • AI
  • Analyse template optional

⌘K jumps, ⌘⇧K acts

Jumps to views, customers, documents and clarifications, and to everything this session has produced. A leading type word narrows the search ("customer becker"), and picking a record opens it — not a list with one row in it.

/

Slash commands

  • /filter set a filter of the loaded view
  • /sortieren sort by a column
  • /spalten switch the column set
  • /speichern keep the view as a list
  • /senden send the checked reminders
/filter status open

No language model, no cost

Filter, sort, choose columns, send what is checked — straight from the composer, without a language model and without cost. Anything ambiguous stays a question rather than becoming a guess.

Both paths know the permissions of the person acting. Opening a view without the right to see it returns nothing — including when an agent asks.

One record

The same data for accounting, sales and service.

Clarification management, dunning and credit decisions do not work on copies but on the same debtor record. A note from sales is visible in dunning, a clarification holds its reminder back, and nobody has to ask which list is the newer one.

Accounting

Open documents, incoming payments, reminder levels and due dates in the structure you configured.

Sales

Balance, limit and open clarifications on the customer — without a detour through accounting.

Service

Clarifications with their reason, their owner and their effect on the running reminder.

Permissions apply to people and agents alike

Access follows the existing roles and data scopes and is closed when in doubt. An analysis never sees more than the person it runs for, and an agent does not read a customer group that person cannot open.

Aging

Your aging ladder, in your own words.

The aging classes come from your configuration — the same boundaries and the same labels as your due-date report. An analysis adopts them instead of giving itself its own.

With no ladder configured, the aging breakdown stays empty. No 30/60/90 days are assumed and no cut-off is guessed — a class you do not have is not reported.

Example of a configured ladder

  • due today 18%
  • 1–14 days 34%
  • 15–30 days 27%
  • over 30 days 21%
Limits

What explicitly does not happen here.

The shortest test of a data foundation is the list of what it does not do. These points are built this way deliberately, and the next model does not change them.

  • There is no second store beside the record — not even a copy for reporting.
  • No view stores rows. The query is what is stored; the figures are computed on opening.
  • Six object types and no more. What is not a type cannot be analysed by an agent.
  • No figure comes from the language model. Counting, totalling and classifying is the database's job.
  • No totals across currencies. Sums stay separated per currency.
  • No invented aging. Without a configured ladder the breakdown is left out.
  • No read access around the roles. Without the permission the result stays empty.
Questions

What finance leaders ask about this first.

As current as the view you open yourself — it is the same path. The import runs in seconds and webhooks keep the record in sync; a query is computed when it is opened rather than answered from an intermediate state. Where only record ids are given, the query goes straight to the database, because a search index always lags behind a write.

No. Every query checks the acting person's read permissions and returns nothing when in doubt. A person without a company assignment gets an empty result rather than a company-wide one. So an agent cannot surface anything that would be hidden in the familiar interface.

Yes, for a structural reason: totals are aggregated rather than estimated, with an exact hit count instead of an approximation, and reported per currency. A total across currencies would be wrong, so it does not exist. The figures are in the result before a model writes a sentence about them.

Yes. Every query is stored as an immutable descriptor and its reuse is counted; refining creates a second query instead of overwriting the first. So you can read back which question was asked — the answer to it is today's state, not the one from back then.

Because a type is more than a name: it brings filters, columns, totals, permissions and the list of permitted actions with it. A seventh type is therefore development work rather than a setting — and that is exactly what makes the boundary verifiable. New types arrive when they carry weight in the domain, not so that an agent has more to talk about.

You are the ones who set them. An analysis adopts the classes and labels of your due-date report unchanged, so the same figure means the same thing in both tools. Without that configuration the aging breakdown is left out rather than computed against an assumed 30/60/90.

Agentic credit management — at your fingertips.