Learning Model
The memory that grows with every model
The second department models the same goods receipt. The same role, the same system, the same transaction – and nobody knows it all exists already. In most tools every model starts from zero.
The problem
Double work is not carelessness
It happens because nobody can see what is already there. Whoever names a role does not know the spelling used next door. Whoever enters a system does not know what it is called three models further on.
“Warehouse manager” or “Manager, warehouse”?
Two spellings, two entries – and every report counts them separately.
The same transaction, captured three times
With different suffixes, different meanings, no connection.
What TAOM does differently
Every model pays into the same memory
On saving, every detail of a process step goes into a company-wide directory: roles, systems, transactions, inspection duties, risks, identifiers. Not as a copy – as evidence of everywhere this term appears.
Four models, one memory – and the path back down is the real benefit.
The evidence
Not claimed – countable
The memory is not a statement of intent but a directory you can query. Every number in it comes from saved models.
These numbers come from our public demo environment – a handful of example processes. In a company it looks different: every project adds further models, and the memory becomes more precise every year. The number of models is not limited; what you see here is evidence of the method – not its scale.
Read from our demo environment
Not maintained but counted: the library grows out of the saved models – roles, activities, decisions and areas with their frequency. The style is measured too, not prescribed: vertical lanes, flow from top to bottom.
Not every model counts. Diagrams produced raw by the agent or created anonymously stay out, as do test models that are too small. Otherwise the memory would learn from practice runs instead of maintained processes – the number is therefore smaller than the total, and dependable for it.
Where is this used?
A role, a system, a transaction – searched across every model in the company, not in the file you have open.
Where is something missing?
Inspection required without a test case, risk without a control, AI-generated without approval – thirteen ready-made questions.
And the twin?
Memory and twin are two things
Das Learning Model is the knowledge base: which terms exist, where they appear, in which meaning. The Digital Twin is the picture of operations: which processes run, who is accountable, which systems carry them, where they stand.
One feeds the other. Without the memory the twin would remain a collection of single pictures – each correct on its own, none connected to the next.
Honestly
What works today – and what does not yet
The directory finds reliably: across every model, every attribute, with free text and thirteen questions. The library stands and the agent uses it when laying out new models. How large it becomes depends on the inventory: in our demo environment it is 17 curated models – in a company with ten years of process work, many times that.
Two searches, two jobs. Across every model the search runs on attributes – roles, systems, transactions, inspection states. Inside an open model the editor and viewer additionally search the body text of the descriptions. One finds where something occurs; the other, what it says there.
In the TAOM framework: TOOL-11 – the method behind this capability.