Generate entity models from a domain and process

Once you’ve described the process and the context, the platform calls Cortex to generate a list of entities. This article explains what generation actually produces, what it’s grounded on, and what to do when the platform’s defaults don’t cover what you need.

Applies to: Domain expert

What generation produces

A generation run produces a list of entities, each tagged with:

A name in business language (Purchase Order Header, Contract Line, Vendor).
A type: Core or Supporting.
A target data product (the platform’s first-pass grouping).
A description of what the entity represents.
A typical field list with data types.

The number of entities varies with the process and the context you provided. A typical Procure-to-Pay run with strong context comes back with 25 to 35 entities. Sparse context produces fewer and more generic ones.

What the platform grounds generation on

The platform doesn’t generate entities from nothing. Cortex is given a structured prompt built from your inputs:

The Industry, Domain, Process, and Functional Area.
The Describe Requirements and List Required Components text.
The Source System and module if you provided them.
The platform’s own knowledge of how the chosen domain and process typically work.

The first source is your context. The last source is what the platform knows about the domain and process from its training. When the domain and process are common (Procure-to-Pay, Order-to-Cash, Hire-to-Retire), the platform’s prior knowledge is strong and a sparse context can still produce reasonable output. When the domain or process is custom, that prior knowledge is thin, and your context does more of the work.

This is why providing context matters more for custom domains and processes.

When the platform’s defaults aren’t enough

Three cases come up.

Your domain or process isn’t on the list. Add it through Settings. The custom entry behaves like a built-in one, but Cortex has less prior knowledge to draw on, so context matters more. See Add a custom domain on the Configuration page and Add a custom process within a domain.

Your process is on the list but uses different terminology. For example, a process called “Procure-to-Pay” but your industry calls it “Source-to-Settle”. You don’t need to add a new process. Use the existing one and name your data products differently after generation. The platform’s process name is a label; the entity output adapts to your context.

You have a library of processes you want to import. You might have an internal catalogue, or be aligning to an industry standard like APQC’s Process Classification Framework. The Settings > Processes page supports a TSV import, so you can load a whole library in one go. See Add a custom process within a domain for the import workflow.

When generation returns nothing useful

If the entity list is empty, suspiciously short, or visibly off-topic, one of three things has usually happened.

Your context was too sparse. Cortex couldn’t ground the output on anything specific. Go back, expand the Describe Requirements text with three or four business questions, and re-run.

Your domain and process don’t match. You picked Procurement and Supply Chain as the domain but a process from a different domain. Confirm both selections match what you actually want.

Your custom domain or process is brand new and undocumented. A custom entry with no description gives Cortex nothing to work with. Expand the description on the Settings page and re-run.