Run a business modeling workshop

A business modeling workshop turns plain business stories into a structured model you can edit and materialize. You describe how the business works in your own words, one run works out the whole picture, and everything it proposes waits for you in a single review. Nothing is written to your model until you select Apply.

There is one run and one review. The workshop never stops halfway to ask you a question, so once you start it, the next thing you do is review the result.

Start a workshop

There are two entry points:

Open a model from Business > Models, go to its Canvas tab, and select Start a workshop.
In the Ask AI panel, while a model’s detail page is open, choose Start a modeling workshop.

Either one opens the workshop launch form inside the Ask AI panel, with the model canvas alongside it. The workshop lives there for its whole run, so the proposals appear on the model they are headed for while you review them.

Before the workshop begins, choose the model you want to build into. That is the one required field. If you start the workshop from a model’s Canvas tab, the form preselects that model for you.

Below the model picker, a read-only Model context field shows the AI context recorded for the model you chose, so you can see what the workshop reasons from before you start. Discovery keeps its proposals inside that context. If the model has no AI context yet, the form tells you discovery will run unscoped, with a link to open the model and add context before you continue.

The story box sits inside one chat-style composer, alongside the effort choice, a dictation microphone, and file uploads. Add a file from the Add context menu and it appears as a removable chip above the box; the workshop reads it for that one run without storing it. Knowledge base documents have their own field, described below.

Select Start the workshop when your stories read the way you would explain the business to a new colleague.

Set the effort for the run

The Effort picker beside the story box controls how much thinking the run does, which also affects how long it takes and what it costs:

Quick is the fastest and least expensive. It suits a first draft or a small model.
Standard is preselected and fits most models.
Deep takes more care over a complicated set of stories.
Expert does the most thorough analysis, at the highest time and cost.

Effort does not change the shape of the workshop. Every run is one pass whatever you choose, so effort decides how capable the model behind that pass is, not how many steps it takes. You can run another workshop at a different effort later. Which model runs each level, and which level your organization treats as its default, are set under Settings > Artificial Intelligence: see Configure AI settings.

Choose what context the workshop sees

Three context switches sit on the launch form under What should we look at while we work?, all on by default:

This model’s context shares the target model’s own description with the workshop.
The things already on this model shares the name and description of everything already on the target model, so proposals build on what exists instead of duplicating it.
Your other models shares a short digest of your other models (their names and what they hold), so the workshop can recognize when something belongs somewhere else.

Turn a switch off when that context would mislead the run, for example when the model descriptions are stale or when you want proposals judged purely on the stories you provide.

A second block, What should we produce?, carries two more switches: Notes on why we made these choices asks the run to record the reasoning behind its proposals (the explanation described under Read how the workshop read your notes), and Discover attributes asks it to propose each concept’s attributes as well as the concepts and relationships.

Knowledge documents

The “What should we look at while we work?” block ends with a Knowledge documents field. It offers your organization’s documents plus the documents of every project the selected model belongs to, grouped by project. Change the model and the list changes with it: documents from projects the new model does not belong to are removed from your selection automatically.

Pick nothing and the workshop searches your knowledge base for whatever helps. Pick documents and it reads those and only those. If a picked document is too long to read in full, the run says so when it finishes, naming the document.

Watch the run work out the model

Once you start, the workshop reads your stories, the context you allowed, and any documents you picked, then works the model out in a single pass. There is nothing to answer while it runs.

The run appears as a live progress feed: two steps, Discovering the business concepts and then Discovering the relationships and facts, each start with a spinner and complete with a check mark and a short result line.

That one pass produces four things together:

The concepts your business talks about, such as Customer, Sales Order, or Warehouse, each classified into the categories your organization defined under Settings > Conventions, and each judged against the model you chose.
The relationships between them, each carrying the verb your stories used, for example “places” between Customer and Sales Order, and the direction it runs in.
The facts each concept records, including the one the business uses to identify it.
A flag on any concept whose identifier your stories never named, so you know which ones are heading for a placeholder key.

When the pass finishes, the run parks and waits for you. Nothing has been written to your model.

Where your organization has set a naming convention, proposed names follow it.

If your stories name no business concepts at all, the run stops and says so rather than parking on an empty review. Add more detail to the stories and run the workshop again.

Review what the workshop found

Everything the run proposed lands in one review in the panel, under a header counting what is there, such as “Review · 7 concepts · 5 relationships”, with Apply beside it and Discard workshop for abandoning the run. A Review / Chat toggle sits under the header: Review is described here, and Chat under Tweak with AI.

Two lines set the terms. The first states the rule: everything here will be added unless you leave it out, and leaving a thing out also leaves out the relationships that need it. The second counts where you are, such as “6 kept · 1 excluded · 2 blocked”.

Concepts are grouped under their category. Each concept is a card carrying its name, any badges it earned, and a check mark and a cross for keeping it or leaving it out. Kept cards are tinted, so what survives reads at a glance.

Underneath each card sit that concept’s relationships, one narrow row each, written from the card above them:

A relationship between two concepts reads as the verb and the concept on the other end, with an arrow for the direction: ”-> contains Order Line” and ”<- places Customer” both sit under the Sales Order card. The card’s own name is never repeated on its rows.
A relationship joining three or more concepts reads as its name and how many others take part, such as “Sales -> with 3 others”. Select it to open the full editor.
A concept with no relationships is simply a header row on its own.

Everything arrives kept. There is no step that resolves the rows you never touched, so leaving a row alone means it is written.

Five badges are worth knowing. The workshop checks whether a concept already exists anywhere on your branch, not just in the model you are building into:

Out of scope means the workshop judged that the concept belongs to another model. These arrive already left out, with the reasoning behind the call under the card’s Rationale toggle. Keep it anyway if you disagree, and its relationships come back with it.
Already in the model means the target model already has this concept. Apply reuses what is there instead of creating a second copy.
Already connected means the model already relates those two concepts, so the row has nothing left to decide and carries no controls.
Exists in {model name} - will be linked means the concept already lives in another model on your branch. Keep it and Apply links that existing concept into this model instead of creating a second copy. When it lives in several models, the badge names the first and adds how many more.
Exists - will be linked means the concept exists on your branch but sits in no model yet. Keep it and Apply links it into this model rather than creating a duplicate.

If the run missed something your stories named, the field at the top of the review adds it: type the name and select Add concept. It joins the review kept, and you can rename or re-categorize it like anything else.

Leave a concept out and its relationships follow

Leaving a concept out is the one decision that reaches other rows. Exclude a concept and every relationship that needs it greys out immediately, wherever it sits in the review, and says why: “Will not be added while Customer is excluded”. Those rows lose their own controls, because their own decision no longer decides anything.

The same holds for what the concept records. Its facts and business key section says “Nothing will be recorded about Customer while it is excluded” and its checkboxes go quiet, so you are never invited to keep a fact that will not be written.

Nothing is rewritten while you do this. Keep the concept again and every affected row comes back exactly as you left it, with its own decision intact.

The rule you see is the rule Apply follows. A row that reads as blocked will not be written, and a row that reads as kept will be, so the review and the result never disagree.

Open a card for the details

Select a concept card and it expands in place. You can:

rename it,
move it to a different category,
read the identifier note, shown where the workshop could not find a name the business uses to identify this concept, which tells you it will get a placeholder key you can rename later,
work through Facts recorded about this: one checkbox per fact, plus the row proposed as the business key, or a preview of the key the concept will be given when none was proposed,
read Also takes part in, which lists the relationships where this concept takes part but another concept anchors it, each naming the concept on the other end.

Select a relationship row and it expands into that relationship’s editor: the reasoning the workshop recorded for it, who takes part, each participant’s verb and direction, whether it is saved as one combined relationship or as separate relationships for each participant, and a control to remove it outright. A row that attaches at a finer grain shows “attached to {concept}” beside it, for example “attached to Order Line”.

Read the model on the canvas

The canvas beside the panel draws every proposal dashed, right alongside anything that already exists on the model, so a proposal is never mistaken for the real model underneath it. Select a card and its node highlights on the canvas; select a node and its card scrolls into view. A concept you leave out dims rather than disappearing.

While you are reviewing, the canvas is never re-arranged for you. Proposals are placed into free space and nothing you have already arranged moves, so the arrangement you set up survives the review. That is deliberate: the canvas under the proposals is your real model, and a preview has no business moving it while you are still deciding.

Apply is the one exception. The moment the workshop writes what you kept, the canvas lays itself out once, because the new concepts have just arrived with no position of their own and leaving them where they landed would give you a broken diagram rather than an arrangement worth keeping. The layout runs once per workshop, on the model that workshop was building into, so reopening one you have already applied never re-packs a canvas you have tidied since.

Any other time you want the whole picture tidied, select Auto layout on the rail in the bottom left corner of the canvas: it re-arranges and re-fits everything, proposals included.

A relationship joining three or more concepts renders as a junction node labeled with the verb, with a spoke to each participant. A relationship between exactly two concepts shows as a single line carrying the verb, in the preview and after Apply alike, no junction involved. Something that belongs at a finer grain draws straight to the concept it belongs with instead of joining the junction: a Product attaches to the Order Line it was ordered on, not to the Order Header.

Read how the workshop read your notes

Above the first concept card sits a plain-language read of what the workshop proposed:

an overview of what it read in your stories,
the recommended shape for any relationship more involved than a simple pair, including any it proposed as a combined relationship rather than separate ones,
the open assumptions worth verifying before you apply the result.

Above that read, the summary notes anything worth knowing about the run itself, such as a knowledge document the workshop could not read in full.

The explanation is advisory: it helps you judge the proposals, and nothing in it is saved to your model. If you change proposals after it was written, whether in the review, in a card’s details, or through chat, the summary notes that it is out of date and offers a Refresh button to bring it back in line. Below the explanation, an expander lists the discovery steps the run took, each with its result.

Tweak with AI

The Chat tab keeps a chat pane beside the review, so you can discuss and adjust the proposed model in plain language instead of clicking through each card. A draft you have typed survives a switch back to the Review tab.

You can start by just asking questions: what the workshop proposed and why, without changing anything. Once you are ready to make changes, the assistant can also act on your request, including to:

rename or re-categorize a proposed concept,
add or remove a concept or a relationship,
edit a relationship’s verb phrase and its direction,
add or remove a participant, and edit that participant’s role,
switch a relationship between one combined relationship and separate relationships for each pair,
keep a proposal or leave it out.

The canvas and the review update immediately after each change, every change stays reversible (edit a card by hand, or just ask again), and nothing is written to your model until you select Apply.

Typical requests look like:

Drop the Promotion concept from the Sales Order.
Add a Payment Method concept.
Rename Customer to Buyer.
Move Warehouse to the Place category.
Why did you combine these three into one relationship instead of three separate ones?
Split that group into separate relationships instead of one combined relationship.

Apply the result

When the review reads the way you want the model to read, select Apply, available throughout. The workshop writes the concepts, relationships, and facts you kept into the model you chose, on your current feature branch, through the same review path as any other change. Anything blocked by an excluded concept is left out, exactly as the review said it would be.

If anything cannot be applied, for example a relationship that conflicts with one already in the model, the workshop applies everything else and reports the items it skipped, so the result is never silently incomplete.

Once the write completes, you return to the model canvas, focused on the model you just built, so you can see everything new in place. This is the one moment the workshop arranges the canvas for you, laying it out once so the concepts it just added land somewhere you can read them.

Runs started by the older workshop

The workshop used to run in stages, pausing twice to ask which things to carry forward. A run left parked by that older flow cannot carry on: open it and you get a short note saying so instead of a review. Start a new workshop with the same stories to pick up where it left off.

Where to go next

Edit the model on the canvas: see Model business entities.
Turn the model into Data Vault tables: see Model a Data Vault on the canvas.
Set up the categories used in the review: see Set your modeling conventions.