Product graph

The product graph is a map of what you sell: your products, their modules and features, the systems they integrate with, the people who buy and use them, and how those things relate. Appricity uses it to find the right knowledge when a buyer names something specific.

What can the graph contain?

Seven kinds of entity:

• Product: a top-level product or product line.
• Module: a major capability area within a product.
• Feature: a specific feature inside a module.
• Screen: a screen in your product's interface.
• Integration: a third-party system you connect to.
• Persona: a buyer or user role, such as a finance director or a field engineer.
• Workflow: an end-to-end process your product supports.

And seven kinds of relationship between them: contains, integrates with, replaces, requires, used by, addresses and depends on.

How does it improve answers?

When a question mentions an entity by name (say, "how does the billing module work with Stripe?"), Appricity recognises Billing and Stripe, includes their direct neighbours in the graph, and ranks passages tagged with those entities higher. The buyer gets an answer built from the passages that are actually about the thing they asked about.

How do I build it?

In Knowledge → Product graph:

1. + Entity: choose a type, give it a name (the one buyers use) and a short description. A short identifier is filled in for you.
2. + Relationship: connect two entities, for example Billing integrates with Stripe.
3. Switch between the List and Graph views to check the structure. The graph view is good for spotting entities with no connections.

Opening an entity shows its incoming and outgoing relationships and how many passages are tagged with it. Deactivate removes an entity from use without deleting its history.

How do passages get tagged?

Two ways:

• By hand: open a source, select passages, and click an entity to tag them all at once.
• By suggestion: when you import in bulk with Suggest entity tags switched on, Appricity proposes tags with a confidence score. You Approve, Reject or Approve all on the batch page. Nothing is tagged until you approve it.

One passage can carry several tags.

How big should the graph be?

You don't need to model your whole product. A good first graph covers:

• every module customers ask about by name;
• every integration people ask about;
• the three to five personas that matter for qualification;
• one or two key workflows per module.

That's usually 15 to 30 entities and 20 to 50 relationships. Use the names your buyers use: matching is on the entity's name, so name entities the way buyers say them.

Can entity details differ between product versions?

Yes. With releases in use, an entity can carry version-specific details, and a release's page counts them and flags gaps compared with the previous release.