Skip to main content

Prompting by category

The best practices apply to everything you ask K-AI. On top of those, each kind of work has a few things worth spelling out. Here's how to prompt well for the three categories you work in: modeling, analysis, and debugging.

Modeling​

Modeling questions turn your data into a solid model: exploring the sources you've connected and building or refining your assets, relations, and processes. Tell K-AI which sources to work from, what you want to end up with, and the leading object to build around. K-AI proposes the changes for you to review, so the clearer your aim, the closer its first proposal lands.

Weak: "Build my model."

Strong: "Based on the sales tables I connected, suggest the objects, activities, and relations for an order process, with Order as the leading object."

What makes it work:

It points K-AI at the right data and says what to build, so its first proposal lands close.

  • Source to use: the sales tables I connected
  • What to build: objects, activities, and relations for an order process
  • Leading object: Order

Try asking:

  • "Look at the schemas of my connected sources and suggest objects and activities I could create."
  • "Which of my tables share columns that could become relationships?"
  • "Based on my current model, what processes could I define?"

Analysis​

Analysis questions draw insights out of your processes. You can ask in plain business language or in more technical, structural terms, and K-AI matches its answer to how you ask. Either way, name the process you mean, set the scope with a time range or a segment, and say what form you want the answer in. Be clear about whether you're asking about a single case or across all of them.

Weak: "How is my process doing?"

Strong: "For the order process, what's the average time spent in each status this quarter?"

What makes it work:

It names the process, the measure, and the period, so K-AI can answer without guessing.

  • Context: the order process
  • Question: average time spent in each status
  • Scope: this quarter

Try asking:

  • "Which customers are ordering less than they used to?"
  • "What are the most common process variants, and how often does each happen?"
  • "Which steps have the longest average duration in the order process?"
  • "How long did order 1043 take from start to finish?"

Debugging​

Debugging questions help you find and fix what's wrong. K-AI can check your model's health, cross-check it against your source data, and surface issues like missing relations, orphaned records, data-quality problems such as nulls or duplicates, or numbers that don't add up. Describe the symptom and where you saw it, and say what you expected instead, so K-AI can diagnose the cause before suggesting a fix.

Weak: "Something is wrong, just fix it."

Strong: "Some orders have no linked invoice, which shouldn't happen. Check whether the Order to Invoice relation is complete and tell me which orders are affected."

What makes it work:

It describes the problem and where to look, so K-AI can diagnose before changing anything.

  • Symptom: orders with no linked invoice
  • Why it's wrong: that shouldn't happen
  • Where to look: the Order to Invoice relation
  • What you want back: which orders are affected

Try asking:

  • "Are all orders linked to a customer? Show me any that aren't."
  • "Validate this object: are there missing attributes or relationships?"
  • "Check this object's source data for nulls, duplicates, or inconsistencies."
  • "Are there modeling gaps, like unlinked objects or missing relationships?"

Now you know your way around K-AI: what it can do, how it works, and how to ask. Time to put it to work on a real question of your own.