Prompting best practices

These are the habits that make the biggest difference when working with K-AI. Each one is small on its own, but together they turn vague questions into reliable answers.

Be specific about what you mean

Name the asset, process, or relation you're asking about. K-AI works with a whole workspace, so "this" or "the data" forces it to guess.

Weak: "Why is this slow?"

Strong: "Why does the order process spend so long between Approve and Ship?"

The strong version points K-AI straight at the right part of your model.

Lead with your goal

Tell K-AI what you're trying to achieve, not just what to fetch. The same data can be shown many ways, and your goal tells K-AI which one is useful.

Weak: "Show me orders."

Strong: "Show me orders by status so I can see where they get stuck."

Set the scope

Bound the question with a time range, a segment, or a single process. Without a scope, K-AI answers across everything, which is rarely what you want.

Weak: "What's my average cycle time?"

Strong: "What's the average cycle time for completed orders in the last quarter?"

Ask for the form you want

If you have an output in mind, say so: a single number, a ranked list, a chart, or a step-by-step explanation.

Weak: "Tell me about bottlenecks."

Strong: "Give me a ranked list of the five slowest steps in the order process."

Ask one thing at a time

A focused question gets a focused answer. Stacking several asks into one request makes it harder for K-AI to do any of them well.

Weak: "Check my model and fix whatever is wrong."

Strong: "List the data-quality issues in the Customer object. We'll go through them one at a time."

Refine instead of restarting

You don't have to get it right in one go. K-AI remembers the conversation, so build on its last answer with a short follow-up instead of rewriting your question. After it lists your slowest steps, for example, you can just add "now only completed orders" or "break that down by region."

Do's and don'ts

A quick recap to scan before you ask:

Do	Don't
Name the assets, process, or relation you mean	Say "this", "it", or "the data" and hope K-AI guesses
Say what you're trying to achieve	Leave your goal unstated
Set a time range, segment, or process	Ask open-ended questions across everything
Ask for the form you want (number, list, etc)	Leave the output to chance
Ask one focused question at a time	Stack several questions into one request
Refine with follow-ups in the same chat	Start over when the last answer was close
Say what looks wrong and where	Say "just check it and fix it"

Want examples tailored to what you're doing? See prompting by category for modeling, analysis, and debugging.