First Trusted Answer

This guide walks through the first complete workflow that demonstrates how Tukun.ai should be used in practice.

The goal is not to get any answer. The goal is to get one answer your team would actually trust enough to discuss.

The example workflow

Use a question like:

How did weekly active accounts change over the last 12 weeks by plan tier?

This is a good first workflow because it has:

• a clear metric
• a clear time window
• a clear segmentation dimension
• a result that most teams can validate against existing intuition

Step 1: confirm the source and account context

Before running the question, make sure:

• you are in the correct account
• the intended data source is selected
• the data source has enough coverage for the metric you are asking about

If the source is wrong, the rest of the workflow is wasted effort.

Step 2: write the first question narrowly

Start narrow. Do not add comparisons, causality claims, and business recommendations in the first turn.

Good:

Show weekly active accounts for the last 12 weeks by plan tier.

Too broad for a first turn:

Why is retention down, what segments are driving it, and what should we do next quarter?

Step 3: inspect the first result

When the first result comes back, ask:

• Did Tukun interpret “active accounts” the way we mean it?
• Is the weekly grain correct?
• Are the plan tiers the expected categories?
• Is the trend plausible given what we already know?

If any of these answers is “not sure,” do not move on yet.

Step 4: refine the question

Use follow-ups to fix scope before adding judgment.

Examples:

• “Exclude internal and test accounts.”
• “Compare the last 4 weeks against the prior 4 weeks.”
• “Break down paid plans separately from free.”

The sequence matters. First fix scope. Then add comparison. Then ask for interpretation.

Step 5: decide whether the evidence is strong enough

At this point, the result may still be unsuitable for a business conclusion. Common reasons:

• the metric definition is unclear
• exclusions are not standardized
• the source tables are incomplete
• a business term maps to multiple candidate fields

If that happens, the correct next step is usually semantic modeling, not more prompting.

Step 6: promote the result if it is stable

Once the answer is conceptually right:

• save it as a card if it is a recurring check
• add it to a dashboard if it belongs in a recurring review
• create or refine the semantic definition if the question will repeat across the team

What success looks like

A successful first trusted answer gives you:

• one reviewed result
• one short list of unresolved business definitions
• one decision about what should be saved
• one concrete next step for semantic cleanup or dashboard reuse

That is the real product value: a controlled path from question to reusable analysis.