Getting the most out of Dawn

Dawn is the agent inside TrueState. You can ask Dawn questions, give it datasets and dashboards to inspect, and ask it to create useful analytics work such as charts, commentary, reports, pipelines, jobs, and dashboard updates.

You do not need to be an analytics expert to get value from Dawn. If you have done a basic SQL course, you already have enough mental model to start: data lives in tables, tables have rows and columns, and useful analysis usually means connecting those tables back to a real business question.

The main trick is not learning a perfect prompt. The main trick is helping Dawn understand the business, the data, and the decision you are trying to make.

The simple mental model

Think of Dawn as a capable analyst who has tools, but still needs a brief.

If you say:

Analyse this data.

Dawn has to guess what matters.

If you say:

Analyse the customer subscriptions dataset for the retention team. We want to understand why churn increased last quarter. Focus on customer segment, plan type, tenure, and recent support activity. Give me the 3 most likely drivers and show the query or chart that supports each one.

Dawn has a much better chance of doing useful work.

Good Dawn work usually has four ingredients:

Ingredient	Plain English meaning	Example
Business goal	Why you care	"Reduce churn"
Data	What Dawn should inspect	"subscriptions and support tickets"
Output	What you want back	"3 drivers with charts"
Audience	Who will use it	"retention team"

The TrueState 360 framework

TrueState 360 is a practical way to make Dawn useful across the whole organisation, not just one chat at a time.

It works in a loop:

Understand the business.
Map the business to data.
Define calculations consistently.
Choose valuable use cases.
Build useful assets.
Review what is live and improve the context.

You do not need to complete every part before asking your first question. But the more of the loop you fill in, the less Dawn has to guess.

1. Understand the business

Start with the **Business Model** in **Agent -> Context**.

This is where you explain, in normal language, how the organisation works:

- Who are the customers?
- What does the business sell?
- How does it make money?
- Which channels matter?
- What costs, partners, or activities are important?

This helps Dawn avoid generic advice. A recommendation that makes sense for ecommerce might be useless for a B2B software company or a services business.

2. Map the business to data

Use the **Data Map** to connect business words to data words.

For example, the business might talk about "customers", "accounts", "subscriptions", "orders", or "renewals". The data might store those ideas across tables with names like `crm_accounts`, `stripe_subscriptions`, or `orders_export`.

The Data Map helps Dawn understand which datasets represent which business concepts.

3. Define calculations once

Use the **Feature Catalog** for metrics and calculations that people should not redefine every time.

Examples:

- Monthly recurring revenue
- Gross margin
- Active customer
- Churned customer
- Lead conversion rate

If these definitions are not written down, two dashboards can show different answers for the same metric. Dawn can help you avoid that by reusing the trusted definition.

4. Pick the best use cases

Use **Use Cases** to capture analytics ideas, then prioritise them by impact and feasibility.

A good use case is not "make a dashboard". A good use case explains the decision.

Better examples:

- "Help the sales team decide which leads to call first."
- "Help finance explain why revenue changed this month."
- "Help support identify customers at risk before renewal."
- "Help marketing compare campaign quality, not just lead volume."

5. Build the right asset

Once you know the use case, choose the right output:

- **Conversation** for exploration and follow-up questions
- **Dashboard** for repeated monitoring
- **Report** for a written narrative
- **Pipeline** for repeatable data preparation
- **Model** for prediction or classification
- **Job** for recurring agent work

6. Review and improve

When Dawn produces something useful, decide whether any of the learning should become reusable context.

For example:

- A metric definition belongs in the Feature Catalog.
- A business rule belongs in Conventions.
- A repeatable process belongs in Skills.
- A recurring task belongs in Jobs.

This is how the workspace gets smarter over time.

What to do first

If you are new to analytics, start small. Do not try to map the whole company in one sitting.

A good first session looks like this:

Pick one business question.
Attach the most relevant dataset.
Ask Dawn to inspect the columns and explain what it can and cannot answer.
Ask for a first analysis.
Ask Dawn what context would make the answer better next time.

Example:

I am new to this dataset. It is supposed to contain sales opportunities from our CRM. Please inspect the columns, explain what each important column seems to mean, and tell me what business questions this dataset can answer safely.

Then:

Now use this dataset to help me understand why closed-won revenue changed last quarter. Keep the explanation beginner-friendly and show the key query or chart behind each conclusion.

Beginner-friendly prompt patterns

Ask Dawn to orient you

Use this when you do not know the dataset well.

Please orient me to this dataset. Explain the important columns in plain English, identify possible IDs and date columns, and suggest 5 useful business questions it could answer.

Ask for an analysis with guardrails

Use this when you have a question but want Dawn to avoid overclaiming.

Analyse this dataset for [business goal]. Use plain English. Separate facts from guesses. If the data cannot answer part of the question, say what is missing.

Ask for a dashboard plan

Use this before building a dashboard.

Design a dashboard for [audience] to monitor [business process]. Recommend the key metrics, charts, filters, and commentary. Explain why each item belongs there.

Ask for a report

Use this when the output should be read by other people.

Create a report for [audience] about [topic]. Include an executive summary, key findings, supporting evidence, risks or caveats, and recommended next steps.

Ask for reusable context

Use this after Dawn has learned something important.

Based on this conversation, what should be saved into Context so future analysis is more accurate? Separate business facts, metric definitions, data mapping notes, and conventions.

Common mistakes

Asking for "insights" without a decision

"Find insights" sounds reasonable, but it is too open. Dawn will do better if you name the decision.

Instead of:

Find insights in the marketing data.

Try:

Help the marketing team decide which campaign types deserve more budget next month. Compare lead volume, conversion quality, and cost where the data allows it.

Trusting a chart without checking the definition

If a chart says "revenue", ask what revenue means.

Good follow-up:

What definition of revenue did you use? Which column did it come from? Are refunds, discounts, or one-off fees included?

Treating every result as final

Analytics is usually iterative. A first answer often reveals a better second question.