Writing better prompts

Prompt detail is the biggest thing you control. The people who get the most out of Scout write a paragraph, not a line.

The quickest way to start is to use a prompt chip above the prompt box. It inserts an editable prompt you can adjust. More ready-made examples are available in the Prompt Library.

Six things to specify

	Specify this	Not this
Company	The ticker	A name that could match several listings
Periods	"last 8 quarters and 3 fiscal years"	"historicals"
Line items	Name them, in the order you want	"the income statement"
Layout	Tabs, rows vs columns, quarters interleaved or grouped	Nothing, so Scout picks
Values or formulas	"hard values", or "calculate X as a formula"	A mix you didn't choose
Boundaries	"don't touch columns H onward", "new tab only"	Nothing, so Scout decides what's safe

Tips for better results

Use the company's own wording — "Family daily active people, or DAP" matches better than "users." Give the abbreviation and the full name.
Separate what to pull from what to calculate — List the line items, then list the margins, growth rates, and percentage-of-revenue lines you want derived.
Rule out inference — Add something like "Do not infer unit volumes or customer counts unless explicitly disclosed." Scout is built to leave blanks rather than guess, and saying so reinforces it.
Ask for commentary as its own row if you want the qualitative read next to the numbers.
Put your formatting conventions in Global instructions — Open the three-dot menu, select Settings, and add them once instead of retyping them every session.
Ask for restated figures if you want them — As-reported is the default.

Keep prompts small

Scout is most likely to stall on long prompts in large financial models. Three focused prompts usually work better than one prompt that tries to do everything:

Build the Daloopa data tab.
Build the statements off it.
Add drivers, ratios and forecast columns.

Fix the plan before Scout builds

For substantial requests, Scout shows a plan before making changes. Review the period range, line items, target tab, and any warnings. If something looks wrong, ask Scout to revise the plan in chat before it starts building.

Fixing a plan takes one sentence. Fixing a finished model can take an afternoon.

When something goes wrong

Symptom	What to do
Scout stopped partway through	Type `continue` or `resume` in the same conversation
Scout stalled without a message	Try the prompt again
The output has the wrong shape	Revert, then rewrite the prompt with the layout specified
Same prompt, different result	Add more detail; vague prompts vary most

Before and after

Thin:

Build me a model for Amazon.

Better:

Build a revenue growth contribution module for AMZN on a new tab. Break out revenue by major business line: Online Stores, Third-Party Seller Services, Advertising, AWS, Subscription Services. For each, last 8 quarters of revenue, YoY growth, and contribution to total company revenue growth, reported figures only. Add a commentary row per segment for demand, pricing or mix. Label the calculated contribution metrics. Don't infer margins, unit volumes or customer counts unless disclosed. Hard values for reported figures, formulas for growth and contribution.

Both prompts may run. The second is much more likely to come back in the shape you wanted.