Prompts and workflows

Once your assistant is connected, the fastest way to get value is to borrow prompts that map cleanly onto the nine tools. You don’t name tools in your prompts: describe what you want and the assistant picks the right ones. This page collects prompts that work well, then combines them into multi-step workflows.

Prompts by tool

You want to know	Try asking
What’s published (list_data_runs)	“List my published Neuralift use cases and when each completed.”
The segments in a run (list_segments)	“Show me all segments in the churn-reduction use case with their personas and top segment tactics.”
Recommended actions (list_actions)	“Summarize the recommended actions for this run, ranked by strategic fit, with each one’s expected impact.”
Plans and activation (list_action_plans)	“For the win-back action, what are each plan’s angle, audience filter, and row count?”
Where the upside is (list_segment_tactics)	“Which segment has the highest revenue lift? Rank its segment tactics by expected impact and show the evidence.”
What the data says (list_insights)	“What are the strongest insights across my segments? Include each one’s confidence and supporting evidence.”
Who a segment is (list_personas)	“Describe the persona for segment 4 in plain language I can share with the creative team.”
What data was used (list_prepared_datasets)	“What tables went into this analysis, and how many rows does each have?”
What a field means (list_prepared_dataset_columns)	“Explain what days_since_last_purchase means and show its sample values and stats.”

Two habits make answers noticeably better:

• Name the use case when you have more than one published run (“in the Q3 loyalty use case, …”), so the assistant filters instead of mixing runs.
• Ask for evidence (“…and show the evidence” / “…with confidence scores”), so answers stay grounded in the scores and features Neuralift computed rather than the assistant’s own generalizations.

Workflow: from segment to campaign brief

Turn one segment into a ready-to-review campaign brief by chaining segments → persona → plans:

1. Find the target. “Across my latest use case, which segment has the highest expected-impact segment tactics for revenue?”
2. Understand who they are. “Describe that segment’s persona and its defining features.”
3. Get the playbook. “What actions and action plans target this segment? Include each plan’s angle and audience row count.”
4. Draft the deliverable. “Draft a campaign brief for this segment using its persona and the plans that target it: audience summary, key message, the angle each plan takes, and how we’ll measure success.”

The result is a brief where every section traces back to published results: the audience description comes from the persona, the message and mechanics from the plans’ angles, and the measurement plan from the KPIs the segment tactics target. Pair it with the activation filters from step 3 when you deliver results to your own warehouse.

Workflow: data dictionary Q&A

When someone asks “what does this field actually mean?”, the answer is in the prepared datasets:

1. “What tables and columns went into the churn use case?”
2. “Explain tenure_bucket: what type is it, what are its sample values, and how is it distributed?”
3. “Which columns most distinguish segment 2 from the rest?”

This works well in review meetings. The assistant reads the same data dictionary you see in the app, so definitions, sample values, and statistics stay consistent between the app and the conversation.

Workflow: comparing a use case with its clone

A clone is a separate use case rerun on a fresh data snapshot. When both the original and its clone are published (for example, before and after a quarter of campaigns):

1. “List my published use cases with their goals and completion dates.”
2. “Compare the segments in the March use case and its June clone: which are similar, which are new, and how did segment sizes shift?”
3. “Compare the top segment tactics across the two use cases. Which tactics from March no longer appear in June, and what new ones emerged?”
4. “Summarize what changed in one page for a leadership update.”

Note. Assistants can still make mistakes when summarizing. Grounded answers make errors rarer and easier to catch. When a number matters, ask where it came from and check it against the app before acting on it.

Next steps

• Tools reference: exactly what each tool returns
• Segment detail: the same content inside the app
• Delivering results: activating segments in your own systems