Story Brief To AI-Ready Narrative Plan

A stakeholder asks for a five slide readout by Friday morning. What they actually need is not slides. They need a decision supported by evidence, scoped to a time window, and framed for a specific audience. Your transferable competency in this course is translating that messy request into a narrative plan that an AI can help draft without inventing business meaning you never validated.

AI is useful here because it can generate candidate structures, draft alternative story arcs, and propose chart menus fast. AI is dangerous here because it will also fill in missing definitions, infer causal reasons, and pick visually persuasive defaults. You own the consequences if a chart misleads or a claim is unsupported. The curriculum builds from brief to plan, from plan to honest visuals, from visuals to narrative, and from narrative to a decision-ready artifact.

Anchor the story in a decision

Start by forcing one sentence of clarity. Who decides, what they will decide, and by when. The business question is not what happened. It is what choice the audience is making using your story.

Define success criteria in operational terms, not vibe. A key performance indicator is a metric with an agreed definition, grain, and refresh cadence. If the KPI is net revenue retention, you need to state whether it is logo or dollar based, monthly or quarterly, and what counts as expansion.

Before you use AI at all, you should be able to point to the acceptance criteria. If the VP says this readout is successful when it explains why Q2 forecast missed, then your plan must include evidence that can support a why claim. If you only have a 12 week time series of bookings, you do not have evidence for seasonality. You have a short run fluctuation.

Take a look at a one slide brief and notice what must be explicit for the story to be defensible.

Build an AI intake package that prevents invented meaning

AI can summarize what you provide. It will invent what you omit. Your intake package is the boundary between your data and the narrative the model will try to write.

At minimum, you provide dataset context. For example, a crm_opportunities export with 48,200 rows from 2025-01-01 to 2025-06-15, one row per opportunity, segmented by region and sales_stage. You also provide metric definitions. If churn is being discussed, you specify whether churn is customer count churn or revenue churn, whether it is gross or net, and what the denominator is.

Caveats are not optional. If finance actuals lag by 10 days, the model cannot be allowed to describe the latest week as final. If pipeline includes renewals and new business, the model cannot assume pipeline equals growth.

Use this checklist style to package the inputs you will hand the model.

Generate a narrative plan with a prompt pattern

You are not prompting for a story. You are prompting for a plan that you can audit. The plan should include the objective, audience, key claims, required charts, and explicit evidence links to fields and tables.

A good plan forces the model to stay inside the data you gave it. You name allowable sources like crm_opportunities, finance_actuals, and product_usage_daily. You require every claim to cite columns like amount, close_date, stage_changed_at, and arr_actual. You also require the model to propose at least one alternative explanation for each key movement. The model can generate candidates. You decide which are plausible given the business context.

Draft a reusable prompt that produces a narrative plan rather than prose.