SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsdata-storytelling

data-storytelling

Data & analytics

View repository

Updated 10-15 21:03

About this skill

Data Storytelling — Data→Story Conversion

Phase 3: Extract insights from data and compose them as a "story" that resonates with the audience.

Use This Skill When

  • Convert data, numbers, and KPIs into a story for presentations
  • Need a narrative that explains the "meaning" of graphs and charts
  • Convey the impact of change through Before / After comparisons
  • Visualize ROI, market positioning, and technology adoption curves
  • Translate research metrics into slides

Data Storytelling Framework（CIIA）

All data insights are composed of the following 4-layer structure:

1. Context (Situation Description)

  • Why this data matters
  • Background and preconditions the audience should know
  • Clarification of the comparison baseline

2. Insight (Discovery)

  • Express the main finding shown by the data in one sentence
  • State statistical significance and confidence intervals if available
  • Put into words the "meaning behind the numbers"

3. Impact (Influence)

  • How this finding affects business/research/policy
  • Quantitative degree of impact (cost savings, efficiency improvement rate, etc.)
  • Specific impact on stakeholders

4. Action (Action)

  • What should be done in response to this finding
  • Specific Next Steps (with priorities)
  • Identification of additional information needed for decisions

Visualization Selection Guide

Purpose Recommended Chart Usage Conditions
Comparison Bar chart / Radar chart Side-by-side comparison of 2 or more items
Trend Line chart / Area chart Time-series data (3 or more points)
Composition Pie chart / Treemap Proportion of the whole (6 or fewer categories)
Relationship Scatter plot / Network diagram Relationship between two variables
Flow Sankey diagram / Flowchart Flow of process/transformation
Distribution Histogram / Box plot Data spread and outliers

Default: If in doubt, use a bar chart (it is the most intuitively understood).

Guide by Slide Pattern

KPI Dashboard slide

  • Arrange 3–5 KPIs side by side (card format)
  • For each KPI: current value / period-over-period change / target value / achievement rate
  • Convey direction instantly with trend arrows (↑↓→)
  • Colors: achieved=green, not achieved=red, maintained=gray

ROI Visualization

  • Overlay investment amount and recovery amount in a time series
  • State the break-even point explicitly
  • Show long-term effects with a cumulative ROI curve

Market Positioning Map

  • Two-axis matrix (e.g., quality × price)
  • Show own position in a highlighted color
  • Place competitors in gray
  • Indicate direction of movement with arrows

Technology Adoption Curve

  • Innovators → Early Adopters → Majority curve
  • State the current position explicitly
  • Annotate the position of the chasm (gap) and how to cross it

Research Output Metrics

  • Time-series trends of paper count, citation count, and h-index
  • Breakdown by field (treemap)
  • Collaborative research network diagram
  • Impact factor distribution

Composition rules for Before / After comparison

  1. Same scale: Always compare Before and After on the same scale
  2. Explicit change amount: Include not only absolute values but also the rate of change (%)
  3. Specify the period: Specify from when to when the change occurred
  4. Annotate causality: Briefly annotate the cause of the change (e.g., initiative name)

Workflow

  1. Data receipt and organization

    • Check user-provided data
    • Determine data type (time series/categorical/numeric)
    • Report missing values and outliers

  2. Insight extraction

    • Analyze each data point with the CIIA framework
    • Narrow down to 3 or fewer key insights
    • Express the "So What?" in one sentence

  3. Visualization design

    • Determine chart type according to the Visualization selection guide
    • Provide code examples for Chart.js / Mermaid
    • Color scheme follows brand colors (use a neutral palette if unknown)

  4. Narrative composition

    • Structure the slide order as "problem → discovery → impact → proposal"
    • Include supplementary data and sources in the speaker notes
    • If there is a Before / After, follow the comparison composition rules

  5. Output: data-insights.md generate

Deliverables

  • data-insights.md — CIIA-structured insight list + Visualization instructions + narrative composition

Quality Gates

  • All insights follow the CIIA 4-layer structure
  • Data sources are specified for all data points
  • Visualization types are consistent with the selection guide
  • Before / After comparisons are composed on the same scale
  • The one-message-per-slide principle is followed
  • Speaker notes include supplementary data
  • KPIs are narrowed down to 5 or fewer
  • Causality and correlation are distinguished

Gotchas

  1. No cherry-picking — Do not select only convenient data. Include unfavorable data and present a fair overall picture. If omitting, annotate the "reason for omission"
  2. Confusing causality with correlation — Do not present correlation as causation. "When A increases, B also increases" is correlation and differs from "A causes B." Claiming causation requires an explanation of the mechanism.
  3. Scale manipulation — Do not exaggerate changes by setting the Y-axis starting point to something other than 0. If using a logarithmic scale, state that explicitly.
  4. Data sources required — Clearly state the source (source name, acquisition date, URL, etc.) for all data. Do not use data of unknown origin.
  5. Pie chart misuse — Do not use pie charts when there are 7 or more categories. Switch to a bar chart or treemap.

Validation Loop

  1. data-insights.md After generating, verify the following:
    • Whether the four layers of the CIIA framework are present in each insight
    • Whether the Visualization selection follows the guide
    • Whether sources are stated for all data
    • Whether there is any scale manipulation (Y-axis starting point, etc.)
  2. If any item fails → fix the relevant insight and re-verify
  3. After all items pass, present to the user and obtain approval ⏸️

Failure Recovery

Failure pattern Recovery procedure
Insights cannot be extracted due to insufficient data Request additional data from the user. At minimum, confirm "what value changed from when to when"
Some data has an unknown source Mark the relevant data as "source unconfirmed." Ask the user to confirm the source. Do not include it in the slides until confirmation is obtained
Cannot determine an appropriate Visualization Create a provisional version with the default (bar chart), present alternatives to the user, and ask them to choose
The Before / After baseline is unclear Confirm the comparison start point with the user. If unclear, use "the oldest available data" as a provisional baseline

Memory Persistence

Add learnings gained from running this skill to the Gotchas section in the following format:

  • Newly discovered pitfalls in data representation
  • User-specific preferences for brand colors and chart settings
  • Domain-specific KPI definitions and thresholds

Prohibited actions

  1. Fabricating data is prohibited — Do not create data points that do not exist
  2. Manipulating impressions is prohibited — Do not intentionally mislead through scale, color, or placement
  3. Using data without a source is prohibited — Do not include data whose source cannot be confirmed in the slides
  4. Posting confidential data without permission is prohibited — Do not include internal company data or customer data without the user's explicit permission
  5. Copyright infringement is prohibited — Do not reproduce third-party graphs or charts without confirming the license