Skip to content
Data & SpreadsheetsAdvanced

Find Insights in a Dataset (Without Overclaiming)

Get trends, anomalies, and next questions from a table of data, with calculations shown and correlation kept separate from causation.

Share on X (opens in a new tab)Share on LinkedIn (opens in a new tab)

Make it yours

Fill in each variable and watch the prompt update. Anything still in {{braces}} needs your input before you use it.

Fill in the blanks

0 of 4 filled

Who will read the findings.

Where the data comes from and anything unusual about the period.

What you'll decide based on this.

The table (CSV or pasted rows). Keep it small enough to read reliably.

Live previewUnfilled
You are a data analyst preparing findings for {{audience}}.

Context about this data: {{context}}
The decision this analysis should inform: {{decision}}

Data:
"""
{{data}}
"""

Work through this in order:
1. **Data check** — describe what the data contains (rows, columns, time span) and any quality issues or gaps that limit what we can conclude.
2. **Key findings** — the 3–5 most important patterns, trends, or anomalies. For each, show the calculation or numbers it's based on so I can verify it.
3. **Possible explanations** — for each finding, 1–2 hypotheses. Label these clearly as hypotheses; do not present correlation as causation.
4. **What I can't tell from this data** — questions it can't answer and what additional data would help.
5. **Recommended next step** — one concrete action or analysis, tied to the decision above.
6. **Chart suggestion** — the single best chart type to show the main finding, and what goes on each axis.

Double-check your arithmetic. If the dataset is too large to analyze reliably by reading it, say so and suggest using a spreadsheet or code for the calculations.
Open in Playground

Example

Example input

Filled prompt
You are a data analyst preparing findings for the owner of a small online plant shop.

Context about this data: Monthly website and sales data for the first half of 2026. We ran a 20%-off promotion in April.
The decision this analysis should inform: whether to repeat the promotion in the autumn

Data:
"""
Month,Visitors,Orders,Revenue
Jan,4200,126,5040
Feb,4000,124,4960
Mar,4600,147,5880
Apr,6900,276,8832
May,4800,139,5560
Jun,4500,131,5240
"""

Work through this in order:
1. **Data check** — describe what the data contains (rows, columns, time span) and any quality issues or gaps that limit what we can conclude.
2. **Key findings** — the 3–5 most important patterns, trends, or anomalies. For each, show the calculation or numbers it's based on so I can verify it.
3. **Possible explanations** — for each finding, 1–2 hypotheses. Label these clearly as hypotheses; do not present correlation as causation.
4. **What I can't tell from this data** — questions it can't answer and what additional data would help.
5. **Recommended next step** — one concrete action or analysis, tied to the decision above.
6. **Chart suggestion** — the single best chart type to show the main finding, and what goes on each axis.

Double-check your arithmetic. If the dataset is too large to analyze reliably by reading it, say so and suggest using a spreadsheet or code for the calculations.

Example output

Illustrative example response — written by Promptly, not generated live

1. Data check 6 months (Jan–Jun 2026), 4 columns, no missing values. Only monthly totals, so we can't see new vs. returning customers or profit margins.

2. Key findings

  • April spike: 276 orders vs. an average of 133.4 in the other 5 months ((126+124+147+139+131) ÷ 5 = 667 ÷ 5) → about 2.1× normal.
  • Conversion rose in April: 276 ÷ 6,900 = 4.0% vs. 3.0% in January (126 ÷ 4,200).
  • Average order value fell: April 8,832 ÷ 276 = $32.00 vs. $40.00 in most other months — consistent with 20% off.
  • No lasting lift: May orders (139) returned close to March (147).

3. Possible explanations (hypotheses)

  • The discount drove extra visits and purchases (timing fits, but seasonality in spring plant buying could also contribute).
  • May's dip may reflect customers who bought early in April.

4. What I can't tell Profit after the discount, whether April buyers were new customers, and whether they returned. Order-level data and costs would answer this.

5. Next step Calculate April's gross profit vs. a typical month before deciding; revenue alone can hide a loss.

6. Chart Combo chart: months on the x-axis, orders as bars, conversion rate as a line on a second axis.

Why it works

Requiring the model to show its calculations makes errors visible — language models can make arithmetic slips, so verify numbers before acting on them. Separating findings from hypotheses prevents the classic correlation-equals-causation trap, and anchoring everything to a specific decision keeps the analysis focused on what's actionable.

Techniques used