Building a chart
Six questions, a live preview on your own data, and fifteen shapes to draw it with.
Six questions
The chart builder asks six questions. The preview on the right redraws on your own data as you answer them, and nothing is written until you press Save.
What do you want to see?
Pick one metric to follow it over time, or up to six to compare them. Metrics are grouped by data source — Advertising (from your Ads reports), Retail (from Seller Partner), or Combined, which is where measures like TACOS live because they need both sides.
Split it into separate lines?
One line for the whole account, or one per campaign, product, portfolio, placement, search term, and so on. The list of dimensions offered is what your chosen metrics share — pick a metric that only exists per campaign and the ASIN split disappears, with a sentence naming the metric that excluded it.
Over what period?
Date range, bucket size (daily or weekly), and an optional comparison against the preceding period of equal length. Three controls that answer one question, so they sit in one row.
Narrow it down
Optional. Marketplace, brand, portfolio, campaign, ad group, ASIN, SKU, ad product (SP / SB / SD / DSP), match type, placement, and campaign state. Set nothing and the chart reads everything you sell, in every marketplace.
How should it look?
Only the shapes that actually fit what you picked are offered first, recommended order. A shape that does not fit tells you what to change to unlock it rather than drawing something misleading.
Name it
A name is proposed from the configuration. Edit it, add a description if the chart needs one, and save.
The chart types
| Family | Types |
|---|---|
| Over time | Line, area, stacked area, sparkline |
| Comparison | Vertical bar, horizontal bar, grouped bar, stacked bar, combo |
| Composition | Pie, heatmap |
| Relationship | Scatter |
| Single value | KPI, gauge |
| Raw | Data table |
Two things that keep charts honest
Low-signal buckets
A bucket built on very little data (a handful of clicks, say) produces a rate that swings wildly and means nothing. Charts flag those buckets rather than drawing them as fact. In strict mode they come back empty and the line draws a gap; in the default lenient mode they keep their value and are marked as uncertain.
Truncation is visible
A chart draws at most 5,000 rows. When a split has more series than a chart can read, the biggest are drawn and the rest roll into a single Other line, and the chart says so. You never get a picture that quietly stopped early.
- ✓Start from a question, not from a metric — the name field is a good test of whether the chart answers one.
- ✓Use the weekly bucket for anything you plan to look at over a quarter; daily noise hides the trend.
- ✓Set the comparison period when you care about direction, not level.
- ✗Put six metrics on one chart because you can — two on a combo usually reads better than six on a line.
- ✗Read a rate on a low-signal bucket as a real number; that is exactly what the flag is warning you about.
- ✗Mix a weekly-only metric into a daily chart — the builder refuses it, and the refusal is the right answer.