Waterfall + flow
my pickA gross-to-recognised waterfall over a Sankey that shows seven source systems converging on the reconciliation engine and splitting into reconciled, contested and quarantined, with band thickness proportional to dollars.
This is the single strongest answer to “make the data easy to take in”. The waterfall answers where did the money go and the Sankey answers where did it come from and what happened to it — together they explain the entire product without a sentence of narration. The receivables ageing histogram and five location sparklines with target lines fill the lower band.
Watch — A Sankey is real work to build well, and it is only honest if band widths are truly proportional. Done casually it becomes decoration.
gpt_image_2 — Proportional bands, labelled decrements, a proper waterfall.
seedream_v5_pro — Looser — the waterfall became floating boxes joined by arrows, and the labels garbled (“$00M”, “Alwallances”).Source-by-check matrix
my pickA grid of seven feeds down the side against the data checks across the top, every cell a pass, warning or fail square — plus a location-by-month revenue heat matrix with the figures printed in the cells.
The fastest possible read of pipeline health: one glance finds the failing feed and the failing check at their intersection. It is also the most honest picture of what Concord actually does — thirty-four checks against seven feeds — and it degrades gracefully to a table.
Watch — Colour-only encoding fails colour-blind viewers. The build needs a shape or letter in each cell as well as a hue.
gpt_image_2 — Legible matrix, failed-check counts, a proper sequential ramp on the revenue grid.
seedream_v5_pro — Handsome, but the heatmap ramp does not track its own numbers — $95,368 is shaded lighter than $21,571.Executive scorecard
strongBullet charts with target lines and qualitative bands per clinic, a donut of the exception mix, a ninety-day calendar heat grid of pipeline runs, and an ageing histogram of open exceptions.
The calendar heat grid is the quiet winner — ninety days of unattended operation shown as ninety shaded cells makes the uptime claim visible instead of asserted. Bullet charts beat plain bars because the target is drawn rather than implied.
Watch — Donuts are weak at comparison and this one carries only three slices; a stacked bar would say the same thing more precisely.
gpt_image_2 — Dense, footnoted, with definitions and a legend. Reads as a real recurring report.
seedream_v5_pro — More elegant typography, weaker rigour — the ageing axis runs 10, 20, 30, 30, 400, 500, 500.Process flow
strongA plain labelled diagram of the pipeline across the top — seven feeds into ingest, model, check, publish, with record counts printed on each connector and a warning triangle on the stage that needs attention — over a grid of per-location small multiples.
Record counts on the connectors make data loss visible: you can see 4,923,978 going into a stage and 4,842,551 coming out, and ask about the difference. That is the reconciliation mindset expressed as a picture.
Watch — It is a diagram that must stay truthful as the pipeline changes — a maintenance promise, not just a design.
gpt_image_2 — Counts on every connector, a status matrix by location and stage, small multiples by month.
seedream_v5_pro — Cleaner diagram, but every connector carries the same count, so the data-loss story disappears.