# Variance Charts > Charts for actual versus plan, forecast, or prior - bullet charts, deviation bars, dumbbell and slope charts, variance bands, and small-multiple budget-vs-actual grids. Covers plotting the variance rather than two lines the reader must subtract, and encoding favourable-versus-unfavourable when direction and goodness disagree. Trigger on "actual vs budget", "BvA chart", "vs plan", "variance chart", "bullet chart", "target", "over/under", "compare to forecast", "gap to plan". - Skill: `lukehle/variance-charts` (Agent Skill) - Install (CLI): `npx skillmds@latest add lukehle/variance-charts` - Raw SKILL.md: https://api.skillmd.com/api/skills/lukehle/variance-charts/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Finance & Business - Author: Lukehle (https://skillmd.com/u/lukehle) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/lukehle/variance-charts --- # Variance charts The core principle, and it is violated constantly: > **Plot the variance. Do not plot two lines and make the reader subtract.** Two overlaid lines show that a gap exists. A variance chart shows how big it is, which direction, and whether it is closing. The second is the question. --- ## Bullet chart — the default for actual vs target Compact, precise, and it replaces both the gauge and the two-bar comparison. ``` Revenue ▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░ │ $4.82M / $5.10M plan (5.5%) under Gross margin ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ │ 76% / 75% plan +1 pt Net burn ▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░ │ $412K / $455K plan better └─ actual (bar) └─ target (tick) ``` - **Bar is the actual, a perpendicular tick is the target.** The tick reads as a threshold; a second bar reads as a second value. - Optional qualitative bands behind (poor / acceptable / good) — keep them very low contrast or they compete with the data. - **One row per measure, stacked**, sharing an axis where the units allow. Where they do not, give each row its own normalized scale and label the values. - Sort by variance magnitude, not alphabetically. Bullet charts fit a dozen measures in the space of one gauge and add the comparison. There is no case where a speedometer is the better choice. --- ## Deviation bars One bar per line item, drawn from a zero line, length = variance. Best when the question is *which items missed and by how much*. - **Zero line is the plan**, not the axis edge. - Sort by variance so the biggest misses are adjacent and readable. - Label with **both absolute and percentage** — one alone misleads. $40K on a $50K line is a catastrophe; on a $4M line it is noise. - Consider two panels: one sorted by absolute impact, one by percentage. They rank differently and both matter. --- ## Favourable versus unfavourable — the encoding trap **Direction is not goodness.** Spend under budget is negative and good. Churn above plan is positive and bad. Decide which the colour encodes, then be consistent and say so: | Approach | Use when | |---|---| | Colour = **goodness** (favourable/unfavourable) | Mixed measures on one chart — the usual case for a P&L variance view | | Colour = **direction** (up/down) | Every measure on the chart shares a polarity | Either way: - **Sign or arrow carries the direction**; colour reinforces (`artifact-accessibility`). - Label the legend explicitly — "green = favourable", not just a colour swatch. - Never let a reader infer polarity from the chart alone. A revenue miss and an expense saving are both "green is good" — but they point opposite ways on the axis, and only an explicit legend prevents the misread. --- ## Variance bands on a time series For actual versus plan over time, plot actual as a line and the plan as a band or a reference line, shading the gap. - Shade the gap area, colour-coded by favourability — the shaded area *is* the cumulative variance, which is more informative than either line. - Mark the forecast boundary where actuals stop and projection begins. A solid line becoming dashed is the clearest convention. - **Name the plan version**: "vs Plan (board approved 2026-01)". Plans get revised; an unlabelled "plan" is unreconcilable three months later. --- ## Dumbbell chart Two dots joined by a line, one row per category. Excellent for **before/after** or **plan/actual** across many categories. - The line length is the variance and reads instantly. - Sort by gap size. - Consistent dot colours across all rows — plan always one colour, actual always the other. - Better than a grouped bar chart above about six categories, because grouped bars force the eye to compare non-adjacent pairs. --- ## Slope chart Two time points, one line per entity. Shows **rank change and rate of change** at once. - Label both ends directly; no legend. - Highlight the two or three entities that matter; grey the rest. A slope chart with 40 equally weighted lines is spaghetti. - Best for "who moved" questions — segment mix shifts, customer ranking changes. --- ## Small-multiple BvA grid One small panel per line item or department, each showing actual against plan over time. - **Shared scale across every panel**, always. Per-panel scales make a department that missed by $2M look identical to one that missed by $20K. - Sort panels by variance magnitude, not alphabetically. - Keep panels small and dense — the pattern across panels is the point, not any single panel. - One shared legend and one shared axis label, not per panel. --- ## What to avoid | Anti-pattern | Why | |---|---| | Two overlaid lines with no variance shown | Makes the reader do arithmetic; the gap is the message | | Grouped bars for actual/plan across many categories | Non-adjacent comparison; use dumbbell | | Percentage variance with no absolute | 340% on a tiny base is noise | | Colour alone for favourable/unfavourable | Fails greyscale and colourblind readers | | Unlabelled plan version | Unreconcilable later | | A gauge | Bullet chart, always | | Truncated axis on an additive measure | Doubles the apparent miss | --- ## Related skills - `chart-selection` — choosing among these - `bridge-charts` — the budget-to-actual bridge, which explains *why* the variance exists - `timeseries-finance` — the time dimension - `chart-annotation` — naming the plan version and stating the finding - `financial-tables` — the variance table that usually accompanies these