Revenue fell 3%: how do you build a contribution waterfall attributing that delta to country segments?
answer
- the bars must reconcile to the headline
- mutually exclusive and exhaustive segments
- gross movement, not only net
- a share above 100% can be correct
- additive metrics only — a rate needs the mix split
basics
~20 sPartition the total into mutually exclusive, exhaustive segments, compute each segment's absolute delta, and reconcile the bars to the headline. Report gross gains and gross losses separately, not just the net, and give entrants and exits their own bars.
solid answer
~50 sRevenue is additive, so `total delta = sum of segment deltas` holds exactly — the waterfall is that identity, ordered by size. First I fix a partition that is mutually exclusive and exhaustive; if countries overlap or an "other" bucket is missing, the bars will not reconcile to the headline and the chart is worthless. Each bar is the segment's delta in currency. The share-of-decline figure `segment delta / total delta` I quote carefully, because with offsetting segments a single contribution can legitimately exceed 100%, and the ratio explodes when the net delta is near zero. So I show gross alongside net: "four countries lost 2.4M, three gained 0.9M, net -1.5M". Countries that entered or exited get their own bars rather than being smeared across the rest. And I would flag that this exactness is a property of additive metrics — a rate needs the mix-versus-within split instead.
go deeper
Know that for a total like revenue, the segment changes add up to the overall change exactly, and that the segments must not overlap. Being able to check that the bars sum to the headline is the bar here.
Be ready to explain the reconciliation check, why unmapped rows need their own bar, and why gross gains and losses tell a different story from the net. Know the formula for share of the decline and its instability near zero.
Show that you handle entrants and exits without breaking the identity, decide drill depth by who owns the dimension, and volunteer that a ratio metric cannot be charted this way. Interviewers expect the caveats unprompted.
Own the standard: which segmentation the company decomposes by default, how much analyst time a deep waterfall deserves, and how you stop the chart becoming a post-hoc search for a flattering bar.
## Why an additive metric decomposes cleanly If a metric is a **sum** over segments — total revenue, total orders, total minutes watched — then the change in the total is the sum of the changes: ``` Total' - Total = sum over segments i of (x_i' - x_i) ``` This is an identity, not an approximation. There is no interaction term, no weighting choice, no residual. That is what makes a contribution waterfall a legitimate chart for additive metrics: each bar is a real quantity of money, and the bars have to land on the headline number. If they do not, you have a partition problem, not a rounding problem. ## Step one: the partition The segmentation must be **mutually exclusive and exhaustive** over the same universe as the headline. Two failure modes recur: - **Overlap.** A customer assigned to both "United Kingdom" and "EMEA" gets counted twice, and the bars overshoot the total. - **Gaps.** Rows with a null or unmapped country silently vanish. If the headline includes them and the chart does not, the reconciliation fails by exactly that amount. The fix is a visible "unmapped" bar, not a filter. Always print the reconciliation line: sum of bars, headline delta, difference. A difference of zero is the chart's licence to be believed. ## Step two: gross versus net A net delta of -1.5M can be produced by many different worlds: one country down 1.5M with everything else flat, or four countries down 2.4M against three up 0.9M. Those call for different responses, and only the gross figures distinguish them. Report both: ``` gross decline = sum of negative segment deltas gross increase = sum of positive segment deltas net = gross decline + gross increase ``` A large gross with a small net means the business is churning underneath a calm headline — usually more urgent than a clean single-segment decline. ## Step three: share of the decline, handled honestly The natural summary is `segment delta / total delta`. Two properties surprise people: 1. **Shares can exceed 100%.** If one country lost 2.4M and the net was 1.5M, that country represents 160% of the net decline, offset by gains elsewhere. This is correct arithmetic, and it is why the sentence needs the word "net" in it. 2. **The ratio is unstable near zero.** As the net delta approaches zero, every share blows up. When the headline barely moved, drop the percentage entirely and quote absolute currency amounts. A safe convention: percentages of the **gross** movement in each direction, plus the absolute values. ## Step four: entrants and exits A country launched this period has no prior value; a country shut down has no current value. Both are genuine parts of the delta, but they are not "a segment that changed" — they are the appearance or disappearance of a segment. Give them their own bars, labelled as such. Folding a launch into the general improvement makes an expansion decision look like organic growth. ## Step five: how deep to drill Each extra dimension multiplies the number of bars and shrinks each one. Two disciplines keep it useful: - **Drill only along dimensions someone owns.** Country, platform, plan tier and acquisition channel usually map to a team that can act. Splitting by an arbitrary hash of user id produces bars that are arithmetically valid and decision-free. - **Stop when the bars stop implying an action.** Below a certain size, a segment's delta is a number on a chart rather than a thing to do, and the temptation to narrate whichever tiny bar fits the story grows with the bar count. ## Step six: know when the tool does not apply All of the above rests on additivity. A **rate** — conversion, retention, margin percent — is not a sum over segments; it is a share-weighted blend. Its segment shares are constrained to sum to one, so a segment's share change is never independent of the others', and a naive waterfall of segment rate differences will not reconcile. Rates need a mix-versus-within decomposition first, and only then can the resulting effects be charted as bars. A quick tell that someone has confused the two: a conversion-rate waterfall whose bars are expressed in percentage points of each segment's own rate. Those quantities cannot sum to a change in the overall rate. ## What good looks like in the room State the identity. Name the MECE requirement and the reconciliation check. Distinguish gross from net before quoting any share. Handle entrants explicitly. Say where you would stop drilling and why. Then, unprompted, note that the whole approach changes for a ratio metric — that last sentence is usually what separates a senior answer from a competent one.
- Your largest segment shows 140% of the decline — is the chart broken?No. Shares are taken against the net delta, so a segment that fell more than the net did exceeds 100% whenever other segments gained. It is a signal that gross movement is much larger than net, which is worth surfacing directly: report the gross decline, the gross increase and the net side by side, and stop quoting percentages entirely if the net is close to zero.
- How deep do you drill before the bars stop being useful?Drill only along dimensions that map to an owner who can act — country, platform, plan tier — and stop once a bar no longer implies a decision. Every extra split shrinks the bars and multiplies the opportunities to narrate whichever one fits the story you already had. A decomposition that nobody can act on is arithmetic, not analysis.
- Would the same waterfall work for an overall conversion-rate drop?No. A rate is a share-weighted blend, not a sum, so segment rate differences do not add up to the change in the overall rate. Segment shares are constrained to sum to one, meaning one segment's share gain forces losses elsewhere. You need the mix-versus-within decomposition first, then chart those two effects and their per-segment contributions.
saying these in an interview costs you the question
- Reports only the net movement and hides offsetting segments
- Uses overlapping segments so the bars overshoot the total
- Drops null or unmapped rows instead of showing them as a bar
- Divides by a near-zero net delta and quotes a wild percentage
- Applies an additive waterfall directly to a rate metric
- Smears a newly launched segment across the existing ones