Knowing a metric moved is step one. Contribution analysis shows which segment drove it, and how VedaX Studio surfaces those drivers for you.
Picture a Monday morning. Someone drops a number in the team channel: revenue is up 12% this month. Good news, everyone nods, and the thread moves on.
But that number isn't an answer. It's a question wearing an answer's clothes. *Up 12% because of what?* A total is the sum of dozens of smaller stories — regions, plans, channels, product lines, customer segments — and they rarely move together. Some are surging, some are flat, a couple might be quietly sinking. The headline 12% is just the net result after all of them cancel out.
If you stop at the headline, you make decisions on a blur. You might pour budget into a channel that's actually dragging, or miss the one region carrying the whole quarter on its back. The headline tells you that something changed. It can't tell you what to do about it.
This is the gap contribution analysis fills. It's the difference between *the metric moved* and *here's the handful of things that moved it.* And it's the natural next step after you've learned to tell a real shift from random wobble — which is its own skill, covered in our guide to reading whether a metric is a trend, an anomaly, or just noise. Once you trust that the change is real, the only useful question left is: who drove it?
Contribution analysis (you'll also hear it called *driver analysis* or *decomposition*) takes a change in a total and splits it across the categories that make up that total. Instead of one number, you get a ranked list of who contributed what.
The mechanics are simpler than the name suggests:
1. Pick the metric that moved — revenue, signups, orders, churned accounts, support tickets.
2. Pick a dimension to slice it by — region, plan, channel, status, product category. A dimension is just a label that groups your rows.
3. Compare each slice's value now versus the prior period — , and measure how much each slice added to or subtracted from the overall change.
The output is a sentence you can actually act on: *"Revenue rose, and most of that rise came from the Enterprise plan; SMB was roughly flat; one region declined and partly offset the gain."* That's three facts you can take into a meeting, versus one number you can only stare at.
The split between metric (the thing you measure) and dimension (the thing you slice by) is the engine underneath all of this. If those two words feel fuzzy, the plain-English guide to metric vs. dimension is the fastest way to make them click — because contribution analysis is literally *a metric, broken down by a dimension, across time.*
One distinction worth holding onto: contribution analysis tells you where the change concentrated, not always why it happened. "Driven mainly by the Enterprise plan" is a where. The why — a price change, a big deal closing, a seasonal pattern — is the follow-up investigation. But knowing the where shrinks the search for the why from the whole business down to one corner of it.
Let's walk a fully hypothetical store so the shape of the analysis is concrete. Imagine an online retailer whose total revenue rose 12% month over month. Leadership is pleased. Then someone runs the breakdown by region.
Imagine the picture comes back like this:
- North — up sharply, and large enough that it accounts for almost the entire company-wide gain.
- West — essentially flat.
- South — flat to slightly down.
- East — actually declining, quietly cancelling out part of North's surge.
Now the headline reads completely differently. "Revenue up 12%" was never a company-wide story. It was a North story that the other three regions were partly masking. The real total movement is bigger than 12% in North and *negative* in East — the 12% is just what's left after they net against each other.
That reframe changes every decision on the table:
- Don't celebrate evenly. Whatever's working in North is your lever. Find it and pull it harder.
- East is a problem the headline hid. A declining region inside a "growing" company is exactly the kind of thing that festers because the top-line number looks fine.
- Your forecast is fragile. If growth depends on one region, a single bad month there erases the whole company's gain. Concentrated growth is riskier than it looks.
Notice that none of this required a fancy model — just splitting one number by one good dimension. The hard part is rarely the math. It's having a clean dimension to split by in the first place, and remembering to actually do the split instead of trusting the headline.
The catch with contribution analysis is that nobody runs it on a metric that looks fine. You only go digging *after* something feels off, which means you find the East-style declines late, if at all. The fix is to make the breakdown automatic rather than something you remember to do.
That's the job of the Insights feed in VedaX Studio. It scans your latest data proactively and surfaces findings without you asking — the broader shift from dashboards-you-check to insights that come find you. Among those findings is a category built specifically for this problem:
- Trends tell you a metric moved X% versus the prior period — the headline.
- Drivers / contribution go the next step and name the segment behind it, in plain language like *"records up, driven mainly by Completed"* or *"revenue up, driven mainly by one region."* That's contribution analysis, pre-computed and handed to you.
- Goals flag whether you're on or off track against a target.
- Anomalies call out unusual spikes and drops a scheduled scan has caught.
Every finding comes with two doors. "Ask Buddy" lets you interrogate it conversationally — VedaX Buddy writes the SQL and returns a chart or table, so you can type *"show me East revenue by week"* in plain English instead of hand-writing a query (the same natural-language-to-SQL workflow, pointed at a specific driver). "View data" drops you straight into the underlying rows so you can verify the claim yourself.
If a finding matters enough to watch, you don't have to keep checking the feed either. The opt-in Insight Digest emails your insights — drivers included — to chosen recipients on a daily or weekly cadence, at an exact time in your workspace's timezone. It's off by default; you turn it on for the metrics that deserve a standing place in your inbox. We cover that setup end to end in the analytics email digest guide, and the wider pattern of automated reports and alerts if you want a metrics ritual that runs itself.
Automatic driver detection is only as good as the columns it has to slice by. Feed it a messy dimension and you get messy drivers — or none at all. A few things quietly wreck a breakdown:
- Inconsistent labels. If your region column holds "North", "north", and "N" as three separate values, your North story gets split into three weaker stories and the real driver hides in the fragments.
- A free-text field pretending to be a category. A column with two thousand unique values isn't a dimension you can decompose — it's noise. Good dimensions have a sane, repeatable set of values.
- A column VedaX has mis-classified. This is the subtle one. To break down a metric by a dimension, the tool first has to know *which column is the metric and which is the dimension* — and that you should never sum an identifier like an order ID.
VedaX handles that last point with semantic column roles, its built-in business dictionary. It classifies each column as a metric, a dimension, an identifier, or a date, so it analyses the right things and won't, say, try to average a postal code. When it gets one wrong, you correct it — *"this isn't a metric,"* or *"track this instead"* — and your correction sticks everywhere the column is used. That's worth understanding in its own right; the semantic layer / business dictionary guide goes deeper.
The practical takeaway: a few minutes spent cleaning a dimension and confirming its role pays off every time the breakdown runs. It's the unglamorous prep that makes "driven mainly by X" trustworthy instead of suspicious.
Contribution analysis has a famous failure mode, and it catches smart people. It's called Simpson's paradox: a trend that holds in every individual group can *reverse* when you combine the groups. The culprit is mix shift — a change in how your total is composed, rather than a change within any one part.
Here's the shape of it, kept fully hypothetical. Imagine a subscription business where, segment by segment, retention actually *improved* this quarter — every plan held onto a higher share of its customers than before. Yet the company's overall retention rate *fell.* How? The mix changed: a flood of new customers landed in a lower-retention plan, so even though each plan got better, the blended average got dragged down by the shift in who's in the pool. Every part went up; the whole went down.
This is exactly where naive breakdowns mislead. A few guardrails:
- Separate rate from mix. When a blended rate (retention, conversion, average order value) moves, ask whether the rates within segments moved, or whether the *proportion* of each segment moved. They imply very different actions.
- Watch for new volume landing in one bucket. A surge of records in a single segment can swing a total even if nothing about that segment's behaviour changed.
- Use the right tool for rates over time. For retention specifically, group customers by when they joined and follow each group forward — the discipline laid out in our cohort and retention analysis guide — so a changing mix of newcomers can't disguise what's really happening to each cohort.
- Be especially careful with churn. A flat overall churn number can hide a high-value segment leaving while a low-value one grows. Reading the drivers underneath, as in spotting churn early from your data, is what surfaces the dangerous case.
The defense isn't to distrust the breakdown — it's to always ask the second question: *did the segments change, or did the mix of segments change?* VedaX's contribution findings give you the starting point in plain language; your judgment about rate-versus-mix is what turns it into a correct conclusion.
Contribution analysis earns its keep only when it changes what you do next. A simple loop to run whenever a metric moves:
1. Confirm the move is real — a genuine shift, not noise or a one-off blip.
2. Break it down by your most decision-relevant dimension — usually region, plan, channel, or status. Let the Insights feed hand you the driver, or ask Buddy for the split directly.
3. Read the concentration. — Is the change spread evenly, or is one or two segments carrying it? Concentrated change means concentrated opportunity — and concentrated risk.
4. Check for the hidden offset. — Is a declining segment being masked by a growing one? That's the East in our example, and it's the finding the headline will never show you.
5. Sanity-check rate versus mix — before you conclude, so Simpson's paradox doesn't send you the wrong way.
6. Decide and assign. — Double down on the driver, dig into the decliner, or revisit a forecast that now looks fragile.
That last step is where analysis becomes action — and a driver finding is far easier to act on than a raw number, because it already points at a specific corner of the business. If you're presenting it to others, framing the contribution as a short narrative (*"the growth is real but it's one region, and here's the region that's slipping"*) lands harder than a chart alone; that's the craft in data storytelling that drives action, and the broader move from dashboards to decisions.
Knowing a metric moved is step one. Knowing which segment drove it is the step that actually tells you what to do — and it's the step VedaX is built to take for you, automatically, before you even think to ask.
Contribution analysis takes a change in a total — like revenue up 12% — and splits it across the categories that make up that total, such as region, plan, or channel. Instead of one headline number, you get a ranked list of which segments drove the change. It answers "which part moved?" rather than just "did something move?"
A trend tells you a metric moved X% versus the prior period — that's the headline. Contribution (or driver) analysis goes one step further and names the segment responsible, e.g. "records up, driven mainly by Completed." The trend tells you that something changed; the contribution tells you what to investigate or act on.
Simpson's paradox is when a trend that holds inside every individual group reverses once you combine the groups — usually because the mix of segments changed, not the behaviour within them. For example, retention can improve in every plan yet fall overall if many new customers land in a lower-retention plan. The guardrail is to always ask whether the segments changed or whether the proportion of each segment changed.
Yes. VedaX's Insights feed scans your latest data and surfaces driver findings in plain language, like "revenue up, driven mainly by one region," alongside trends, goals, and anomalies. Each finding includes "Ask Buddy" to interrogate it conversationally and "View data" to inspect the underlying rows. You can also receive these findings by email through the opt-in Insight Digest on a daily or weekly cadence.
A breakdown is only as good as the column you slice by. Inconsistent labels (North vs north vs N) fragment a real driver into weak pieces, and free-text fields with thousands of unique values can't be decomposed at all. VedaX's semantic column roles also need to know which column is a metric versus a dimension or identifier so it analyses the right things — and you can correct any misclassification, with the correction applying everywhere.
See it in VedaX: AI business intelligence for small businesses