Skip to content
Saiku Cloud

Getting insight from your data

The Saiku workbench is a tool, not an answer. The other pages cover mechanics — drag here, click there. This page covers what to actually do with the mechanics: how to take a business question and turn it into a Saiku analysis that answers it.

Start with the question, not the data

The biggest mistake analysts make in OLAP tools is starting from “what data do we have?” and clicking around until something interesting shows up. That’s exploration, not analysis. Useful sometimes — usually not.

A better start: write the question down before opening Saiku.

  • “Which regions grew revenue more than 20% year-over-year in Q4?”
  • “What’s our gross margin trend by product line over the last 12 months?”
  • “Which customers haven’t bought anything in the last 60 days but bought regularly before that?”

With the question explicit, you know what you need:

  • A measure (revenue, margin, customer count).
  • A dimension to slice by (region, product line, customer).
  • A time filter (Q4, last 12 months, last 60 days).
  • A comparison (vs prior year, vs trend, vs population).

Now the workbench moves are obvious.

The four-step pattern

Most analyses follow the same shape:

  1. Pick the measure. What number answers the question?
  2. Pick the slicing dimension. What do you want it broken down by?
  3. Pick the time window. What period?
  4. Pick the comparison. Vs what?

Step 1 + 2 + 3 give you a number. Step 4 is what makes it an insight rather than just a value.

Worked example: regional revenue growth

The question: Which regions grew revenue more than 20% year-over-year in Q4?

Decompose:

  • Measure: Revenue (sum).
  • Dimension: Region.
  • Time window: Q4 2024 vs Q4 2023.
  • Comparison: % change between the two.

In the workbench:

  1. Drag Region onto ROWS.
  2. Drag Revenue onto MEASURES.
  3. Drag Time › Year onto FILTER; pick 2023 and 2024.
  4. Drag Time › Quarter onto COLUMNS, filtered to Q4.
  5. You now have Q4 2023 and Q4 2024 revenue side by side, by region.
  6. Add a calculated member for the ratio: ([Q4 2024 Revenue] - [Q4 2023 Revenue]) / [Q4 2023 Revenue], formatted as a percentage. (For a plain period-over-period figure, Growth calculation… on the measure chip does it without the formula.)
  7. Open the Rows axis options and Order… by that column, descending.

The top of the table answers the question directly. Right-click a region and Include Level its products to see what drove the growth.

Patterns

A few moves come up over and over.

Period-over-period. Two date filters + a calculated-member ratio. The shape above — works for “this year vs last year”, “this month vs same month last year”, whatever.

Top-N + the rest. “Top 10 customers by revenue, with everyone else collapsed into Other.” Filter the dimension to top 10; add a calculated member for total - sum(top 10) labelled “All others”.

Drill-then-pivot. Drill into a parent that looks interesting, then swap the axes (toolbar button) to see the children side-by-side instead of stacked. Quick way to reorient when the top-level shape leads you to a specific question one level down.

Composition inside a comparison. When each bar also needs to show what it’s made of, a stacked bar carries both — totals across categories, breakdown within each. For a hierarchy, a sunburst or treemap does the same job radially.

Anti-patterns

A few moves to avoid.

Pie charts with 12+ slices. Unreadable. Switch to treemap or sort + bar instead.

Drilling without filtering. Drilling Year → Quarter → Month → Day on a 5-year cube gives you ~1825 rows. Filter to a window first.

Trusting the default order. The default is whatever the cube returned — usually the dimension’s natural order, not anything meaningful about the numbers. Use Order… in the axis options before drawing conclusions from “the top of the list”.

Auto-run on a slow query. If a query takes thirty seconds and you’re still assembling the layout, every drag fires a full query nobody wanted. Untick Auto-run in the Run dropdown, compose, then run once.

When to use a chart vs the table

  • Table when you want exact numbers. “What was Q4 revenue in the South-West?” wants $2,381,492, not a bar of approximate height.
  • Chart when you want pattern. “Which regions grew, and which shrank?” wants the visual gestalt of a bar chart, not the row of numbers.

A common workflow: build in the table to confirm the data is right, switch to a chart for the final visual, then switch back to verify specific numbers if questioned.

Sharing your finding

Once you’ve found the answer, Save as on the toolbar keeps the workbook. Pick a folder the people who’ll want it can see.

Three ways to pass it on, in increasing order of permanence:

  • Send the URL. The whole query lives in the address bar. Copy, paste, done — no saving, no folder permissions, nothing to set up. Easily the fastest way to say “look at this”.
  • Share the saved workbook. Anyone with access to your workspace opens the same query and gets live numbers, because it re-runs each time. Right for something people will return to.
  • Export. Excel, CSV, PDF, or straight to email. A snapshot of the numbers as you saw them — right for people outside the platform, and for anything going into a report where the figures shouldn’t shift under you.

When Saiku isn’t the right tool

A few cases where Saiku’s strengths don’t help:

  • Row-level operational queries. “Show me orders 47281 through 47290.” That’s a SQL query, not an OLAP analysis — right-click a data cell to drill through to the rows instead.
  • Real-time dashboards. Saiku runs queries on demand. If you need a number that updates every 30 seconds with no human in the loop, a different tool fits better.
  • Free-text search across data. Saiku doesn’t search rows; it aggregates them. For “find every order mentioning ‘urgent’ in the comment field”, look elsewhere.

For everything else — slicing aggregations along dimensions, with or without a chart — Saiku is what you want.

Where to go next