Global IT Spend in Power BI

Where do plan, forecast and actuals drift apart?

1. Cost distribution: how spend splits across 7 business areas, 5 IT areas (38 sub-areas), 6 regions (30 countries) and 5 cost-element groups (23 sub-groups).
2. Budget and forecast alignment: where are the largest gaps between actual, forecast and plan, over time and geography?
3. Structural drivers: what drives deviations in areas such as internal versus external labour, shared services and R&D planning?

Power Query prep, a clean model, DAX measures

• Append and tag. Combine the two sheets (81,610 + 12,119 rows) and tag each with a dataset period, so a single click can return to the original view.
• Replace nulls with 0. Each ERP row fills only one of actual, forecast or plan. Dropping rows with nulls would erase the budget.
• Sort months properly. Add a month number and quarter, and sort by column, so Power BI doesn't order months alphabetically.
• Keep the credits. 1,463 negative actual rows (about −$126M) are valid chargebacks and accrual reversals. Removing them would inflate net spend.
• Measure. Variance is defined once as Actual − Plan, so negative means underspend, and used everywhere.

A Python EDA notebook sits alongside the dashboard for checks that are easier in code.

Full-year actuals land 3.77% under plan

• $900.40Mfull-year plan
• $890.54Mforecast
• $866.41Mactual
• −$33.99Mactual vs plan (−3.77%)

The drill-downs are where the questions live: variance by business area and region, internal versus external labour, and a clear flag on Europe shared services overspend.

Plan is a full year; actuals are not

The first version of the numbers looked alarming: actuals 38% below plan. That is a comparison of eight months of actuals against twelve months of plan. The fair like-for-like view is January to August only:

• Use the month slicer, not the dataset-period filter, for a fair year-to-date comparison.
• Replacing nulls with 0 is safe here only because each row fills exactly one measure.
• This is a practice dataset, so the real-world causes of the variances are not known.