Your Power BI Report
Takes 30 Seconds to Open
Users stopped waiting months ago and went back to Excel. We find the actual cause — usually the model, rarely the data volume — and fix it.
Symptoms
Does this describe
your environment?
Users have quietly stopped using it and gone back to exporting to Excel. Adoption dies silently before anyone reports a problem.
Every interaction re-queries the model. Usually a relationship or measure problem rather than hardware.
Almost always calculated columns on a fact table, or a flattened model repeating dimension values on every row.
A full reload of all history every day. Runtime grows until one day it does not finish.
Nothing was rebuilt; the data simply grew past what the design could carry. This is the most common version.
Undocumented DAX written by someone who has left. Every change is a risk nobody wants to take.
Why It Happens
It is almost never
the data volume
The assumption is usually that there is too much data and the fix is more capacity. Occasionally true. Far more often the model is asking the engine to do work it was never designed for, and no amount of capacity fixes a design problem — it just makes it more expensive.
Three causes account for most of what we find. Calculated columns on large fact tables, which are computed and stored at refresh and consume memory permanently. Flattened single-table models, which defeat the columnar compression the engine relies on and frequently double or triple the model size. And DAX that forces row-by-row evaluation where a set-based expression would do the same job in a fraction of the time.
The fourth is bidirectional relationships switched on to solve a filtering problem, which then apply ambiguous filter paths to every query in the model. It fixes one visual and slows everything else.
All four are fixable without rebuilding, in most cases. The diagnosis is what determines which one you have, and guessing is expensive.
What We Do
What the audit
actually delivers
Measured analysis using DAX Studio and VertiPaq Analyzer: what consumes memory, which columns are most expensive, where cardinality is hurting you.
Actual timings for your slowest visuals before any change, so improvement is measured rather than asserted.
Every issue ranked by impact against effort. You see what a day of work buys before committing to a week.
The measures costing you the most, rewritten, with the before and after timing for each.
Whether your model needs tuning or restructuring, stated plainly. Sometimes the honest answer is that a rebuild is cheaper than the patching.
The findings and the reasoning written down, so the next person does not repeat the archaeology.
Engagement
Timeline
and Cost
A performance audit is a fixed-price two-week engagement — you know the cost before it starts and it does not drift. If the diagnosis shows the model needs restructuring rather than tuning, that is a separate proposal and we will say so in week one rather than at the end. Full model rebuilds typically run six to ten weeks depending on source count.
FAQ
Questions
We Get Asked
Usually yes. A copy of the .pbix and a sanitised sample of the data is normally enough for the diagnosis. Implementation eventually needs access, but the audit rarely does, which removes the slowest approval step from the start of the project.
We tell you in week one rather than at the end, with the reasoning and an estimate. A rebuild is sometimes genuinely cheaper than tuning a model that was structured wrongly from the start, and you should hear that early enough to act on it.
It depends entirely on what we find, and anyone quoting a number before looking is guessing. What we commit to is measurement: we record your timings before, and you see the same measurement after. If the improvement is marginal, you will see that too.
Frequently not. Buying capacity to fix a model problem is the most expensive way to not solve it. If capacity genuinely is your constraint we will say so, but we check the model first.
Smit Chandarana, directly. There is no team to hand it to. Seven years of this inside Honda, General Motors and Microsoft.