Power BI consultant

A Power BI consultant builds the data model, the DAX measures and the reports on top, so any figure traces to a formula and a table. It all stays in your own Microsoft tenant, written up so your team can change it after handover.

Last reviewed

Gross margin 38.4% in October, traced to four measures and three tables

Example data

38.4%

Gross margin

October 2026

Revenue
£487,500
Cost of sales
£300,300
Gross profit
£187,200

Gross margin % → Gross profit, Revenue

Gross margin % = DIVIDE ( [Gross profit], [Revenue] )

Gross profit = [Revenue] - [Cost of sales]

Revenue = SUM ( Sales[Net amount] )

Cost of sales =
SUMX (
    Sales,
    Sales[Quantity] * RELATED ( Product[Unit cost] )
)

Sales[Net amount], Sales[Quantity], Product[Unit cost]

Star schema. Date: dimension, marked as date table, one row per day, columns Date, Month, Year. Sales: fact, one row per order line, columns Date key, Product key, Quantity, Net amount. Product: dimension, one row per product, columns Product key, Product, Unit cost. Relationships: Date 1 → * Sales and Product 1 → * Sales, both single direction. The trace runs to Sales, filtered by Date[Month] = October 2026.DateDimension, marked as datetableOne row per dayDateMonthYearSalesFactOne row per order lineDate keyProduct keyQuantityNet amountProductDimensionOne row per productProduct keyProductUnit cost1*1*filtered by Date[Month] = October 2026Star schema. Date: dimension, marked as date table, one row per day, columns Date, Month, Year. Sales: fact, one row per order line, columns Date key, Product key, Quantity, Net amount. Product: dimension, one row per product, columns Product key, Product, Unit cost. Relationships: Date 1 → * Sales and Product 1 → * Sales, both single direction. The trace runs to Sales, filtered by Date[Month] = October 2026.DateDimension, marked as datetableOne row per dayDateMonthYearSalesFactOne row per order lineDate keyProduct keyQuantityNet amountProductDimensionOne row per productProduct keyProductUnit cost1*1*filtered by Date[Month] = October 2026Star schema. Date: dimension, marked as date table, one row per day, columns Date, Month, Year. Sales: fact, one row per order line, columns Date key, Product key, Quantity, Net amount. Product: dimension, one row per product, columns Product key, Product, Unit cost. Relationships: Date 1 → * Sales and Product 1 → * Sales, both single direction. The trace runs to Sales, filtered by Date[Month] = October 2026.DateDimension, marked asdate tableOne row per dayDateMonthYearSalesFactOne row per order lineDate keyProduct keyQuantityNet amountProductDimensionOne row per productProduct keyProductUnit cost1*1*filtered by Date[Month] = October 2026Star schema. Date: dimension, marked as date table, one row per day, columns Date, Month, Year. Sales: fact, one row per order line, columns Date key, Product key, Quantity, Net amount. Product: dimension, one row per product, columns Product key, Product, Unit cost. Relationships: Date 1 → * Sales and Product 1 → * Sales, both single direction. The trace runs to Sales, filtered by Date[Month] = October 2026.DateDimension, marked as date tableOne row per dayDateMonthYearSalesFactOne row per order lineDate keyProduct keyQuantityNet amountProductDimensionOne row per productProduct keyProductUnit cost1*1*filtered by Date[Month]= October 2026
Example data: an invented month of sales. Revenue of £487,500 less cost of sales of £300,300 leaves gross profit of £187,200, which is 38.4% of revenue.

What this fixes

Four signs a Power BI model needs rework.

Two reports, two answersThe sales report and the finance pack give different revenue for the same month, and the meeting turns into working out which one is right.
The slow refreshThe overnight refresh fails some mornings, or runs for an hour, and nobody can say which step is slow.
One person's reportOnly one person can change the report, and they're on holiday, or they've left.
Finance won't sign it offThe numbers look plausible, but finance won't put them in the board pack because nobody can show how they're calculated.

What you get

The model, measures and reports you keep.

The data modelA star schema with one fact table per business process, its grain written at the top of the table description, and one date table marked as the date table.
The measures tableEvery figure as an explicit, named DAX measure in one table, each with a description.
The reportsPages built for the questions agreed at scoping, where every visual uses a measure from that table.
Refresh and securityScheduled refresh running in your workspace, and row-level security checked with “View as” for every role before you sign it off.
HandoverA recorded walkthrough of the model, and a one-page map from each figure on the reports to the measure and table behind it.

If a figure can't be traced to its measure and its table in under a minute, the report isn't finished.

Power BI consultant or Power BI developer: how a project runs

How a consultant's job differs, and how a project runs.

A Power BI developer is usually hired to build reports to someone else's specification. A consultant also decides how the model is shaped and which licence it needs. The wrong shape makes totals move with a slicer. The wrong licence runs out of refreshes, or has every reader paying for Pro.

  1. 1

    Scoping

    The questions the reports must answer, the sources, the grain of each fact table, and how many people will read the reports, because that last number decides the licence. You approve this in writing before anything is built.

  2. 2

    Build

    Model first, then measures, then report pages. You review the model and the measures table before any page is designed, while a disputed definition is still cheap to change.

  3. 3

    Handover

    The walkthrough and the map, then a check that someone on your side can add a measure and publish it without help.

  4. 4

    Afterwards

    You own it. The workspace, the model and the reports were in your tenant from the first day, so there's nothing to transfer.

Where Power BI dashboards go wrong

Common model and licence mistakes, with the guard for each.

Totals that move when they shouldn'tA flat table imported from a spreadsheet, a many-to-many join between two fact tables, or a filter set to both directions makes totals shift with a slicer in ways nobody can explain. The guard is the star schema: facts at a stated grain, shared dimensions between them, relationships that filter one way, and a single marked date table.
Margins nobody can reproduceA margin worked out per row in a calculated column and then averaged gives a different answer from total gross profit over total revenue, and an implicit sum dragged onto a visual hides how either was made. The guard is the measures table. Every ratio is a ratio of sums, written with DIVIDE so a month with no sales shows a blank where a division error would be, and every measure says in its description what it counts.
A licence picked before the data was measuredA model that grows past 1 GB, or a business that needs data fresher than eight refreshes a day, outgrows Pro. A large group of people who only read reports changes the sums again. The guard is measuring model size, refresh need and the number of readers at scoping, then choosing between Pro, Premium Per User and Fabric capacity from those numbers.

Everything is built in your Microsoft tenant from the first day. If the work stopped tomorrow, you'd lose a contact and nothing else.

Industries and examples

Worked examples and the sectors this work serves.

The finance dashboard in our KPI library shows the kind of report this work produces: a monthly pack on example data, with the formula behind every line. Power BI fits ecommerce businesses whose gross and net sales must both reconcile to the finance ledger, and financial services firms where row-level security decides who sees which book of business.

Questions people ask

Answers on licences, tenants, timings and existing models.

Do you build Power BI dashboards for retail and ecommerce?

Yes. The usual model has sales at order-line grain, with products, customers and a date table around it, and returns kept as their own fact table so gross and net sales both reconcile to the finance ledger. If the order data from the site can't be trusted yet, fixing the tracking behind it comes first.

Can you fix an existing Power BI model rather than rebuild it?

Often, yes. The work starts with a review of the model you have: its relationships, filter directions, implicit measures and calculated columns. If the grain of the fact tables is right, the fixes go measure by measure and your reports keep working. If the grain is wrong, rebuilding the model is usually quicker, and the existing report pages are pointed at the new one.

Do we need Premium or Fabric capacity?

It depends on the size of the model, how often it must refresh, and how many people only read reports. Pro covers a model up to 1 GB refreshed up to eight times a day, and everyone who views a report needs a Pro licence. Premium Per User raises those limits to 100 GB and 48 refreshes. A large audience that only reads reports is where Fabric capacity starts to make sense, because from F64 upwards those readers can view with a free licence. All of this gets measured at scoping, before any licence is bought.

Do you work in our tenant?

Yes, always. You create an account for the work, the workspace sits in your tenant from the first day, and you remove the account when the project ends. On day one you can open the empty workspace in your own Power BI service, and every model and report appears there as it's built.

How long does a Power BI project take?

Length depends on how many sources the model joins, how much cleaning the source data needs before it can be modelled, how many measures finance has to agree, and how quickly reviews come back. One clean source with a dozen agreed measures is a short project. Several sources with disputed definitions take longer, and most of the extra time goes on agreeing what the numbers mean.

Book a call about your Power BI

Bring the report people argue about. The first call works out whether the trouble is in the model or the measures.

Book a call