Power BI & Business Analysis

Weeks 1–2 — the analyst mindset and the data layer. Most dashboards fail at this stage, not at the chart.

1. What a business analyst is actually asked

   The request is never "build a dashboard". It is "why did returns go up in the North?" — and the deliverable is an answer, with the numbers as supporting evidence.

   Example Starting from the tool and building something pretty is the classic junior mistake: a beautiful dashboard that does not answer the question in the meeting gets rebuilt by somebody else the following week.

   Practice Write the exact question, the decision it changes, and the three numbers needed to answer it — before opening any tool.

2. Shaping data in Power Query

   Power Query records your cleaning steps and replays them on every refresh. That is the difference between a one-hour job every Monday and a report that maintains itself.

   Example Removing a top row, promoting headers, changing a type and filtering blanks are four recorded steps. Retyping the same cleanup by hand next month is what makes an analyst look slow.

   Practice Import a messy CSV, clean it entirely in Power Query — no formulas — and save the file. Then add a row to the source and refresh.

3. A star schema in plain language

   One fact table of events and a few dimension tables of descriptions. It sounds academic until the first time a single flat file cannot answer "sales by region by month" without duplicating rows.

   Example A flat file with the region name repeated in every sales row doubles every total the moment it is joined to a region table. Splitting region into its own table is what fixes the double count, and it is the single most common interview question in this area.

   Practice Split your dataset into one fact table and two dimension tables, then draw the relationships before building anything.

Weeks 3–4 — measures, filters and time. This is where Power BI stops being a drawing tool.

1. Measures versus calculated columns

   A calculated column is stored once per row and grows with the table; a measure is computed at the moment you look at it. Totals and percentages almost always belong in a measure.

   Example A "profit margin" calculated column gives you the average of the per-row margins, which is not the margin. As a measure — total profit divided by total sales — it gives the number a manager expects.

   Practice Build one of each for the same figure, put them on a card side by side, and find the difference in a filtered view.

2. CALCULATE and time intelligence

   CALCULATE changes the filters a measure is evaluated under, which is how "this month", "same month last year" and "share of total" all become one line each.

   Example Total sales and sales for the same period last year are the same measure with one extra filter argument. Writing them as two disconnected formulas works until the fiscal year changes.

   Practice Add four measures: total, previous month, month-over-month growth percentage and share of the total.

3. The three charts that are hard to get wrong

   A column or bar chart for comparison, a line for time and a small table for detail. Almost every business question fits one of the three, and choosing correctly prevents most bad dashboards.

   Example A pie chart with nine slices answers nothing. The same data as a sorted bar chart makes the answer obvious in one glance, which is the only reason to draw a chart at all.

   Practice Rebuild one of your own charts as the correct type and write a sentence on what changed about what you noticed.

Weeks 5–6 — the artefact and the words that get you hired. Analysts are chosen on the write-up, not the visuals.

1. The dashboard that answers one question

   One screen, one question, one row of headline numbers, one trend and one breakdown. Everything else is a second page that nobody opens.

   Example A dashboard with the answer written at the top in one sentence is used in meetings. The same dashboard without it gets a follow-up call asking what it means.

   Practice Rebuild your dashboard so the answer is readable in five seconds by someone who saw it for the first time.

2. Writing the recommendation, not the description

   "North returns rose from 3% to 6% between July and September, driven by the two SKUs with the highest damage rate. I would recheck the packing line for those two lines before the festive season." That is a recommendation. "Returns by region" is a description.

   Example The numbers are the evidence; the recommendation is the deliverable. Report writers who describe get asked for the analysis again by email.

   Practice Write a three-sentence recommendation above one of your dashboards, naming the cause and the next action.

3. Three case studies a fresher can defend

   Pick three questions you can answer end to end: a sales decline, a returns spike and a cost comparison. Depth in three beats a portfolio of twelve shallow dashboards.

   Example For each, be ready for "how did you clean it?", "why that chart?" and "what would you do next?" — the follow-ups are where the interview actually happens.

   Practice Write one page per case study, publish it, and rehearse each answer out loud in under two minutes.

Ranking, distinct counts and who is allowed to see which rows.

1. Ranking and TOPN

   Ranking answers "which products are in the top ten" without sorting a table by hand. RANKX does it as a measure, and TOPN returns just the rows that matter.

   Example RANKX(ALL(dim_product[product]), [Total Sales]) numbers every product by sales, so a table filtered to rank <= 10 shows the top ten.

   Practice Build a measure that ranks products by sales and show only the top five on a visual.

2. Distinct counts and semi-additive measures

   Some numbers must not be added across time: a headcount of 100 in January plus 100 in February is not 200 people. Distinct counts and the latest-value pattern keep those numbers honest.

   Example DISTINCTCOUNT(fact_sessions[user_id]) counts unique users; averaging it across days is wrong unless the report says "average daily users".

   Practice Create a distinct-count measure for users and a month-end measure for headcount, and explain in the tooltip why they are not summed.

3. Row-level security: the right rows for the right person

   Row-level security lets one report serve many people while each sees only their own region or team. The filter is a role you assign, applied by the service on top of the query.

   Example A role with a DAX filter dim_region[region] = USERPRINCIPALNAME() shows a manager their own region and nothing else, from the same file.

   Practice Create one region role, test it with "View as role", and note where you would assign people in the service.

Publish, refresh, share, and write the analysis a stakeholder acts on.

1. Publishing, refresh and gateways

   A published report is only as good as its refresh. Schedule it, watch the first few runs, and know how the service reaches a database that sits behind a firewall.

   Example A scheduled refresh at 7am needs the gateway credential to be entered correctly once; a failed refresh sends a mail and the first you hear of a stale report is a decision made on old numbers.

   Practice Publish one report, schedule a refresh, and confirm it succeeded by checking the history.

2. Sharing: workspaces, apps and one source of truth

   Share an APP, not a file. An app is one published bundle everyone opens, so the number is the same for everyone, and you update it in one place.

   Example The sales team opens the app; you edit the report and press Update app, and every viewer sees the new version without downloading anything.

   Practice Publish an app to one group, set its permissions, and write one line telling them where to open it.

3. A BA case study you can defend

   A business analyst is hired on one story: the question, the data, the finding and what changed. Build one case study end to end and you can answer every question an interviewer asks.

   Example A report that shows time-to-fill by role, the bottleneck in the interview stage, and the recommendation to add a second interviewer - that is a case study in one page.

   Practice Write your own case study: the question, the data, the finding, the recommendation and one limitation.