31.9 Genpact, Mu Sigma, Fractal, Amazon and FedEx India

M84. Dataflow or Power Query Editor: which is better for transformations?

Reported for: Genpact [S29]

They use the same Power Query language. Power Query in Desktop transformations belong to one semantic model. Dataflows run the transformations in the Service and store the result, so many models and reports can reuse the same cleaned table (for example one cleaned "Blinkit Stores" table for every city team) and it is refreshed once. Use dataflows for shared, reusable entities and Desktop Power Query for model-specific shaping.

M85. What is the difference between the Contributor and Member workspace roles?

Reported for: Genpact [S29]

Workspace roles: Admin, Member, Contributor, Viewer. A Contributor can create, edit, delete and publish content in the workspace. A Member can do all of that and also add people (with Member or lower roles), share items and publish or update the workspace app. Admins additionally manage the workspace settings and can delete it (Module 25.3).

M86. Show the top 5 products and group the rest as "Others".

Reported for: Mu Sigma [S30]

Create a table of product names plus an "Others" row, relate nothing (disconnected), and use a measure:

Product Axis = UNION(DISTINCT(Product[Product Name]), ROW("Product Name", "Others"))

Sales Top5 + Others =
VAR p = SELECTEDVALUE('Product Axis'[Product Name])
VAR top5 = TOPN(5, ALL(Product[Product Name]), [Total Sales])
RETURN IF(p = "Others",
    [Total Sales] - CALCULATE([Total Sales], top5),
    IF(p IN top5,
       CALCULATE([Total Sales], Product[Product Name] = p)))

Put Product Axis on the chart axis and this measure in values. Products outside the top 5 return blank and disappear, while "Others" shows their combined sales. For a static view, a Power Query grouping can also work.

M87. How would you design a customer-churn dashboard? Which KPIs?

Reported for: Mu Sigma [S30]

Define churn first (for example "no Blinkit order in the last 30 days"). KPIs: active customers, churned customers, churn rate, retention rate, repeat customer %, average days between orders, and AOV of churned vs retained customers. Visuals: a monthly trend line of churn rate, a matrix by city and area, cohort retention (signup month × months since signup), and a table of at-risk customers (last order 20–30 days ago) for action. Protect personal data with RLS.

M88. Power Query vs Power Pivot, and Power Query vs DAX: when do you use each?

Reported for: Mu Sigma [S30] · also Accenture [S21], Deloitte [S1] (Power Query)

Power Query (M) extracts and cleans data before loading (split, merge, unpivot, fix types). Power Pivot is the in-memory data model (tables, relationships) that DAX runs on. It is the same engine as Power BI's model. Rule of thumb: shape and add static row-level columns in Power Query; write aggregations and filter-responsive logic as DAX measures. Use DAX calculated columns only when the column depends on model relationships or DAX-only logic.

M89. Messy date columns come in different formats. How would you clean them?

Reported for: Mu Sigma [S30] · also Deloitte [S2]

Standardise in Power Query: trim, replace separators ("." and "/" → "-"), then use Change Type › Using Locale with English (India) so 03-04-2026 is read as 3 April. For mixed sources, convert each source with its own locale before appending, or use try Date.FromText([Raw], [Format="dd-MM-yyyy"]) otherwise Date.FromText([Raw], [Format="yyyy-MM-dd"]). Check the errors column with Keep Errors (Module 7.7).

M90. What is the difference between CALCULATE and CALCULATETABLE? What does TREATAS do?

Reported for: Fractal [S31]

CALCULATE returns a scalar (a number or text); CALCULATETABLE returns a table, evaluated in a modified filter context. Both accept the same filter arguments. TREATAS applies the values of one table as filters on columns of another, like a virtual relationship. It is useful when Targets are at City + Month level and have no physical relationship:

Target Sales =
CALCULATE(SUM(Targets[Target]),
    TREATAS(VALUES(DarkStore[City]), Targets[City]),
    TREATAS(VALUES('Date'[Year Month]), Targets[Year Month]))

M91. What does KEEPFILTERS do?

Reported for: Fractal [S31]

A normal CALCULATE filter such as DarkStore[City] = "Pune" replaces any existing filter on City. KEEPFILTERS(DarkStore[City] = "Pune") intersects with the existing filter instead. If the user has selected Nashik, the normal version still shows Pune's sales, while the KEEPFILTERS version shows blank (Nashik ∩ Pune is empty).

M92. Write a customer retention measure.

Reported for: Fractal [S31]

Retained Customers =
VAR thisMonth = VALUES(Orders[Customer ID])
VAR lastMonth = CALCULATETABLE(VALUES(Orders[Customer ID]),
                    DATEADD('Date'[Date], -1, MONTH))
RETURN COUNTROWS(INTERSECT(thisMonth, lastMonth))

Retention % =
DIVIDE([Retained Customers],
       CALCULATE([Active Customers], DATEADD('Date'[Date], -1, MONTH)))

Use it in a visual by Month: of the customers who ordered last month, the share who ordered again this month.

M93. Why should you use variables (VAR) in DAX?

Reported for: Fractal [S31]

Variables make formulas readable, avoid calculating the same expression twice (performance), make debugging easy (return a variable to inspect it), and "freeze" a value computed in the outer context before the filter context changes (this replaces EARLIER).

Gross Margin % =
VAR s = [Total Sales]
VAR c = [Cost of Goods Sold]
RETURN DIVIDE(s - c, s)

M94. What is OLTP vs OLAP?

Reported for: Amazon [S32]

OLTP (online transaction processing) systems record day-to-day transactions (the Blinkit order app writing each order): many small inserts and updates, normalised tables. OLAP (online analytical processing) systems are built for analysis: large reads, aggregations, star schemas, history. Power BI models and data warehouses are OLAP-style, so report from a warehouse or replica, not directly from the busy transactional database.

M95. When would you use a line chart vs a pie chart, and what makes a good visualization?

Reported for: FedEx India [S33] · also Deloitte [S1] [S4], EY [S5] (visuals used and why)

Line chart: trends over continuous time (daily orders in Pune over three months). Pie/donut: part-to-whole with only a few categories (payment mode share: UPI, Card, Cash, Wallet), and only when exact comparison isn't needed. Good visualization: answer one question per visual, choose the simplest chart that fits, sort bars, label clearly, use consistent colours, avoid 3-D and clutter, and start bar axes at zero (Module 23).