You support 220 analysts and managers using 14 Power BI semantic models and 90 dashboards over 1.2 TB of sales and finance data; the annual platform budget is $180,000, freshness must stay under 30 minutes, and the architecture must last 12 months. Would you build one enterprise model or domain models?

I would build owned domain semantic models with conformed dimensions and thin reports, not one enterprise model. - Sales and finance each get a star schema with a named BI owner, while Date, Customer, Product, and the approved revenue definition come from versioned shared contracts. - I would target at least 70 of 90 dashboards as thin reports on 6 to 8 certified models; moving from 14 to 8 is a 43 percent reduction in model surfaces, but refresh savings must be measured rather than inferred from that count. - An architecture decision record would cap a model at a clear business grain and 20-minute refresh budget, reserving 10 minutes for ingestion and validation within the $180,000 envelope.

A Fabric estate serves 320 analysts through 18 semantic models and 140 Power BI dashboards on 8 TB in OneLake; it has a $260,000 annual capacity budget, a 15-minute freshness SLO, and an 18-month planning horizon. How would you organize the Direct Lake semantic layer?

I would place domain-owned Direct Lake models over curated OneLake tables and keep cross-domain definitions in explicit shared contracts. - Finance, sales, operations, and customer teams each own model grain, measures, and on-call metadata, while conformed dimensions are published once with compatibility tests. - I would consolidate 18 models to roughly 10 reusable models and keep reports thin; the 44 percent reduction removes eight ownership and release surfaces, while actual CU and support savings remain measured outcomes. - The design record would require p95 visual latency under 5 seconds and data age under 15 minutes, with Capacity Metrics reviewed at 70 percent sustained utilization before spending beyond $260,000.

A company has 180 analysts, 12 LookML models with 24 Looker Explores, 75 dashboards, and 3 TB in BigQuery; annual Looker and warehouse spend is $220,000, dashboard queries need p95 under 20 seconds, and the models must support the next year. How would you structure the LookML projects?

I would use a governed core LookML project plus domain projects that import stable shared definitions. - The core project owns canonical users, dates, accounts, fiscal logic, and 12 board metrics; domain CODEOWNERS approve sales, product, and finance views at their declared grain. - I would reduce 24 overlapping Explores to 10 to 12 curated Explores and measure reused-field coverage above 80 percent, trading some local freedom for fewer conflicting joins. - A project manifest, semantic catalog, and query-cost check would block a merge if p95 exceeds 20 seconds or estimated BigQuery spend raises the $220,000 annual run rate by more than 10 percent.

Six Looker domain projects import a shared LookML project with 140 governed fields; a base-view change must reach 110 dashboards without another breaking rollout, and interactive p95 must remain under 15 seconds. How would you version and release the dependency?

I would treat the shared LookML project as a versioned contract and migrate consumers through pinned dependency references rather than an unbounded latest version. - The platform owner classifies additive and breaking changes, publishes a release manifest for all 140 fields, and keeps the prior major reference available during a 60-day migration window. - CI tests each of the 6 consuming projects with LookML validation, Content Validator, 30 golden queries, access fixtures, and p95 comparisons before a dependency reference can advance. - The rollout closes when all 110 dashboards use an approved reference, metric parity is exact, p95 remains below 15 seconds, and no consumer remains on the retired major version.

A Cube deployment serves 450 analysts and managers through 16 Cube data models and 130 embedded dashboards over 12 TB in Snowflake; the annual budget is $260,000, p95 response time must be under 3 seconds, and the design horizon is 18 months. How would you shape the semantic layer?

I would define domain cubes at stable business grains and expose consumer-facing views backed by selective pre-aggregations. - Orders, subscriptions, marketing spend, and support cases get accountable owners, tested joins, and shared dimensions, while views expose only approved measures to the 130 dashboards. - Query history would drive daily and monthly rollups for the top 80 percent of requests; I would not pre-aggregate every combination because storage and refresh cost grow multiplicatively. - A model manifest would require p95 below 3 seconds, pre-aggregation freshness below 15 minutes, and Snowflake plus Cube annual run rate at or below $260,000, with a review at 70 percent of either limit.

An enterprise has 380 BI users, 20 semantic models split between Power BI and Looker, 200 dashboards, and 25 TB in Snowflake; total BI spend is $600,000 per year, cross-tool metrics need 99 percent agreement, and the target state is due in 12 months. Where should canonical semantics live?

I would anchor business grain and additive facts in governed warehouse marts, then keep presentation semantics native to each BI tool. - Versioned contracts define order, customer, fiscal calendar, and currency conversion once; DAX measures and LookML fields may differ only where the interaction model requires it. - A reconciliation suite compares 25 critical metrics across both tools every day and blocks release below 99 percent agreement, with one named domain owner resolving each exception. - Rebuilding both tools around 8 shared marts reduces the count of isolated transformation paths from 20 by 60 percent, but measured compute, cache behavior, and support cost decide whether the design stays inside $600,000.

Five business units have 500 Power BI users, 22 semantic models, 240 dashboards, and 4 TB of imported data; annual cost is $350,000, finance close dashboards must be ready by 06:00, and the architecture must serve two years. How would you set model boundaries and ownership?

I would align model boundaries to accountable business processes and share dimensions through contracts rather than chained ownership. - Finance owns ledger and close models, sales owns pipeline, and operations owns fulfillment; every model declares grain, steward, support tier, and downstream dashboard count. - Customer, legal entity, product, and fiscal calendar receive versioned keys and reconciliation tests, while cross-domain reports combine certified outputs instead of recreating 22 variants. - I would target 12 to 14 models and retire the rest only after 95 percent of 240 dashboards map to an owner, preserving a 90-minute close refresh window inside the 06:00 SLO and $350,000 budget.

A Looker organization supports 140 analysts with 12 LookML models, 65 dashboards, and 2 TB in BigQuery; annual spend is $160,000, a new domain must publish governed analytics within 10 business days, and the operating model must last 18 months. Who should own what?

I would use federated domain ownership with a small platform team owning the semantic contract and release path. - Four domain teams own views, measures, dashboard quality, and user support, while the BI platform team owns project templates, shared dimensions, access rules, and production deployment. - CODEOWNERS requires one domain reviewer and one platform reviewer for shared LookML; the 10-day target becomes 3 days for modeling, 3 for validation, and 4 for user acceptance. - A quarterly ownership matrix flags models with no active owner or fewer than 5 monthly users, keeping support within the $160,000 budget instead of centralizing every change in one team.

A Power BI tenant has 300 users, 18 semantic models, 150 dashboards, and 9 TB of source data; annual cost is $280,000, metric changes must reach all governed reports within 5 business days, and the roadmap spans two years. How would you increase measure reuse without building a giant model?

I would standardize measure contracts and calculation patterns while leaving facts in domain-sized semantic models. - A metric registry records formula, grain, owner, dimensions, and effective date; reusable calculation groups handle time intelligence and currency rather than copying DAX into 18 models. - TMDL source files and automated dependency scans identify the 150 affected dashboards, with a target of 90 percent of governed measures generated or reviewed from the registry. - The release template budgets 2 days for impact analysis, 2 for model PRs, and 1 for deployment, accepting limited duplication when centralization would breach model size, latency, or the $280,000 capacity limit.

A SaaS company serves 80 internal analysts and 400 customer users through 14 Cube models and 80 embedded dashboards over 5 TB; annual platform cost is capped at $200,000, p95 queries must stay under 2 seconds, and the product horizon is 24 months. How would you design ownership and reuse?

I would separate internal domain ownership from a curated embedded contract that product teams can depend on. - Domain owners maintain orders, billing, usage, and support cubes, while an embedded analytics owner publishes stable views and a versioned measure catalog for the 80 dashboards. - Shared pre-aggregations cover the 20 query shapes producing at least 80 percent of traffic; tenant-specific rollups are added only when they beat shared rollups on measured latency and storage. - A compatibility suite and monthly cost scorecard block breaking changes, p95 above 2 seconds, or annual run rate above $200,000, trading custom customer fields for a predictable 24-month API surface.

You have 260 Power BI users, 9 semantic models, 70 dashboards, and 1.8 TB in Azure SQL; annual cost is limited to $140,000, data may be 30 minutes old, and the decision must hold for 18 months. Would you choose Import or DirectQuery?

I would choose Import with incremental refresh because a 30-minute freshness allowance does not justify source-bound interactive queries. - Partition the large facts by date, refresh only the latest 2 days every 30 minutes, and keep dimensions fully imported so dashboard queries use the in-memory engine. - Capacity testing must keep refresh within 20 minutes and p95 visuals under 4 seconds; DirectQuery remains an exception for drill-through fields that cannot be retained. - The architecture record compares imported model memory and 48 refreshes per day against Azure SQL concurrency cost, choosing the option that stays below $140,000 per year.

A Fabric program serves 350 users with 12 semantic models and 100 dashboards over 6 TB of Delta tables in OneLake; it has a $240,000 annual F64 budget, needs 10-minute freshness, and is planned for two years. Would you use Direct Lake or Import?

I would make Direct Lake the default for curated Delta tables and isolate unsupported patterns in small Import models. - Direct Lake avoids copying 6 TB into scheduled imports and lets semantic models read OneLake data, but every table must have stable grain, optimized files, and supported security behavior. - A proof with the 20 busiest dashboard pages must hold p95 below 5 seconds and data age below 10 minutes; any DirectQuery fallback or guardrail hit is recorded rather than accepted silently. - Capacity Metrics must remain below 70 percent sustained load within $20,000 per month, and an Import exception is justified when its compression and predictable cache cost less than tuning the lake path.

A dispatch operation has 180 users, 6 Power BI semantic models, 45 dashboards, and 900 GB in Azure SQL; annual cost is $200,000, operational data must be under 2 minutes old with p95 visuals below 5 seconds, and the design horizon is one year. Is DirectQuery appropriate?

I would use DirectQuery for the narrow operational slice and remove historical analytics from that path. - The source gets indexed serving views at dispatch, vehicle, and 5-minute grain, while 12 months of history is imported into a separate model for trend analysis. - I would load-test 180 users at twice expected peak, requiring p95 below 5 seconds and fewer than 30 concurrent source queries after visual and page-query reduction. - Azure SQL and capacity telemetry share the $200,000 annual ceiling; if source compute rises more than 20 percent, an Import aggregation or shorter live window is cheaper than scaling DirectQuery indefinitely.

A retailer has 420 Power BI users, 15 semantic models, 180 dashboards, and 7 TB of order data; the annual budget is $300,000, today's sales need 5-minute freshness while older data may refresh daily, and the architecture must last 24 months. How would you use a composite model?

I would combine imported history and aggregates with DirectQuery only for today's detailed orders. - Dual-mode Date, Store, and Product dimensions serve both paths, an Import aggregate holds store-hour metrics for two years, and DirectQuery covers the current business day. - Aggregation mappings must answer at least 90 percent of 180 dashboard queries from memory with p95 under 4 seconds, leaving live detail as an intentional slower drill-through. - The design document caps the live slice at 50 million rows and total monthly run rate at $25,000, trading 5-minute freshness for more source complexity only where users can act on it.

A marketplace serves 500 Power BI users through 10 semantic models and 220 dashboards querying 30 TB; annual warehouse and BI cost is $450,000, p95 visuals must be under 4 seconds with 15-minute freshness, and the plan covers 18 months. How would you design aggregations?

I would build Import aggregations from observed dashboard grains and retain DirectQuery detail for exceptional exploration. - Order-day-category-region and seller-hour rollups cover the dominant measures, while relationships and aggregation mappings preserve access to the 30 TB detail table. - Query logs must show at least a 90 percent aggregation hit rate across the top 50 pages and p95 under 4 seconds; unused rollups are removed after 30 days. - Each rollup needs a storage, refresh, and warehouse-scan estimate, and the portfolio is rejected if annualized cost exceeds $450,000 even when another aggregation improves one dashboard.

A payments team has 300 Power BI users, 8 semantic models, 95 dashboards, and 12 TB growing by 500 GB per day; annual cost is capped at $220,000, the latest hour needs 5-minute freshness while history may be daily, and the horizon is one year. How would you configure incremental refresh?

I would use incremental refresh for bounded historical partitions and a hybrid DirectQuery partition only for the current hour. - At 500 GB per day, I would not import 24 months of raw transactions; the model keeps only a benchmarked recent detail window plus 24 months of daily merchant aggregates, while raw history remains in the warehouse for drill-through. - RangeStart, RangeEnd, a reliable change-detection column, and partition reconciliation are source-controlled; late-arriving detail partitions refresh daily and the indexed live partition is restricted to the latest 60 minutes. - Capacity and source cost must remain at or below $220,000 per year, and I would remove the hybrid path if fewer than 10 percent of users consume five-minute data or source cost rises 20 percent.

A company has 40 Power BI authors and 360 viewers, 20 semantic models, 160 dashboards, and 5 TB of source data; it can spend $330,000 over three years, requires 99.5 percent availability, and expects stable headcount. Would you license Pro, Premium Per User, or Fabric capacity?

I would select a reserved F64 capacity plus Pro licenses for 40 authors if the workload proof shows that shared Pro limits are insufficient. - At the quoted prices, Pro for all 400 users is 400 times $14 times 36, or $201,600; PPU is 400 times $24 times 36, or $345,600, so PPU already exceeds the $330,000 cap. - An F64 reservation quoted at $95,000 per year plus author Pro licenses costs $305,160 over three years; viewers still need Fabric Free identities and access to content hosted on F64 or larger capacity. - The decision must verify model size, refresh frequency, and concurrency in a workload test and verify availability against the service contract and recovery drills; Pro wins if its shared-capacity limits pass because it is $103,560 cheaper.

A Fabric tenant serves 480 users with 25 semantic models, 210 dashboards, and 18 TB in OneLake; the 18-month capacity budget is $500,000, 99 percent of visuals must render within 5 seconds, and peak use is 1,200 visual queries per minute. How would you size capacity?

I would size from a representative concurrency benchmark and Capacity Metrics, not from the 18 TB storage number. - Replay the 30 busiest pages at 1,200 visual queries per minute while scheduled refresh and Direct Lake framing run, then record interactive and background CU consumption by item. - Start with the smallest SKU that keeps p99 interactive delay at zero, visual p99 under 5 seconds, and sustained utilization below 70 percent; test one SKU above it for the growth margin. - The sizing memo compares reserved and scale-up cost across 18 months and keeps a 20 percent budget reserve, so the approved baseline cannot exceed $400,000 before contingency.

A group has 350 Power BI users, 18 semantic models, 140 dashboards, and 8 TB of finance plus self-service data; annual capacity spend is $300,000, finance refresh needs 99.9 percent completion by 06:00, and the layout must work for 24 months. Would you isolate workloads across capacities?

I would isolate production finance from development and bursty self-service work, while keeping low-risk domains together. - One reserved production capacity hosts certified finance models and 60 critical dashboards; a smaller pay-as-you-go capacity hosts development and 80 self-service dashboards and can pause outside working hours. - Workspace assignment, deployment pipelines, and a capacity-routing matrix become the control artifacts, with finance maintaining 30 percent headroom during its 04:00 to 06:00 refresh window. - I would approve the split only if both capacities plus operations stay under the total $25,000 monthly limit; otherwise scheduled workload separation on one larger capacity is the cheaper trade-off.

A regional company has 240 BI users, 12 Fabric semantic models, 90 Power BI dashboards, and 4 TB in OneLake; its two-year compute budget is $180,000, production runs continuously while development is used 50 hours per week, and dashboard p95 must stay under 6 seconds. How would you buy and schedule Fabric capacity?

I would reserve steady production capacity and use pausable pay-as-you-go capacity for development. - A vendor quote of $90,000 for two years reserved versus $150,000 pay-as-you-go makes reservation rational for the 24 by 7 production baseline. - Development uses about 2,600 hours per year rather than 8,760; at a quoted $5 per hour it costs $26,000 over two years, provided automation pauses it with no active jobs. - The capacity plan includes a monthly p95 under 6 seconds and 70 percent utilization threshold, leaving $64,000 of the $180,000 budget for short production scale-ups and growth.

BI Developer interview questions

100 real questions with model answers and explanations for Senior candidates.

See a BI Developer resume example →

Practice with flashcards

Spaced repetition · Hunter Pass

Questions

semantic-layerbipower-bi

I would build owned domain semantic models with conformed dimensions and thin reports, not one enterprise model.

Sales and finance each get a star schema with a named BI owner, while Date, Customer, Product, and the approved revenue definition come from versioned shared contracts.
I would target at least 70 of 90 dashboards as thin reports on 6 to 8 certified models; moving from 14 to 8 is a 43 percent reduction in model surfaces, but refresh savings must be measured rather than inferred from that count.
An architecture decision record would cap a model at a clear business grain and 20-minute refresh budget, reserving 10 minutes for ingestion and validation within the $180,000 envelope.

Why interviewers ask this: The interviewer is testing whether you can balance semantic reuse, domain ownership, refresh isolation, and cost at enterprise scale.

capacityslosemantic-layer

I would place domain-owned Direct Lake models over curated OneLake tables and keep cross-domain definitions in explicit shared contracts.

Finance, sales, operations, and customer teams each own model grain, measures, and on-call metadata, while conformed dimensions are published once with compatibility tests.
I would consolidate 18 models to roughly 10 reusable models and keep reports thin; the 44 percent reduction removes eight ownership and release surfaces, while actual CU and support savings remain measured outcomes.
The design record would require p95 visual latency under 5 seconds and data age under 15 minutes, with Capacity Metrics reviewed at 70 percent sustained utilization before spending beyond $260,000.

Why interviewers ask this: A strong answer uses Direct Lake as a serving mode inside an owned semantic architecture rather than treating OneLake as governance by itself.

querieswarehousebigquery

I would use a governed core LookML project plus domain projects that import stable shared definitions.

The core project owns canonical users, dates, accounts, fiscal logic, and 12 board metrics; domain CODEOWNERS approve sales, product, and finance views at their declared grain.
I would reduce 24 overlapping Explores to 10 to 12 curated Explores and measure reused-field coverage above 80 percent, trading some local freedom for fewer conflicting joins.
A project manifest, semantic catalog, and query-cost check would block a merge if p95 exceeds 20 seconds or estimated BigQuery spend raises the $220,000 annual run rate by more than 10 percent.

Why interviewers ask this: The interviewer is evaluating LookML project boundaries, ownership, reuse, and control of BigQuery cost and query latency.

dependenciesbilooker

I would treat the shared LookML project as a versioned contract and migrate consumers through pinned dependency references rather than an unbounded latest version.

The platform owner classifies additive and breaking changes, publishes a release manifest for all 140 fields, and keeps the prior major reference available during a 60-day migration window.
CI tests each of the 6 consuming projects with LookML validation, Content Validator, 30 golden queries, access fixtures, and p95 comparisons before a dependency reference can advance.
The rollout closes when all 110 dashboards use an approved reference, metric parity is exact, p95 remains below 15 seconds, and no consumer remains on the retired major version.

Why interviewers ask this: The interviewer is testing contract versioning and coordinated rollout for shared LookML across independently owned projects.

modelingsnowflakedesign

I would define domain cubes at stable business grains and expose consumer-facing views backed by selective pre-aggregations.

Orders, subscriptions, marketing spend, and support cases get accountable owners, tested joins, and shared dimensions, while views expose only approved measures to the 130 dashboards.
Query history would drive daily and monthly rollups for the top 80 percent of requests; I would not pre-aggregate every combination because storage and refresh cost grow multiplicatively.
A model manifest would require p95 below 3 seconds, pre-aggregation freshness below 15 minutes, and Snowflake plus Cube annual run rate at or below $260,000, with a review at 70 percent of either limit.

Why interviewers ask this: The interviewer is checking Cube modeling, pre-aggregation economics, ownership, and consumer-oriented reuse.

monitoringbipower-bi

I would anchor business grain and additive facts in governed warehouse marts, then keep presentation semantics native to each BI tool.

Versioned contracts define order, customer, fiscal calendar, and currency conversion once; DAX measures and LookML fields may differ only where the interaction model requires it.
A reconciliation suite compares 25 critical metrics across both tools every day and blocks release below 99 percent agreement, with one named domain owner resolving each exception.
Rebuilding both tools around 8 shared marts reduces the count of isolated transformation paths from 20 by 60 percent, but measured compute, cache behavior, and support cost decide whether the design stays inside $600,000.

Why interviewers ask this: A senior BI architect must separate portable business contracts from tool-specific semantic behavior without promising an unrealistic universal model.

semantic-layerbipower-bi

I would align model boundaries to accountable business processes and share dimensions through contracts rather than chained ownership.

Finance owns ledger and close models, sales owns pipeline, and operations owns fulfillment; every model declares grain, steward, support tier, and downstream dashboard count.
Customer, legal entity, product, and fiscal calendar receive versioned keys and reconciliation tests, while cross-domain reports combine certified outputs instead of recreating 22 variants.
I would target 12 to 14 models and retire the rest only after 95 percent of 240 dashboards map to an owner, preserving a 90-minute close refresh window inside the 06:00 SLO and $350,000 budget.

Why interviewers ask this: The interviewer is testing whether model boundaries follow business accountability while conformed dimensions preserve enterprise consistency.

bilookerbigquery

I would use federated domain ownership with a small platform team owning the semantic contract and release path.

Four domain teams own views, measures, dashboard quality, and user support, while the BI platform team owns project templates, shared dimensions, access rules, and production deployment.
CODEOWNERS requires one domain reviewer and one platform reviewer for shared LookML; the 10-day target becomes 3 days for modeling, 3 for validation, and 4 for user acceptance.
A quarterly ownership matrix flags models with no active owner or fewer than 5 monthly users, keeping support within the $160,000 budget instead of centralizing every change in one team.

Why interviewers ask this: The interviewer is evaluating an ownership model that scales LookML delivery without sacrificing reusable controls.

monitoringbipower-bi

I would standardize measure contracts and calculation patterns while leaving facts in domain-sized semantic models.

A metric registry records formula, grain, owner, dimensions, and effective date; reusable calculation groups handle time intelligence and currency rather than copying DAX into 18 models.
TMDL source files and automated dependency scans identify the 150 affected dashboards, with a target of 90 percent of governed measures generated or reviewed from the registry.
The release template budgets 2 days for impact analysis, 2 for model PRs, and 1 for deployment, accepting limited duplication when centralization would breach model size, latency, or the $280,000 capacity limit.

Why interviewers ask this: The interviewer is checking whether you can standardize semantics without confusing code reuse with one physical semantic model.

dashboardsdesignqueries

I would separate internal domain ownership from a curated embedded contract that product teams can depend on.

Domain owners maintain orders, billing, usage, and support cubes, while an embedded analytics owner publishes stable views and a versioned measure catalog for the 80 dashboards.
Shared pre-aggregations cover the 20 query shapes producing at least 80 percent of traffic; tenant-specific rollups are added only when they beat shared rollups on measured latency and storage.
A compatibility suite and monthly cost scorecard block breaking changes, p95 above 2 seconds, or annual run rate above $200,000, trading custom customer fields for a predictable 24-month API surface.

Why interviewers ask this: The interviewer is assessing whether you can make Cube a reusable product contract with explicit ownership, performance, and cost limits.

bipower-bisql

I would choose Import with incremental refresh because a 30-minute freshness allowance does not justify source-bound interactive queries.

Partition the large facts by date, refresh only the latest 2 days every 30 minutes, and keep dimensions fully imported so dashboard queries use the in-memory engine.
Capacity testing must keep refresh within 20 minutes and p95 visuals under 4 seconds; DirectQuery remains an exception for drill-through fields that cannot be retained.
The architecture record compares imported model memory and 48 refreshes per day against Azure SQL concurrency cost, choosing the option that stays below $140,000 per year.

Why interviewers ask this: A strong answer chooses storage mode from freshness, concurrency, model size, and source cost rather than from a preference for live data.

dashboardssemantic-layer

I would make Direct Lake the default for curated Delta tables and isolate unsupported patterns in small Import models.

Direct Lake avoids copying 6 TB into scheduled imports and lets semantic models read OneLake data, but every table must have stable grain, optimized files, and supported security behavior.
A proof with the 20 busiest dashboard pages must hold p95 below 5 seconds and data age below 10 minutes; any DirectQuery fallback or guardrail hit is recorded rather than accepted silently.
Capacity Metrics must remain below 70 percent sustained load within $20,000 per month, and an Import exception is justified when its compression and predictable cache cost less than tuning the lake path.

Why interviewers ask this: The interviewer is testing current Direct Lake benefits and constraints, including the need to validate performance and fallback behavior.

designsemantic-layerbi

I would use DirectQuery for the narrow operational slice and remove historical analytics from that path.

The source gets indexed serving views at dispatch, vehicle, and 5-minute grain, while 12 months of history is imported into a separate model for trend analysis.
I would load-test 180 users at twice expected peak, requiring p95 below 5 seconds and fewer than 30 concurrent source queries after visual and page-query reduction.
Azure SQL and capacity telemetry share the $200,000 annual ceiling; if source compute rises more than 20 percent, an Import aggregation or shorter live window is cheaper than scaling DirectQuery indefinitely.

Why interviewers ask this: The interviewer is evaluating whether DirectQuery is constrained to a genuinely fresh operational workload with source and concurrency controls.

semantic-layerbipower-bi

I would combine imported history and aggregates with DirectQuery only for today's detailed orders.

Dual-mode Date, Store, and Product dimensions serve both paths, an Import aggregate holds store-hour metrics for two years, and DirectQuery covers the current business day.
Aggregation mappings must answer at least 90 percent of 180 dashboard queries from memory with p95 under 4 seconds, leaving live detail as an intentional slower drill-through.
The design document caps the live slice at 50 million rows and total monthly run rate at $25,000, trading 5-minute freshness for more source complexity only where users can act on it.

Why interviewers ask this: The interviewer is testing correct composite-model boundaries, dual dimensions, aggregation hit rate, and the cost of a live detail path.

queriesaggregationwarehouse

I would build Import aggregations from observed dashboard grains and retain DirectQuery detail for exceptional exploration.

Order-day-category-region and seller-hour rollups cover the dominant measures, while relationships and aggregation mappings preserve access to the 30 TB detail table.
Query logs must show at least a 90 percent aggregation hit rate across the top 50 pages and p95 under 4 seconds; unused rollups are removed after 30 days.
Each rollup needs a storage, refresh, and warehouse-scan estimate, and the portfolio is rejected if annualized cost exceeds $450,000 even when another aggregation improves one dashboard.

Why interviewers ask this: A senior answer designs Power BI aggregations from measured query shapes and accounts for refresh and storage cost.

configbipower-bi

I would use incremental refresh for bounded historical partitions and a hybrid DirectQuery partition only for the current hour.

At 500 GB per day, I would not import 24 months of raw transactions; the model keeps only a benchmarked recent detail window plus 24 months of daily merchant aggregates, while raw history remains in the warehouse for drill-through.
RangeStart, RangeEnd, a reliable change-detection column, and partition reconciliation are source-controlled; late-arriving detail partitions refresh daily and the indexed live partition is restricted to the latest 60 minutes.
Capacity and source cost must remain at or below $220,000 per year, and I would remove the hybrid path if fewer than 10 percent of users consume five-minute data or source cost rises 20 percent.

Why interviewers ask this: The interviewer is checking knowledge of incremental-refresh partitions, hybrid real-time data, and when the added complexity is economically justified.

capacitysemantic-layerbi

I would select a reserved F64 capacity plus Pro licenses for 40 authors if the workload proof shows that shared Pro limits are insufficient.

At the quoted prices, Pro for all 400 users is 400 times $14 times 36, or $201,600; PPU is 400 times $24 times 36, or $345,600, so PPU already exceeds the $330,000 cap.
An F64 reservation quoted at $95,000 per year plus author Pro licenses costs $305,160 over three years; viewers still need Fabric Free identities and access to content hosted on F64 or larger capacity.
The decision must verify model size, refresh frequency, and concurrency in a workload test and verify availability against the service contract and recovery drills; Pro wins if its shared-capacity limits pass because it is $103,560 cheaper.

Why interviewers ask this: The interviewer is testing current per-user and Fabric licensing economics together with the technical limits that make the cheapest sticker price insufficient.

capacitysemantic-layerqueries

I would size from a representative concurrency benchmark and Capacity Metrics, not from the 18 TB storage number.

Replay the 30 busiest pages at 1,200 visual queries per minute while scheduled refresh and Direct Lake framing run, then record interactive and background CU consumption by item.
Start with the smallest SKU that keeps p99 interactive delay at zero, visual p99 under 5 seconds, and sustained utilization below 70 percent; test one SKU above it for the growth margin.
The sizing memo compares reserved and scale-up cost across 18 months and keeps a 20 percent budget reserve, so the approved baseline cannot exceed $400,000 before contingency.

Why interviewers ask this: The interviewer is assessing whether capacity sizing uses measured workload behavior, smoothing headroom, and an explicit financial reserve.

capacitysemantic-layerbi

I would isolate production finance from development and bursty self-service work, while keeping low-risk domains together.

One reserved production capacity hosts certified finance models and 60 critical dashboards; a smaller pay-as-you-go capacity hosts development and 80 self-service dashboards and can pause outside working hours.
Workspace assignment, deployment pipelines, and a capacity-routing matrix become the control artifacts, with finance maintaining 30 percent headroom during its 04:00 to 06:00 refresh window.
I would approve the split only if both capacities plus operations stay under the total $25,000 monthly limit; otherwise scheduled workload separation on one larger capacity is the cheaper trade-off.

Why interviewers ask this: The interviewer is testing workload isolation as a reliability decision that must still pass a capacity cost comparison.

capacitysemantic-layerbi

I would reserve steady production capacity and use pausable pay-as-you-go capacity for development.

A vendor quote of $90,000 for two years reserved versus $150,000 pay-as-you-go makes reservation rational for the 24 by 7 production baseline.
Development uses about 2,600 hours per year rather than 8,760; at a quoted $5 per hour it costs $26,000 over two years, provided automation pauses it with no active jobs.
The capacity plan includes a monthly p95 under 6 seconds and 70 percent utilization threshold, leaving $64,000 of the $180,000 budget for short production scale-ups and growth.

Why interviewers ask this: The interviewer is evaluating whether you match reservation and pause economics to distinct steady and intermittent Fabric workloads.

Locked questions

21
A BI platform serves 300 analysts and managers through 14 semantic models and 120 dashboards over 6 TB; annual spend is $280,000, business domains request freshness from 15 minutes to daily, and SLOs must be operational within one quarter. How would you define refresh SLOs?
slosemantic-layerdashboards
22
A BI estate has 450 users, 25 semantic models, 200 dashboards, and 10 TB across Fabric and Looker; annual cost is $500,000, governance reviews happen quarterly, and critical freshness SLO attainment must exceed 99.5 percent. What would you put on the BI scorecard?
slosemantic-layerbi
23
11 semantic models and 80 dashboards serve 220 analysts over 3 TB of customer and billing data; the annual budget is $180,000, freshness must stay under 30 minutes, and source teams plan schema changes over the next 12 months. What data contract would you require?
data-contractssemantic-layerschema
24
A company has 380 BI users, 20 semantic models, 160 dashboards, and 14 TB spanning Fabric and Snowflake; lineage tooling has a $420,000 annual platform budget, critical data must be under 1 hour old, and 90 percent lineage coverage is due in 9 months. How would you build lineage?
lineagesemantic-layercoverage
25
A Power BI team has 160 developers and analysts, 16 semantic models, 110 dashboards, and 5 TB of source data; annual platform cost is $240,000, governed releases must ship weekly, and the CI/CD design must last 18 months. What pipeline would you establish?
ci-cdbipower-bi
26
A Looker deployment supports 190 analysts with 14 LookML models, 85 dashboards, and 4 TB in BigQuery; annual cost is $200,000, approved changes must reach production within 2 business days, and the workflow is planned for one year. How would you design CI/CD?
ci-cddeploymentbi
27
A Power BI organization has 320 users, 18 semantic models, 140 dashboards, and 8 TB of source data; annual cost is $320,000, releases occur weekly with a 30-minute rollback objective, and the setup must serve two years. What should deployment pipelines own, and what must stay outside them?
deploymentrollbackbi
28
A Cube platform serves 140 analysts and 160 application users through 13 data models and 70 dashboards over 9 TB in Snowflake; annual cost is $190,000, p95 must stay under 3 seconds, and semantic releases happen daily for the next year. How would you build observability into CI/CD?
modelingsnowflakeci-cd
29
Six teams support 500 BI users through 28 semantic models and 230 dashboards over 20 TB; annual spend is $600,000, a common semantic standard must be approved within 6 months, and critical reports have a 30-minute freshness SLO. What would your cross-team RFC contain?
slosemantic-layerdecision-making
30
A BI platform serves 400 users with 22 semantic models, 180 dashboards, and 16 TB across Fabric, Looker, and Snowflake; annual cost is $480,000, critical freshness must meet 99.5 percent at 15 minutes, and observability must be delivered in one quarter. What would you instrument?
observabilitybilooker
31
A sales organization has 300 BI users, 16 semantic models, 145 dashboards, and 5 TB in Power BI; annual cost is $300,000, consolidation must finish in 6 months, and critical pipeline data must remain under 30 minutes old. How would you reduce duplication?
bici-cdpower-bi
32
A Power BI tenant serves 420 users with 24 semantic models, 260 dashboards, and 11 TB of data; annual spend is $450,000, zombie-report retirement has a 90-day window, and tier-one dashboards need 99.5 percent monthly freshness. What retirement policy would you implement?
bipower-bidashboards
33
A company has 180 analysts, 10 semantic models, 75 dashboards, and 7 TB in Snowflake; it can spend $250,000 over three years, needs p95 queries under 4 seconds, and must choose within 8 weeks between Cube Cloud and a custom metrics API. What would your build-versus-buy memo recommend?
queriessnowflakeapi
34
A business plans to move 500 users, 25 semantic models, and 300 Tableau dashboards over 20 TB to Power BI and Fabric; the three-year cap is $1.2 million, migration must finish in 18 months, and critical dashboards need 99.5 percent freshness under 1 hour. What would your migration memo propose?
bipower-bitableau
35
A company is considering moving 260 users, 18 LookML models, and 130 Looker dashboards over 9 TB in BigQuery to Power BI; the three-year budget is $700,000, the transition window is 12 months, and dashboard p95 must remain under 8 seconds. How would you decide?
bipower-bilooker
36
A legacy SSAS Multidimensional estate has 14 cubes, 430 MDX calculated members, 190 Excel reports, and 11 TB of data; the company wants Power BI semantic models within 18 months without changing audited finance totals. How would you plan the migration?
excelbipower-bi
37
A Snowflake and Power BI estate serves 400 users through 20 semantic models and 200 dashboards over 40 TB; leaders are considering Fabric, current three-year TCO is $1.5 million, evaluation lasts 18 months, and critical freshness is 15 minutes. Would you migrate the warehouse as well as BI?
warehousesnowflakebi
38
A 620-million-row Power BI model serves 46 reports, but a new calculation-group item pushes p95 from 4 to 28 seconds because it wraps 180 measures and repeats context transitions inside an iterator. What architecture fix do you approve?
bipower-biarchitecture
39
A global company has 450 BI users, 26 semantic models, 310 dashboards, and 18 TB split between Power BI and Looker; current three-year TCO is $1 million, consolidation must finish in 12 months, and critical dashboards have a 30-minute freshness SLO. How would you consolidate the portfolio?
bipower-bilooker
40
A manufacturer must move 200 analysts, 20 Qlik semantic models, and 170 dashboards over 8 TB to Looker; the three-year migration budget is $750,000, completion is required in 15 months, and operations dashboards need 20-minute freshness. How would you structure the migration factory?
bilookermigrations
41
A Power BI tenant serves 500 users in finance, sales, and operations through 18 semantic models and 160 dashboards over 12 TB; annual cost is $400,000, entitlement changes must apply within 4 hours, and the security design must last two years. How would you architect multi-domain RLS?
designsemantic-layerbi
42
A company has 320 Power BI users, 14 semantic models, 120 dashboards, and 5 TB of finance and HR data; annual cost is $300,000, access changes must appear within 24 hours, and monthly-close reporting must remain available for three years. How would you combine RLS and OLS?
bipower-bidashboards
43
A Looker instance serves 260 users through 16 LookML models and 100 dashboards over 7 TB; annual spend is $250,000, entitlement updates need a 30-minute SLO, and the architecture must work for 18 months. How would you enforce multi-domain access?
bilookerarchitecture
44
A Cube platform serves 60 internal analysts and 400 customer users through 12 data models and 90 embedded dashboards over 10 TB; annual cost is $350,000, tenant-scoped queries need p95 under 2 seconds, and the isolation model must last three years. How would you enforce tenant isolation?
queriesmodelingdashboards
45
A BI program has 500 users, 24 semantic models, 220 dashboards, and 15 TB across Fabric and Looker; annual cost is $500,000, analysts need to publish useful content within 5 business days, and the governance model must be in place within 12 months. How would you govern self-service?
bilookerdashboards
46
A Power BI organization has 360 users, 20 semantic models, 150 dashboards, and 8 TB of source data; annual platform spend is $300,000, certification is reviewed quarterly, and critical model freshness must exceed 99.5 percent. How would you manage endorsed semantic models?
bipower-bidashboards
47
A Fabric and Power BI estate serves 300 users through 15 semantic models and 130 dashboards over 6 TB; the three-year resilience budget is $400,000, business requires an 8-hour RTO and 24-hour RPO, and the recovery design must be tested twice a year. What would you implement?
designsemantic-layerbi
48
A Looker deployment supports 180 analysts through 12 LookML models and 80 dashboards over 4 TB in BigQuery; the two-year resilience budget is $250,000, RTO is 4 hours, RPO is 1 hour, and recovery must be rehearsed quarterly. How would you protect the semantic layer and content?
deploymentbilooker
49
A BI estate serves 420 users through 25 semantic models and 190 dashboards over 13 TB; annual cost is $500,000, breaking metric changes require 30 days notice, and the change policy must work for 18 months. How would you manage semantic-layer evolution?
dashboardsmonitoringsemantic-layer
50
An enterprise has 480 users, 30 semantic models across Power BI, Looker, and Cube, 240 dashboards, and 25 TB of data; annual cost is $700,000, governed changes release quarterly, and the operating model must remain stable for two years. How would you coordinate semantic change across tools?
bipower-bilooker
51
At 06:20, the global sales model has missed its 06:00 refresh SLO after one partition fails, 1,800 representatives are opening it, and a full rerun is estimated at 52 minutes. What do you do?
estimationpartitioningslo
52
A finance model has missed its 07:00 refresh SLO on 4 of the last 10 business days, with completion ranging from 07:18 to 08:05 and payroll approval starting at 07:30. What change do you approve?
slo
53
At 09:05, a Microsoft Fabric F64 capacity reaches 92% sustained utilization, interactive report p95 rises from 6 to 48 seconds, and 700 executives and managers are active. How do you respond?
capacity
54
A Fabric F128 capacity starts throttling at 08:00 because 140 scheduled refreshes overlap with 2,400 morning viewers, and the CFO dashboard takes 75 seconds. What do you change today?
dashboardscapacity
55
An XMLA deployment at 14:10 makes a shared Power BI semantic model unavailable, breaking 86 reports used by 4,200 employees before a 15:00 operating review. Do you roll back or fix forward?
deploymentrollbackbi
56
A column rename in the customer dimension causes a Tableau published data source and a Looker model to fail simultaneously, taking 63 executive dashboards offline at 07:40. How do you recover?
bitableaulooker
57
At 05:50, all three nodes in an on-premises data gateway cluster fail outbound TLS after the enterprise proxy root certificate expires, blocking 47 Power BI refreshes due by 07:00. What is your incident plan?
proxygatewaybi
58
One gateway node starts rejecting Snowflake connections after a firewall change, causing 19 of 120 refreshes to fail intermittently while the other nodes look healthy. How do you contain it?
gatewaynetworkingsnowflake
59
A Tableau Server deployment serving 3,600 users loses two VizQL processes and its Backgrounder queue grows to 240 jobs, pushing workbook load p95 to 58 seconds. What do you do?
deploymentbitableau
60
A Looker production deploy causes connection pool exhaustion and 95% of Explores to return errors for 1,100 users, while 34 PDT builds are still running. What do you do?
deploymentbilooker
61
The CEO sees $128.4 million ARR in Power BI, the CFO sees $124.9 million in Tableau, and Sales sees $131.2 million in Looker two hours before the board pack is locked. How do you resolve it?
bipower-bitableau
62
Two executive dashboards report gross margin as 31.8% and 28.6% because one model allocates cloud cost by revenue and the other by usage; a $20 million investment vote is tomorrow. What do you decide?
dashboardscss
63
A fiscal-calendar update shifts weekly revenue by 9% on 14 executive dashboards, while 22 operational dashboards still use the old week boundary and quarterly guidance is due in 24 hours. What do you do?
dashboards
64
An RLS regression lets 214 regional managers view compensation totals for all 18 regions for 37 minutes in Power BI. What is your response?
bipower-birow-level-security
65
A deployment removes OLS from four payroll columns, exposing national ID and bank-account fields to 62 report authors for 3 hours even though no dashboard displays them. What do you do?
dashboardsdeploymentschema
66
A Power BI Embedded token bug lets 17 customers open another tenant's usage dashboard for 11 minutes, affecting up to 46,000 account rows. How do you lead the incident?
bipower-biincidents
67
A Looker access_filter change exposes one retailer's inventory to 8 users from a second retailer for 26 minutes, and two CSV downloads occurred. What do you do?
bilooker
68
A dbt model rename breaks lineage for 63 certified dashboards, but cached results continue to display for 9 hours and no alert fires. How do you restore trustworthy lineage?
alertingdbtcaching
69
A corrected sales-credit rule reduces quarterly commission by $2.4 million across 380 employees after payroll files were approved but before payment. What do you do?
70
Internal Audit asks why 126 salespeople were overpaid $740,000 across two quarters and wants reproducible evidence in 5 business days. How do you respond?
reproducibility
71
A Tableau-to-Power BI migration has moved 140 of 220 dashboards, but 18 percent of sampled KPIs differ and 600 active users still generate 70 percent of weekly sessions in Tableau three weeks before 2,400 licenses expire. What do you do?
bipower-bitableau
72
A consolidation from Power BI, Looker, and Qlik to Looker is six months late, has spent $1.8 million, and only 22% of 900 users have moved. What recommendation do you make?
bipower-bilooker
73
A DirectQuery rollout moves 75 reports to Snowflake, but morning p95 rises from 9 to 64 seconds and warehouse credits triple with 1,300 concurrent users. What do you do?
warehousesnowflakeconcurrency
74
After moving 28 semantic models to Direct Lake, 11 fall back to DirectQuery, Fabric CU peaks at 96%, and executive report p95 reaches 41 seconds. How do you recover the rollout?
semantic-layer
75
A Looker refactor replaces 420 duplicated measures with 38 reusable views, but the first deploy breaks 27 Explores and 160 scheduled deliveries. How do you proceed?
deploymentbilooker
76
A reusable LookML measure refactor passes compilation but changes net revenue by 2.7% across 84 dashboards because filter inheritance changed. What do you do?
refactoringownershipoop
77
Your BI vendor raises annual licensing from $1.4 million to $2.35 million with 90 days' notice, while 2,800 users depend on 760 dashboards. What do you recommend?
dashboardsprocurement
78
A Looker contract changes embedded-view pricing and increases forecast cost by 42% for 65 customers and 9 million monthly sessions. How do you respond?
bilookersessions
79
A CFO refuses to retire a 9-year-old dashboard used by 14 people because it supported a past audit, although it costs $86,000 annually and duplicates a certified report. What do you do?
dashboards
80
Operations blocks retirement of 46 legacy Tableau workbooks because peak season starts in 8 weeks, but only 11 have weekly users and security support ends in 12 weeks. What plan do you approve?
bitableau
81
Nine hundred users open 380 Power BI reports each Monday, pushing p95 to 33 seconds and causing 12% visual errors on a shared F128 capacity. What architecture change do you make first?
capacitybipower-bi
82
At 09:00, 350 Looker users run 1,800 queries in 10 minutes, the database connection pool reaches 100%, and 70 scheduled deliveries miss their window. What do you change?
databasequeriespooling
83
One self-service Looker Explore generates 28 percent of Snowflake spend and has queries running up to 19 minutes across 240 users. What query budget do you enforce?
queriessnowflakebi
84
A single Power BI report consumes 31% of F64 CU during month-end, with 42 visuals and six measures each taking more than 8 seconds. How do you govern it?
bipower-bimeasures
85
Four hundred eighty semantic-model refreshes must finish between 01:00 and 06:00, but the current window ends at 07:25 and delays 38 country dashboards. How do you redesign the window?
dashboards
86
BI warehouse spend rises from $120,000 to $230,000 per month while active users grow only 6%, and 54% of cost has no dashboard owner tag. What do you do?
dashboardswarehouse
87
Tableau extract storage reaches 94 TB, daily refresh uses 68% of the maintenance window, and 1,600 users still need 15-minute morning freshness. What do you change?
bitableau
88
A Looker estate runs 62 dashboards over 40 TB in BigQuery; scan cost rises 68 percent, p95 reaches 24 seconds, 38 percent of queries miss partition filters, and freshness must stay under 30 minutes. What architecture change do you make?
queriespartitioningbigquery
89
An administrator deletes a production Fabric workspace with 12 semantic models and 74 Power BI reports; source data still exists, the business claims a 4-hour RTO and 1-hour RPO, and the deployment pipeline no longer has a production stage. Can the pipeline restore it?
deploymentbici-cd
90
Four business units consume $1.1 million of annual BI capacity, but one unit drives 57% of CU and rejects chargeback because its dashboards are company-wide. What allocation do you implement?
dashboardscapacity
91
Finance wants one enterprise revenue measure, four product teams require contract-specific variants that create an $18 million annual difference, and the RFC is blocked after 28 comments and 6 weeks. What decision model and migration do you impose?
measuresmigrationsdecision-making
92
Security mandates Power BI for identity control, Data Science wants Sigma for warehouse-native analysis, and Procurement will fund only two BI tools for 3,200 users. How do you close the RFC?
bipower-bidecision-making
93
A self-service program produces 73 uncertified semantic models and 260 dashboards in one quarter; 19 KPI conflicts reach executives and support tickets rise 64%. What do you change?
dashboardskpisemantic-layer
94
The certified badge appears on 44 Power BI dashboards even though 17 have no owner, 9 miss freshness SLOs, and 6 use local measures outside the semantic layer. What do you do?
slosemantic-layerbi
95
An upstream customer contract removes country_code with 24 hours' notice, breaking RLS and tax logic in 31 semantic models used across 12 markets. What do you do?
row-level-securitysemantic-layer
96
At 08:40, the COO asks you to remove the stale-data banner from a 7-hour-old operations dashboard before a lender call because the numbers look bad. What do you do?
dashboards
97
The CEO asks you to publish a 14.2 percent growth figure even though the certified metric is 9.6 percent and the larger number excludes $4.8 million of refunds; the earnings draft closes in 90 minutes. What do you do?
monitoring
98
A mid-level BI developer must choose the pilot domain for a Fabric migration affecting 90 models, but stakeholders provide conflicting priorities and no model has complete lineage. How do you mentor them?
mentoringcommunicationmigrations
99
During a capacity incident affecting 2,000 users, a mid-level engineer wants to resize immediately, but evidence for the 3-fold CU spike is incomplete and the CFO dashboard is failing. How do you coach and decide?
dashboardsincidentscapacity
100
You inherit 520 dashboards across Power BI and Tableau, 46 semantic models, 1,400 Microsoft 365 users, 80 percent of active sessions already in Power BI, a $1.9 million annual budget, 30-minute executive freshness, and a mandate to cut cost 25 percent in 12 months. What architecture do you choose?
bipower-bitableau

Questions

2A Fabric estate serves 320 analysts through 18 semantic models and 140 Power BI dashboards on 8 TB in OneLake; it has a $260,000 annual capacity budget, a 15-minute freshness SLO, and an 18-month planning horizon. How would you organize the Direct Lake semantic layer?

3A company has 180 analysts, 12 LookML models with 24 Looker Explores, 75 dashboards, and 3 TB in BigQuery; annual Looker and warehouse spend is $220,000, dashboard queries need p95 under 20 seconds, and the models must support the next year. How would you structure the LookML projects?

4Six Looker domain projects import a shared LookML project with 140 governed fields; a base-view change must reach 110 dashboards without another breaking rollout, and interactive p95 must remain under 15 seconds. How would you version and release the dependency?

5A Cube deployment serves 450 analysts and managers through 16 Cube data models and 130 embedded dashboards over 12 TB in Snowflake; the annual budget is $260,000, p95 response time must be under 3 seconds, and the design horizon is 18 months. How would you shape the semantic layer?

6An enterprise has 380 BI users, 20 semantic models split between Power BI and Looker, 200 dashboards, and 25 TB in Snowflake; total BI spend is $600,000 per year, cross-tool metrics need 99 percent agreement, and the target state is due in 12 months. Where should canonical semantics live?

7Five business units have 500 Power BI users, 22 semantic models, 240 dashboards, and 4 TB of imported data; annual cost is $350,000, finance close dashboards must be ready by 06:00, and the architecture must serve two years. How would you set model boundaries and ownership?

8A Looker organization supports 140 analysts with 12 LookML models, 65 dashboards, and 2 TB in BigQuery; annual spend is $160,000, a new domain must publish governed analytics within 10 business days, and the operating model must last 18 months. Who should own what?

9A Power BI tenant has 300 users, 18 semantic models, 150 dashboards, and 9 TB of source data; annual cost is $280,000, metric changes must reach all governed reports within 5 business days, and the roadmap spans two years. How would you increase measure reuse without building a giant model?

10A SaaS company serves 80 internal analysts and 400 customer users through 14 Cube models and 80 embedded dashboards over 5 TB; annual platform cost is capped at $200,000, p95 queries must stay under 2 seconds, and the product horizon is 24 months. How would you design ownership and reuse?

11You have 260 Power BI users, 9 semantic models, 70 dashboards, and 1.8 TB in Azure SQL; annual cost is limited to $140,000, data may be 30 minutes old, and the decision must hold for 18 months. Would you choose Import or DirectQuery?

12A Fabric program serves 350 users with 12 semantic models and 100 dashboards over 6 TB of Delta tables in OneLake; it has a $240,000 annual F64 budget, needs 10-minute freshness, and is planned for two years. Would you use Direct Lake or Import?

13A dispatch operation has 180 users, 6 Power BI semantic models, 45 dashboards, and 900 GB in Azure SQL; annual cost is $200,000, operational data must be under 2 minutes old with p95 visuals below 5 seconds, and the design horizon is one year. Is DirectQuery appropriate?

14A retailer has 420 Power BI users, 15 semantic models, 180 dashboards, and 7 TB of order data; the annual budget is $300,000, today's sales need 5-minute freshness while older data may refresh daily, and the architecture must last 24 months. How would you use a composite model?

15A marketplace serves 500 Power BI users through 10 semantic models and 220 dashboards querying 30 TB; annual warehouse and BI cost is $450,000, p95 visuals must be under 4 seconds with 15-minute freshness, and the plan covers 18 months. How would you design aggregations?

16A payments team has 300 Power BI users, 8 semantic models, 95 dashboards, and 12 TB growing by 500 GB per day; annual cost is capped at $220,000, the latest hour needs 5-minute freshness while history may be daily, and the horizon is one year. How would you configure incremental refresh?

17A company has 40 Power BI authors and 360 viewers, 20 semantic models, 160 dashboards, and 5 TB of source data; it can spend $330,000 over three years, requires 99.5 percent availability, and expects stable headcount. Would you license Pro, Premium Per User, or Fabric capacity?

18A Fabric tenant serves 480 users with 25 semantic models, 210 dashboards, and 18 TB in OneLake; the 18-month capacity budget is $500,000, 99 percent of visuals must render within 5 seconds, and peak use is 1,200 visual queries per minute. How would you size capacity?

19A group has 350 Power BI users, 18 semantic models, 140 dashboards, and 8 TB of finance plus self-service data; annual capacity spend is $300,000, finance refresh needs 99.9 percent completion by 06:00, and the layout must work for 24 months. Would you isolate workloads across capacities?

Locked questions