I Reviewed 19 Power BI Portfolios for Dubai Clients — the 7 Steps That Stopped Me Hiring Dashboard Decorators

Power BI is the easiest technical role in the UAE market to hire badly, because the artefact candidates show you is the one part of the job that does not matter much. A screenshot of a dark-themed dashboard with well-chosen KPIs tells you the candidate has taste. It tells you nothing about whether the numbers on it are right, whether the model behind it will survive a second entity being added, or whether the refresh will still finish before the finance team arrives in the morning. Here is the process I use instead, built after a run of 19 portfolio reviews for Dubai clients in which the three best-looking submissions were the three I would not have hired.

Step 1 — Decide Which of the Four Power BI Jobs You Are Hiring For

“Power BI developer” covers four distinct jobs. Most employers I talk to need the middle two and write an advert for the first.

• Report builder. Takes an existing, trusted model and produces reports and visuals against it. Valuable, fast to find, and the cheapest of the four.
• Data modeller. Shapes the tables, relationships, grain and measures so that the answers are correct and the same question returns the same number in two places. This is where most of the risk lives.
• Semantic layer owner. Owns the definitions themselves — what “active customer” and “recognised revenue” mean — and the governance around changing them. Half technical, half political.
• Platform administrator. Capacity, workspaces, deployment pipelines, gateways, tenant settings. Infrastructure work that happens to use Power BI vocabulary.

Write down which one you are hiring. If the honest answer is “all four”, you are hiring a senior and you should price accordingly, because the people who can genuinely do all four are a small group and they know it.

Step 2 — Specify the Data Sources, Not the Tool

Every Power BI advert in the region lists the same skills, which is why none of them filter anyone. The three facts that actually filter are: where the data comes from, how fresh it has to be, and who owns the definitions.

Name the systems. “Dynamics 365 Finance, two Oracle instances from the acquisition, a Google Sheet the sales team maintains, and a monthly bank statement export” is a specification. “Experience with various data sources” is not. The candidate who has reconciled two ERP instances into one model will self-identify immediately, and the one who has only ever connected to a clean warehouse will be honest about it if you ask directly.

State the refresh window as a deadline rather than a frequency. “Daily” means nothing; “complete by 07:00 Gulf Standard Time, source systems are only available after 02:00” is a five-hour engineering constraint that immediately changes the conversation about incremental refresh and about whether you need a gateway at all.

If you are standing up the underlying pipelines at the same time, that is a separate hire with a separate pool — our process for hiring data engineers in Dubai covers that side, and the FAQ below explains why mixing the two roles is the most expensive mistake in this category.

Step 3 — Settle Capacity and Licensing Before You Interview

This step gets skipped and it causes more post-hire friction than any technical gap. The licensing and capacity model you land on determines what the role actually is.

On per-user licensing with no dedicated capacity, your new hire will spend a meaningful share of their time working around dataset size and refresh limits, and the skill you need is pragmatic model optimisation. On dedicated capacity, the same person inherits a platform administration job they may not want and may not be good at: workload management, monitoring, deployment pipelines, and arguments about who gets to publish. Those are different people, or at least different conversations.

Decide, write it in the specification, and say it out loud on the first call. Candidates who have administered capacity will tell you quickly whether they enjoyed it. That answer is worth more than their certification list.

Step 4 — Source Against the UAE Reality

Searching for the job title returns the shallowest part of the market. The four pools that actually produce good hires here are:

1. ERP implementation consultancies. People who have spent three years extracting reports out of Dynamics, SAP or Oracle for UAE clients understand group structure, intercompany entries and fiscal calendars without being taught. Many are ready to move in-house for better hours. If your stack is ERP-centric, our guide to hiring an SAP or ERP developer in Dubai maps the same pool from the other direction.
2. Group finance MIS teams. The analyst who has been producing the monthly pack by hand for two years knows exactly which numbers are contested and why. They often have stronger DAX than you expect and weaker engineering habits than you need, which is a coachable combination.
3. Audit and advisory analytics practices. Strong on reconciliation discipline and documentation, usually weaker on performance tuning. Excellent if your main risk is numbers disagreeing.
4. Self-taught operations analysts. The highest-variance pool. The best are outstanding because they built something that was genuinely used; the worst have assembled a portfolio from tutorials. Step 5 is designed to tell them apart in under half an hour.

One sourcing note specific to this market: a large share of strong candidates are already in the UAE on a visa sponsored by an employer, and your timeline is governed by notice periods and transfer logistics rather than by interest. Build that into the plan before you promise a start date.

Step 5 — Run the 25-Minute Data Model Screen

This is the step that does the work. Twenty-five minutes, no slide deck, no take-home yet, and four questions in a fixed order. I have run it often enough to know that the answers separate the pools from Step 4 almost immediately.

Question 1 — “Describe the shape of the last model you owned”

Not what it reported on. Its shape. You are listening for a fact table, a grain, conformed dimensions and a date table they built deliberately. A modeller answers this in under a minute and usually draws it. A decorator describes the pages of the report instead, or says “I imported the extract and built relationships as I needed them”. Both answers are informative and neither is disqualifying on its own — but you now know which conversation you are in.

Question 2 — Hand them a measure and ask what is wrong with it

Write a deliberately broken measure on a whiteboard before the call. My standard one divides a sum of margin by a sum of revenue inside an iterator, so it produces the right number at the row level and a meaningless average at the total. Ask them what the total will show and why. Candidates who have been burned by this explain filter context without being prompted. Candidates who have only ever dragged fields will tell you the formula looks fine.

Question 3 — “Six legal entities, two free zones, one mainland. How do you stop a finance manager seeing another entity?”

This is the UAE-specific question and it is the single most predictive one in the set. You want to hear about a security table, a mapping of users to entities maintained outside the report, dynamic row-level security expressions, and — the detail that separates experience from theory — what happens to the totals a restricted user sees in a consolidated visual. Anyone who has shipped this knows that the consolidation line is where the argument starts.

Question 4 — “The refresh takes four hours and finance arrives at eight. Where do you look first?”

Good answers start with what is actually being pulled: query folding broken by a transformation applied too late, a full reload where an incremental policy belongs, calculated columns doing work that belongs upstream, a gateway on an overloaded machine. Weak answers go straight to “upgrade the capacity”. The difference between those two instincts, over a year, is a five-figure licensing decision.

Score it crudely: two or more convincing answers and the candidate goes to Step 6. One, and they are a report builder, which may still be exactly what you need if Step 1 said so. Zero, and no portfolio should change your mind.

Step 6 — Set a Paid Take-Home on Your Own Messy Extract

Every portfolio I reviewed was built on clean data. None of your data is clean. So the take-home is not a modelling exercise with a tidy CSV — it is a smaller version of your actual problem.

Take a real export from the system the role will depend on. Leave the duplicates in. Leave the entity names inconsistent — the same company spelled three ways is normal in UAE group data and it is exactly the thing that quietly breaks a consolidation. Strip or hash anything commercially sensitive, then set the brief in one paragraph: build a model that answers these three questions, document your assumptions, four hours maximum.

Pay for it. At four hours, paying a professional rate costs less than one wasted interview panel, it roughly doubles completion rates among employed candidates, and it changes the tone of the conversation that follows. In this market, where strong candidates are already employed and juggling notice periods, an unpaid take-home filters for availability rather than ability.

Grade it in this order, and tell the candidate the order in advance:

1. Assumptions document first. Did they notice the duplicates, the three spellings, the missing month? Did they write down what they decided and why? A candidate who flags five data problems and resolves three has done better work than one who silently produced a clean-looking answer.
2. Model shape second. Grain, relationships, a proper date table, no bidirectional relationship added to make a visual behave.
3. Measures third. Do the three answers survive being sliced by entity and by month?
4. Visuals last, and briefly. You are checking that it is legible, not that it is beautiful. If you catch yourself admiring the colour choices, go back to item one.

One practical warning: do not accept a submission that uses a different dataset than the one you sent. It happens, and it always means you are looking at something they built earlier.

Step 7 — Tie Probation to a Replaced Process, Not a Delivered Dashboard

The default 90-day goal for this role is “deliver the executive dashboard”. It is the wrong target, because a dashboard can be delivered and ignored, and because it rewards the visual layer again.

Set the target as a manual process that no longer happens. The monthly pack that somebody rebuilds by hand in Excel over three days. The weekly receivables email. The consolidation that requires a person to remember which rate table to use. Pick one, name it in the offer, and measure the outcome in hours returned to the finance team and in whether the manual version has actually stopped being produced.

That framing does three things. It forces the new hire to talk to the people who own the numbers, which is where the real definitions live. It makes success visible to someone outside the BI function. And it surfaces a bad hire early: a decorator will present a beautiful replacement that finance quietly keeps double-checking by hand, and you will know by week eight.

A 90-day plan that works on this role looks roughly like: weeks one to three, shadow the manual process and document the definitions; weeks four to seven, build the model and reconcile it line by line against the hand-built version; weeks eight to eleven, run both in parallel and close the gaps; week twelve, switch off the manual one with the owner’s agreement. If you want the general structure for this market, our Dubai data engineering hiring guide uses the same parallel-run pattern for pipeline roles, and the compensation bands below assume it.

Where This Differs Outside the UAE

The seven steps travel. The constraints do not. Group structure across free zones and mainland, the housing allowance share of package, visa transfer timelines and bilingual Arabic reporting are all specific to this market, and they are the four things that make a Dubai BI hire slower than the equivalent hire elsewhere. If your roadmap also covers Southeast Asia, our Singapore team’s process for hiring engineers who handle multilingual APAC data covers the same consolidation problem under completely different compensation and pass rules, and their fintech engineering team guide is the better reference if your reporting sits behind a regulated product.

Frequently Asked Questions

What does a Power BI developer cost in Dubai in 2026?

In the roles we have benchmarked and placed, a Power BI developer with two to four years of experience who is mainly building reports generally sits between 12,000 and 20,000 AED per month in base salary. Someone who owns the data model and the semantic layer for a group, and can be trusted with row-level security across legal entities, sits between 20,000 and 32,000 AED per month. A candidate who also administers Fabric capacity, manages deployment pipelines and handles gateway infrastructure prices above that, mostly because the pool is small. As with every UAE package, base understates cost: housing allowance commonly adds 4,000 to 9,000 AED per month at these levels, and you should budget separately for medical cover, annual flights and end-of-service gratuity. Banking, group holding companies and government-related entities pay a premium over retail and hospitality, largely because of audit and compliance exposure.

Is a Power BI developer the same thing as a data engineer?

No, and conflating them is the most expensive specification error we see in this role. A data engineer builds and operates the pipelines that land data somewhere queryable, works mostly in SQL, Python and orchestration tools, and is measured on reliability and latency. A Power BI developer turns landed data into a model that answers business questions correctly, works mostly in Power Query, the tabular model and DAX, and is measured on whether two reports agree with each other. Hiring a data engineer to own a semantic layer usually produces technically clean pipelines feeding a model nobody trusts, and hiring a report builder to own ingestion produces fragile manual refreshes. If you genuinely need both, hire the modelling profile first unless your data is not landing at all.

What makes UAE reporting requirements different for a BI hire?

Four things show up in almost every brief we handle. First, group structure: most UAE businesses of any size operate several legal entities across free zones and mainland, which means consolidation rules, intercompany elimination and row-level security that respects entity boundaries. Second, multi-currency: AED is pegged to the dollar but reporting still has to handle transactional currencies and a chosen rate convention, and getting the rate table wrong is the most common source of two reports disagreeing. Third, calendars: the fiscal year is not always January to December, Hijri dates appear in some government reporting, and weekend definitions have changed within living memory of your historical data. Fourth, bilingual output: Arabic labels with right-to-left rendering in the same report as English is a normal requirement, and it is a layout problem the candidate either has solved before or has not.

How does the UAE e-invoicing mandate affect this hire?

It changes the data you will be reporting on, and the timing is close enough to matter for a hire made now. Under the UAE Federal Tax Authority programme, large taxpayers with annual revenue of AED 50 million or more must appoint an accredited service provider by 30 October 2026, with mandatory business-to-business and business-to-government e-invoicing from 1 January 2027, and other VAT-registered businesses following from 1 July 2027. Invoices move to structured formats exchanged through accredited providers and reported to the authority within a defined window, with long retention obligations. The practical consequence for a BI hire is that your invoice data is about to become more structured, more complete and more auditable than the exports your current reports were built on. A candidate who asks what your e-invoicing readiness looks like is thinking about the next eighteen months. One who does not is thinking about this quarter.

Panos Petropoulos

Web Development Expert at HireDeveloper.ae. Reviews technical portfolios and runs the model and architecture screens for data and web roles across UAE employers, and keeps a list of the questions that candidates cannot prepare for.