Hiring a Data Intelligence Engineer: What Evidence Matters?

Hiring a data intelligence engineer starts with deciding who owns the meaning, verification, and permitted use of an AI-generated answer. Define that responsibility before choosing the title. Then ask candidates to investigate a plausible wrong answer using supplied business definitions and synthetic data, so you can examine how they establish what is trustworthy.

Picture a familiar disagreement. Sales has one recurring-revenue figure. Finance has another. An assistant produces a confident explanation and a query that runs successfully. Someone still has to determine whether the query answers the right question, whether its evidence holds, and who can approve the number for a board presentation.

That is a concrete hiring problem. A new title can help describe it, but the title cannot settle it.

What does “data intelligence engineer” mean here?

Jordan Tigani proposes the role in his November 11 MotherDuck article. His version brings together business context, data models, agent coordination, known-answer evaluations, and responsibility for consequential outputs. The linked proposed job description also allows for existing data engineers, analytics engineers, or analysts to take on the responsibility.

Treat this as a vendor’s proposal about how data work could change. It is useful material for designing a job, not evidence that a standard new occupation has arrived. The title already appears in other job listings with different responsibilities. You need to explain what it means at your company.

The useful part of Tigani’s argument is the accountability question. If an agent can generate a query and an explanation, your team still needs someone who can connect those outputs to a business definition, find exceptions, and establish checks. That is a hiring inference from the proposal, rather than a measured claim about which candidates will succeed.

A recognizable title may serve you better. You might advertise for an analytics engineer and make ownership of AI-assisted answers explicit. You might assign the responsibility to an existing colleague. The decision depends on the work and authority you need, rather than enthusiasm for the name.

Define the responsibility before opening the vacancy

Write down the decisions this person will make and the decisions they must escalate. A broad promise to “own data quality” gives candidates too little information and interviewers too much room to invent their own standards.

Start with a small charter:

This is a starting charter you can adapt, not an industry standard. If finance owns a revenue definition, the engineer should not quietly rewrite it to make a dashboard agree with sales. They should identify the conflict, explain its effect, and get a decision from the appropriate owner.

The need for shared meaning predates agents. For example, dbt’s Semantic Layer documentation describes centrally defined metrics as a way to reduce inconsistent calculations. You do not need that product to make the same organizational choice: establish which definition applies and who can change it.

Separate entry requirements from training. If you expect to teach your subscription model, provide its rules in the exercise. If diagnosing SQL errors is an essential starting skill, ask candidates to demonstrate it. The US Office of Personnel Management’s work-sample guidance makes this distinction between competencies expected on entry and those taught later. This is federal assessment guidance, not a universal legal requirement or validation of the exercise below.

Put one plausible wrong answer in the work sample

Give candidates an answer worth challenging, plus enough context to investigate it. A small discrepancy can reveal more about verification than an open-ended request to build an impressive analytics demo.

The following exercise is fictional. Every account and amount is invented. Its definitions are instructions for the task, not universal accounting rules.

Your brief says that monthly recurring revenue, or MRR, is the monthly amount from recurring subscriptions active at the specified month-end. Exclude one-off setup charges and subscriptions cancelled before that date. For this exercise, annual recurring revenue, or ARR, equals MRR multiplied by twelve.

Supply this synthetic packet:

Include a flawed generated query that sums every recurring amount and setup charge without filtering subscription status. Its answer is MRR of $650 and ARR of $7,800. Add a note that sales has been using that larger figure.

Keep the known answer in the assessor’s packet: MRR is $400 and ARR is $4,800 under the supplied rules. Account A’s setup charge and account B’s cancelled subscription explain the discrepancy. Do not make the exercise depend on guessing an undisclosed finance convention.

Ask the candidate to provide a corrected calculation, a short explanation of the error, checks that would catch it again, and a handoff stating whether the answer is ready to use. They should distinguish facts established by the packet from questions that would require a business decision. Sales preferring the larger number does not change the approved definition.

For a technical role, provide a small executable dataset and query in a familiar SQL environment. State any required dialect knowledge beforehand. For a role centered on analytical ownership, decide whether SQL implementation is necessary or whether a reproducible explanation meets the requirement. Assess the job you intend to offer.

Make tool use clear, too. If AI assistance is allowed, provide comparable access or a route that requires no personal subscription. Ask what the candidate checked after using it. The useful evidence is how they tested the result, including whether a second generated explanation simply repeated the first error.

Agree what counts as evidence before reviewing candidates

Use observable behavior to anchor the review. A polished narrative and the correct final number are both useful, but neither tells you how the candidate reached the answer.

Agree on criteria before running the exercise:

These are proposed review anchors, not a validated hiring instrument. Pilot the task with people who understand the actual job. Check whether the instructions are ambiguous, the effort is reasonable, and the required skills match the vacancy. Tell candidates the expected effort, submission format, and accommodation route; reconsider or compensate substantial work as appropriate to your process.

Have reviewers record their observations independently before discussing the candidate. “Excluded B because its cancellation preceded month-end” gives your team something concrete to examine. “Strong business instincts” leaves everyone guessing what happened.

Use the follow-up to explore the limits of the answer. Ask what changes if a cancellation becomes effective next month, how a newly introduced charge type would enter the checks, or what happens when finance disputes the definition. Give every candidate the same core follow-up conditions. You are examining how they adapt their reasoning, not rewarding whoever guesses the interviewer’s hidden exception.

A score should help organize that discussion without becoming an unexplained hire-or-reject rule. Record what remains uncertain and whether another job-related question could resolve it.

Decide who owns the answer after hiring

The responsibility needs operating conditions as well as a capable person. Someone must approve definitions, grant appropriate access, receive escalations, and decide when an answer can be used.

Before making an offer, explain how those relationships will work. Who resolves a sales-finance disagreement? Who can pause a recurring report? Who corrects an already distributed number? What happens when a check fails while its owner is away?

The exercise can test whether a candidate notices these questions. It cannot compensate for your organization leaving them unanswered. Nor can a tiny synthetic dataset prove that someone can manage your entire data system. Use it as one bounded piece of evidence alongside relevant experience and structured discussion.

Keep the hiring evidence together in Kit

Turn the charter into the job description, interview stages, and review criteria your team will actually use. Kit’s ideal candidate profiles can help connect expected outcomes and competencies to evaluation. Its process template configuration includes Work Sample stages for a brief, candidate submissions, and assigned human reviews.

Use those records to keep the task conditions and each reviewer’s reasoning available for the hiring discussion. Kit organizes the process; your assessors still establish the expected answer, examine the work, and decide what it supports. It does not determine analytical correctness for you.

Start with one consequential answer your business needs to trust. Name its definition, checks, and decision owner. You will have a clearer vacancy and a more useful interview, whether you ultimately call the person a data intelligence engineer or something your candidates already recognize.