ECBS5256 – Managing Data Science Teams

Hiring & Team Formation

Day 1 — Block B | 13:30–15:10

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Learning Outcomes

By the end of this block you will be able to:

  • Produce a structured hiring packet that improves signal and fairness
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Why Hiring Is Your Highest-Leverage Activity

  • A great analyst makes every PM smarter — they reframe questions and surface data no one asked for
  • A great data engineer makes every analyst faster — reliable pipelines mean insight, not CSV cleaning at 2 AM
  • A great analytics manager attracts other great people — talent compounds on itself
  • A bad analyst ships wrong numbers — stakeholders lose trust, the team loses its seat at the table
  • A bad data engineer builds fragile pipelines — every analyst inherits tech debt
  • Trust in analytics is uniquely fragile — one bad number in a board deck and your credibility is gone for a year
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Hiring Returns Compound

Hiring is the one activity where the returns — positive or negative — compound long after the decision is made.

  • Engineering ships code — if it breaks, you can see the error
  • Analytics ships numbers — if they are wrong, sometimes nobody notices for months
  • By the time the damage surfaces, trust is already gone
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The State of Analytics Hiring

The Market Reality

  • Data roles have grown ~650% in the past decade — demand has exploded, but hiring practices have not kept up
  • Most data job descriptions are wish lists, not role designs
  • The majority of analytics interviews are unstructured — a casual chat that predicts almost nothing
  • Top candidates have multiple offers and they are evaluating you as much as you are evaluating them
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What the Best Companies Have Figured Out

  • Stripe built structured interviews for data roles with calibrated rubrics and blind-reviewed take-homes
  • Airbnb invested in making every candidate interaction reflect their values — even rejection emails
  • These companies win talent not because they pay the most, but because their process signals competence and respect

When a candidate goes through a well-structured interview, they think "these people know what they are doing — I want to work here."

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What We'll Build This Block

Four Deliverables, Connected

  Role Design         Work Sample         Scoring Rubric       Interview Loop
  ───────────    →    ───────────    →    ──────────────   →   ──────────────
  90-day outcomes     Realistic task      Predefined           Who evaluates
  drive the JD        tests real work     dimensions + scale   what, and how
                                                               you decide
  • Job Description → outcomes, not skills | Work Sample → tests real work
  • Rubric → fair evaluation | Interview Loop → who evaluates what
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Why These Four Connect

  • The outcomes in your JD determine what your work sample tests
  • Your rubric dimensions map to what the work sample reveals
  • Your interview loop assigns evaluators to rubric dimensions

Later This Block

Your hiring packet will be used in a role-play exercise where you will conduct a structured interview with a classmate acting as a candidate. You will experience both sides — interviewer and candidate — and stress-test your rubric against a real human being.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Cost of a Bad Hire

  • Average cost of a mis-hire: 1.5–2x annual salary (SHRM, Bradford Smart)
  • In analytics specifically, a wrong hire means:
    • 6 months of bad dashboards that erode stakeholder trust
    • Broken credibility with the business — "analytics never delivers"
    • Team morale damage — good people leave when surrounded by poor performers
  • The "brilliant jerk" problem: technical skill without collaboration destroys team culture faster than incompetence

"The best thing a manager does is hire well. The second best thing is fire fast when they don't." — Ben Horowitz

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Role Design Before Recruiting

Don't start with a job description. Start with outcomes.

The Role Scorecard Approach

  1. What does this person need to accomplish in their first 90 days?
  2. Define 3–5 concrete outcomes, not a laundry list of skills
  3. Work backward from outcomes to required capabilities
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Role Design: An Example

90-Day Outcome Capability
Instrument top-3 user flows SQL + analytics engineering
Deliver weekly KPI dashboard Visualization + communication
Scope first A/B test Experimental design basics

The resulting JD looks nothing like a skills shopping list.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

JD Anti-Pattern: The 25-Bullet Requirements List

"Required: Python, R, SQL, Spark, Hadoop, Tableau, Power BI, Looker, dbt, Airflow, Docker, Kubernetes, AWS, GCP, machine learning, deep learning, NLP, time series forecasting, A/B testing, causal inference, stakeholder management..."

What this signals: "We do not know what this person will actually do."

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

JD Anti-Pattern: The Unicorn

"PhD in Statistics required. 10+ years of experience. Must thrive in a fast-paced startup." — Compensation: €75K–€85K

What this signals: "We want Staff-level talent at Junior-level comp."

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

JD Anti-Patterns

Pattern 3: The Copy-Paste

A JD clearly written for a software engineer, with "data scientist" find-and-replaced in. Mentions "code reviews" and "shipping features" but nothing about analysis or stakeholders.

What this signals: "We do not understand what this role is. You will spend your first six months explaining your job to your own manager."

Each of these anti-patterns actively repels the candidates you want most.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What Level Do You Actually Need?

Level What They Do Autonomy Scope
Junior Executes well-defined analyses Needs clear specs Single tasks
Mid Scopes own work, communicates findings Self-directed on known problems Projects
Senior Influences strategy, mentors others Navigates ambiguity Team-wide
Staff/Principal Org-wide impact, shapes practice Sets direction Cross-team
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Common Leveling Mistakes

  • Hiring senior when you need mid — you pay for strategy but need execution
  • Hiring mid when you need senior — they cannot navigate your ambiguity

Leveling mistakes are expensive because the wrong-level hire will either be bored or overwhelmed, and both lead to attrition.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Leveling Conversation

Having an Honest Discussion with Your Hiring Partner

The leveling conversation is the one you have with your recruiter, HR partner, or finance lead before you open the role. It is often uncomfortable.

The tension:

  • You need a Senior Analyst who can navigate ambiguity and influence PMs
  • Finance approved headcount for a Mid-level role
  • Recruiter: "post as Senior, see who applies" — it's a trap
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Leveling Conversation: Pressure to Level Down

  • "We can train them up" — Sometimes true for Junior to Mid. Rarely true for Mid to Senior. The gap is judgment, not skills.
  • "Let's hire two juniors instead of one senior" — Two juniors without a senior mentor means two people making the same mistakes.
  • The hidden cost: A mis-leveled hire costs more than the salary difference between Mid and Senior.
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Leveling Conversation: How to Have It

  1. Bring the outcomes list — "Here is what this person needs to accomplish in 90 days. What level can do this?"
  2. Name the trade-offs — "If we hire Mid, I need to budget 30% of my time for mentoring. Is that what we want?"
  3. Get alignment in writing — A verbal "sure, hire a senior" turns into "why did you approve that comp?" without documentation
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Your First Hire: Analyst vs. Analytics Engineer

Before writing your JD, ask: Does my case context need someone to analyze data, or someone to make data analyzable?

Signal Analyst Analytics Engineer
Data state Clean, warehouse exists Messy, no single source of truth
First 90 days Deliver insights Build pipelines + metrics layer
Key skill SQL + storytelling SQL + dbt + orchestration

If the data is a mess, your first hire is the person who cleans it — not the person who analyzes it.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Hiring in the AI Era

Your candidates will use AI. Plan for it.

  • LLMs can write SQL, draft analyses, and generate visualizations
  • If your work sample tests execution speed, you are testing the AI, not the candidate
  • Allow AI use — then evaluate: did they validate the output?
  • Test framing — Can they define the right question before touching any tool?
  • Test judgment — Can they spot when the AI is subtly wrong?
  • Test communication — Can they explain findings to a non-technical stakeholder?
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Research on Structured Interviews

Predictive Validity (Schmidt & Hunter, 1998)

Method Correlation Method Correlation
Unstructured interviews ~0.38 Cognitive ability tests ~0.51
Structured interviews ~0.51 Reference checks ~0.26
Work samples ~0.54 Years of experience ~0.18

Years of experience is the worst predictor — yet most JDs filter on it first.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What the Research Tells Us

Key insight: Gut feeling is mostly noise. Structure = fairness + signal.

  • Structured interviews are the single biggest improvement you can make — free
  • Work samples outperform every other method because they directly observe the work
  • Years of experience is the worst predictor — yet it's the top filter on most JDs
  • "I know a good hire when I see one" = noise as intuition
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Structure Also Means Fairness

  • Unstructured interviews amplify affinity bias — you hire people who remind you of yourself
  • Structured interviews force you to evaluate evidence against criteria, not vibes
  • This is not bureaucracy. This is how you find signal in a process riddled with bias.

When you are just chatting, you gravitate toward people who remind you of yourself. Same school, same hobbies, same communication style. Structure breaks that pattern.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Designing Work Samples

What Makes a Good Work Sample?

  • Reflects actual job tasks — not LeetCode puzzles for analytics roles
  • Has clear evaluation criteria — defined before candidates see the exercise
  • Respects candidate time — 2–4 hours maximum, clearly communicated
  • Tests judgment, not just technique — "What would you do next?" matters more than "Can you write a LEFT JOIN?"

Example: "Here is a messy dataset of user activity logs and a business question from the VP of Product: 'Should we invest in improving our onboarding flow?' Walk us through your approach, your analysis, and your recommendation."

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What Work Samples Really Evaluate

How they frame the problem > How they execute the analysis > How they communicate the findings

  • Framing is the most diagnostic — do they ask clarifying questions? Do they identify what is out of scope?
  • Execution — is the analysis sound? Do they handle missing data thoughtfully?
  • Communication — can they explain findings to a non-technical stakeholder?

The framing question reveals whether someone can operate independently or needs everything spelled out.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Work Sample Design: Realism and Respect

  • The work sample should feel like a real day on the job, not a homework assignment
  • State the expected time upfront: "This should take 2–4 hours. We do not expect or want you to spend more."
  • Provide the dataset, the business question, and the context — do not make candidates guess what you want
  • Do not assume tool access — "Use whatever tool you are comfortable with"
  • Provide the rubric upfront — candidates should know what dimensions they are being evaluated on
  • Allow flexible formats — slide deck, written memo, Jupyter notebook, or recorded video
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Work Sample Design: Setting Up for Success

  • Include a data dictionary and note known data quality issues
  • Provide a contact for clarifying questions — set up for success, not gotchas
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

When Candidates Go Above and Beyond

  • Some candidates will spend 15 hours on a 4-hour exercise. Do not reward this. Score against the rubric, not effort or volume.
  • Doing the minimum well is a signal of prioritization and judgment — exactly what you want in an analyst.
  • Going above and beyond is not a negative, but it should not substitute for quality on core dimensions.
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Why a Rubric?

  • Score on predefined dimensions, not overall impression
  • Forces interviewers to articulate what they are evaluating

Without a rubric:

  • The debrief becomes a storytelling contest — whoever tells the most compelling anecdote wins
  • Interviewers default to overall impression: "I liked them" or "Something felt off"
  • Halo effect takes over — one strong dimension inflates all scores
  • "Culture fit" becomes a catch-all for affinity bias in disguise
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What a Rubric Makes Possible

  • Each interviewer independently rates specific dimensions with behavioral anchors
  • The debrief shifts to calibration: "I gave a 2 on Business Context because they never connected the analysis to a decision"
  • Disagreements become productive — they reveal information, not just preferences

Without a rubric, disagreements are opinion collisions. With a rubric, they surface the most important information about a candidate.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Dimensions for Analytics Roles

Dimension What You Are Looking For
Technical Execution Sound methodology, clean analysis, handles edge cases
Communication Clarity Explains findings to non-technical audience
Business Context Awareness Connects analysis to decisions and outcomes
Intellectual Curiosity Asks good questions, explores beyond the obvious
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The 1–4 Scale (No Fence-Sitting)

Use a 4-point scale. Not three, not five.

Score Meaning
1 Below the bar — significant concerns
2 Approaching the bar — some gaps
3 Meets the bar — effective in this role
4 Exceeds the bar — exceptional, raises the team

Most scores should be 2s and 3s. If you are giving many 1s and 4s, your calibration is off or your pipeline needs work.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Calibrating Scorers: The Problem

  • Interviewer A thinks "3" means "could do the job"; Interviewer B thinks "3" means "among the best I have seen"
Interviewer Tech Comm Biz Context Curiosity
Hiring Manager 3 3 2 3
Tech Peer 2 2 2 2
PM Partner 3 4 3 3
Skip-Level 3 3 3 4

Average ranges from 2.0 to 3.25. Is this a hire? Without norming, your debrief becomes an argument about what the numbers mean rather than what the candidate demonstrated.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Calibrating Scorers: The Fix

The Norming Session

Before the interview loop begins:

  1. Score a practice candidate together (use a past work sample or a fabricated one)
  2. Discuss every disagreement — "Why did you give a 2 on communication?"
  3. Align on what each score means for each dimension
  4. Document calibration notes for reference
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

What Norming Prevents

  • Anchoring to the first interviewer's opinion
  • Halo effect — one strong dimension inflates all scores
  • Recency bias — the last candidate always seems freshest
  • Similarity bias — rating candidates higher when they remind you of yourself
  • Contrast effect — a mediocre candidate looks great after a terrible one

A norming session takes 20–30 minutes and saves hours of circular debrief arguments. You compare candidates to a standard, not to each other.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Interview Loop: Who Interviews for What

Interviewer Focus Area Why
Hiring Manager Role fit + team dynamics Owns the decision
Technical Peer Depth of craft Evaluates real skill
Cross-Func Partner Collaboration + comms Works with them daily
Skip-Level Judgment + growth Longer-term view
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Why Four Interviewers

  • Four is the sweet spot — fewer than three lacks signal diversity; more than five wastes everyone's time
  • Each interviewer has a distinct focus — no one evaluates "general impressions"
  • The cross-functional partner is often the most diagnostic — a PM who will work with this analyst daily can evaluate collaboration in ways no hiring manager can

Google's People Operations research found that four interviews predict hiring outcomes almost as well as eight or more.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Interview Loop: Independent Scores

The Critical Rule: Independent Scores Before Discussion

  1. Each interviewer submits scores before the debrief meeting
  2. No peeking at others' scores
  3. Debrief starts with a round-robin: each person shares scores and top observations
  4. Hiring manager makes the final call — this is not a consensus vote
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Why Independence Matters

  • The first strong opinion in a debrief anchors everyone else
  • Independent scoring ensures you get four genuine data points, not one opinion amplified by social pressure
  • Disagreements are signal, not noise — when one interviewer says "strong hire" and another says "no hire," dig into why

The hiring manager makes the final call. This is not a democracy. Consensus selects for "no one objected" rather than "someone was enthusiastic."

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Debrief Meeting

  1. Before the meeting: All scores submitted independently. Hiring manager reviews for major disagreements.
  2. Round-robin: Each interviewer shares scores and one key insight per dimension — not a 10-minute monologue.
  3. Discuss disagreements first: "You gave a 2 on Business Context and you gave a 4. Walk us through what you saw."
  4. Separate signal from preference: "I did not like their style" is preference. "They could not connect analysis to a decision" is signal.
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Debrief: Reading the Patterns

  • Strong Hire vs. No Hire on the same dimension — the most informative signal. One interviewer may have probed deeper.
  • Consistent 3s across all dimensions — often a "weak hire" signal. No one is excited. Be cautious.
  • One dimension drags the average down — ask: is this coachable? A 2 on Technical Execution is harder to fix than a 2 on Communication.
  • You have heard all the data. You decide. Do not hide behind consensus.
  • When in doubt, do not hire. A false positive (bad hire) is far more expensive than a false negative.
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Candidate Experience: Your Employer Brand

  • Timeline: < 1 week between stages. Silence is a decision — the decision to lose good candidates.
  • Communication: Proactive updates even when there is no update. "We are still reviewing" beats silence.
  • Rejection: Specific, kind, and fast.
  • You will interview ~20 candidates for each hire — 19 of them will tell their network about the experience
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Candidate Experience: Non-Negotiables

The Minimum Standard

  • Acknowledge every application within 48 hours
  • Provide a clear timeline at every stage
  • Give feedback with rejections (even brief: "We went with a candidate with deeper SQL experience")
  • Never ghost a candidate who completed a work sample
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Going Beyond the Minimum

  • Send the work sample rubric in advance — this signals that you respect their time
  • Offer a 15-minute feedback call to any candidate who completed a work sample
  • Close the loop quickly — the best candidates have other offers. A 6-week process loses them.
  • Be honest about the role — if the data infrastructure is a mess, say so. Candidates who join with realistic expectations stay longer.

Even candidates you reject will walk away thinking "that was the most professional interview process I have been through." That is your employer brand.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Building Your Employer Brand in Analytics

  • Open-source contributions — nothing says "serious data work" like a well-maintained GitHub repo
  • Blog posts about your stack — write about problems you solved, architecture decisions, mistakes you learned from
  • Speaking at meetups — encourage team members to present at local data meetups or conferences
  • Treating rejected candidates well — every person who thinks "that was fair" becomes an ambassador
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

The Employer Brand Compounding Effect

  • Year 1: Nobody knows who you are. You compete on comp and title.
  • Year 2: A few blog posts, a meetup talk, good rejection experiences. Inbound pipeline improves.
  • Year 3: Candidates apply specifically because they heard your team is well-run. You select from a stronger pool.

The best analytics teams spend less time recruiting because their brand does the work for them.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Activity: Build Your Hiring Packet

25 Minutes | templates/job-description.md + templates/rubric.md

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Activity Brief: Build Your Hiring Packet

Use the company scenario you chose in Block A (small / medium / large)

In-Class (25 min)

  1. Job Description (templates/job-description.md)
  2. Scoring Rubric (templates/rubric.md)

Portfolio Homework (complete between Day 1 and Day 2)

  1. Work Sample Exercise (templates/work-sample.md)
  2. Interview Loop Design (templates/interview-loop.md)
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Activity Tips

  • Start with the 90-day outcomes — everything else follows
  • Be honest about "required" vs. "preferred" qualifications
  • Design the rubric so a stranger could use it and reach similar conclusions
  • You will use this rubric in the role-play next — make it specific enough to actually score a candidate

If you are stuck, start with: "What would make me confident this person can succeed in their first 90 days?" and work backward.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Role-Play: Structured Interview Practice

Pair up with someone — ideally a different case context.

Time What happens
Setup 2 min Candidate picks a profile, reads it silently
Interview 6 min Interviewer runs structured interview using their rubric
Debrief 2 min Interviewer shares scores. Candidate reveals profile and gives feedback

Round 1 (10 min): Person A is the candidate. Person B interviews.
Round 2 (10 min): Swap roles. Person B picks a different profile.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Candidate Profiles

Candidate: Pick ONE file. Read it silently. Do not show it to your interviewer.

Profile File
A — Strong candidate materials/candidate-a-priya.md
B — Borderline materials/candidate-b-marcus.md
C — Wrong level materials/candidate-c-sarah.md
D — Culture fit trap materials/candidate-d-jordan.md
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Role-Play: The Goal

You are testing whether your rubric works, not whether the candidate is a fit.

  • Interviewer: Use your rubric's "How to Probe" questions. Score in real time.
  • Candidate: Play the profile, not yourself. Lean into the weaknesses.
  • After each round: Interviewer shares scores. Candidate reveals profile and gives feedback.
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Debrief

Reflection Questions

  • What signals were hardest to evaluate?
    • Where did your rubric give clear answers? Where did it fail?
  • Where did your rubric break down?
    • Missing dimensions? Ambiguous score definitions?
  • What would you change?
    • About your questions, rubric, or scoring process?
CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Debrief: The Candidate Perspective

  • What surprised you about being the candidate?
  • What felt fair? What felt arbitrary?

The reveal: Each profile had a hidden "tell" — a signal a well-designed rubric should catch. Did yours surface it?

A rubric is a living document. After every real interview loop: add a missing dimension, sharpen a vague score, remove what wasn't diagnostic.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Hiring Anti-Patterns

  • "Culture fit" as a criterion — This is how you hire clones of yourself. Replace with "culture add" or specific collaboration behaviors.
  • The 20-hour take-home — Disrespects candidate time, biases toward people without caregiving responsibilities. Two to four hours, clearly scoped.
  • The "rockstar/ninja" JD — Signals immaturity. Senior candidates run from these.
  • Ghosting candidates — Especially after a work sample. The analytics community is small.
  • Consensus-based decisions — "Everyone needs to agree" selects for inoffensive mediocrity.
  • Hiring for potential without structure — "I see myself in them" is bias.

All of these share a common root: substituting gut feeling for structured evaluation.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Key Takeaways

  1. Start with outcomes, not skills — the 90-day scorecard drives everything
  2. Structure beats intuition — 0.51 vs. 0.38, every time
  3. Work samples are your best tool — test real judgment, not trivia
  4. Calibrate before you interview — 20 min of norming prevents hours of debate
  5. Candidate experience is employer brand — 19 rejections for every 1 hire

Between Now and Day 2

Start your work sample design and interview loop — these are portfolio deliverables. Your JD and rubric from today are the foundation.

CEU Vienna | Day 1 – Block B
ECBS5256 – Managing Data Science Teams

Up Next: Block C

Roadmaps, Bets, and Alignment

15:30–17:10

CEU Vienna | Day 1 – Block B

⏱ Expected: 13:30 (min 0/100)

Talk track: Block B — Hiring and Team Formation. This is the post-lunch block, so bring energy and start with the provocative cost-of-bad-hire framing.

Talk track: Welcome back from lunch. This block is about the single highest-leverage activity you will do as a manager — hiring. A great hire compounds for years; a bad hire compounds damage for months. By the end of this block, you will draft a job description and scoring rubric in class, then complete the full hiring packet — including a work sample exercise and interview loop design — as portfolio work between sessions. Let's get into it.

⏱ Expected: 13:30 (min 0/100)

Talk track: One clear outcome for this block — you'll leave with a hiring packet you could actually use.

⏱ Expected: 13:30 (min 0/100)

Talk track: Hiring is the single highest-leverage thing you do as a manager. A great analyst makes every PM smarter. A great data engineer makes every analyst faster. And it compounds the other way too — a bad analyst ships a wrong number in a board deck and the CFO stops trusting anything from your team. Analytics is uniquely fragile because our product is trust. One bad number and your credibility is gone for a year.

⏱ Expected: 13:30 (min 0/100)

Talk track: This is the key insight. Engineering has a feedback loop — broken code throws errors. Analytics does not. A wrong number can circulate through a board deck, influence a strategy decision, and nobody catches it for months. By the time they do, the damage is done. That is why hiring well in analytics is even more critical than in other functions. The returns, positive or negative, compound long after the decision is made.

⏱ Expected: 13:30 (min 0/100)

Talk track: Let's talk about the hiring market for a moment. Data roles have exploded — something like 650 percent growth over the last decade depending on which report you read. But here is the problem: hiring practices have not kept up with the demand. Most data job descriptions are still wish lists. Most analytics interviews are still unstructured — a hiring manager chats with the candidate for 45 minutes and then makes a gut call. Meanwhile, the best candidates have three or four offers. They are interviewing you as much as you are interviewing them.

⏱ Expected: 13:30 (min 0/100)

Talk track: The companies that figured this out — Stripe, Airbnb, and others — they win talent not by paying the most but by running a process that signals competence and respect. When a candidate goes through a well-structured interview, they think "these people know what they are doing — I want to work here." That is the competitive advantage of a good process.

⏱ Expected: 13:30 (min 0/100)

Talk track: Your hiring packet has four connected deliverables that form a system. The job description grounds everything in 90-day outcomes. The work sample tests whether a candidate can do the job. The rubric ensures consistent, fair evaluation. The interview loop structures who evaluates what and how you decide. Each piece feeds the next. In this block, you draft the JD and rubric in class. The work sample and interview loop you complete as portfolio work between sessions.

⏱ Expected: 13:30 (min 0/100)

Talk track: These four pieces form a system. The outcomes in your JD determine what your work sample tests. Your rubric dimensions map to what the work sample reveals. Your interview loop assigns evaluators to rubric dimensions. And here is why this matters: later this block, you will use this packet in a role-play. You will actually conduct a structured interview with a classmate playing a candidate from a profile card. So build something you are willing to use for real.

⏱ Expected: 13:30 (min 0/100)

Talk track: The average cost of a mis-hire is one and a half to two times annual salary. But in analytics, the real cost is worse — bad dashboards erode trust, and the brilliant jerk drives away every collaborative person on your team. As Ben Horowitz puts it, the best thing a manager does is hire well. The second best thing is fire fast when they don't.

⏱ Expected: 13:35 (min 5/100)

Talk track: Most hiring processes start wrong. Someone says "we need a data scientist" and the hiring manager opens a Google Doc and starts listing skills — Python, SQL, Tableau, machine learning, PhD preferred. That is a shopping list, not a role design. Instead, start with the question: what does this person need to accomplish in their first 90 days? Three to five concrete outcomes. Then work backward to what capabilities are actually required.

⏱ Expected: 13:35 (min 5/100)

Talk track: Here is a concrete example. A seed-stage startup needs their first analytics hire. Instead of listing 25 skills, you write down three 90-day outcomes: instrument the top user flows, deliver a weekly KPI dashboard, and scope the first A/B test. Now work backward — what capabilities are actually required? SQL and analytics engineering, visualization and communication, and basic experimental design. That is your job description. Notice how different this looks from "PhD required, 10 years of Python."

⏱ Expected: 13:35 (min 5/100)

Talk track: Before you write your JD, let me show you the anti-patterns. I have collected these from real job postings — names removed to protect the guilty. Pattern one: the 25-bullet requirements list. Python, R, SQL, Spark, Hadoop, Tableau, Power BI — it goes on and on. What this tells a candidate is that you have no idea what this person will actually do. Good candidates self-select out because they do not have all 25, even though no human does.

⏱ Expected: 13:35 (min 5/100)

Talk track: Pattern two: the unicorn. PhD required, ten years of experience, startup mentality — for 80K. This tells senior candidates you either do not understand the market or you are hoping to find someone desperate.

⏱ Expected: 13:35 (min 5/100)

Talk track: Pattern three: the copy-paste. This is a software engineering JD with "data scientist" swapped in. It mentions code reviews and sprint velocity but nothing about stakeholders, analysis, or decision support. The candidate reads this and thinks "they do not even know what this role is." Each of these anti-patterns actively repels the candidates you want most. The best people have options — they will close the tab and move on.

⏱ Expected: 13:35 (min 5/100)

Talk track: Before you write the job description, get honest about what level you actually need. A seed-stage startup does not need a Staff Data Scientist — they need a strong mid-level hire.

⏱ Expected: 13:35 (min 5/100)

Talk track: Leveling mistakes are expensive. A Series B company with ambiguity needs a senior who can influence without authority. The wrong-level hire will either be bored or overwhelmed, and both lead to attrition.

⏱ Expected: 13:35 (min 5/100)

Talk track: This is the conversation nobody teaches you to have. You have done your role design, you know you need a Senior Analyst who can navigate ambiguity and influence product managers. You take this to your recruiter or your finance partner and they say "the budget is for a mid-level hire." Now what? Some managers cave and post a mid-level role hoping to find a unicorn. Others post "Senior" in the title but offer mid-level comp, which is the unicorn JD anti-pattern we just talked about.

⏱ Expected: 13:35 (min 5/100)

Talk track: Let me talk about the pressure to level down. You will hear "we can train them up" — sometimes true for junior to mid, but rarely true for mid to senior because the gap is judgment, not skills. You will hear "let's hire two juniors instead of one senior" — but two juniors without a senior mentor means two people making the same mistakes. The hidden cost of a mis-leveled hire is always more than the salary difference.

⏱ Expected: 13:35 (min 5/100)

Talk track: Here is how to handle it. Bring your outcomes list. Say "here is what this person needs to accomplish in their first 90 days — scope their own work, influence product roadmaps, present to the exec team. What level of hire can do this?" Make the trade-offs explicit. If you hire mid-level, you need to budget 30 percent of your own time for mentoring and air cover. Is that what the organization wants? Sometimes the answer is yes — just make sure everyone agrees. And get the alignment in writing because memory is short when the comp review comes around.

⏱ Expected: 13:35 (min 5/100)

Talk track: Before you write your job description, I need you to ask yourself an honest question. Does your case context need an analyst — someone who takes clean data and turns it into insights? Or does it need an analytics engineer — someone who takes messy, scattered data and turns it into something an analyst can actually use? MSBA students consistently underestimate the data engineering reality. If you picked DataPulse, your data lives in Firebase Analytics, Amplitude, and a PostgreSQL database. An analyst cannot analyze data that does not exist in a queryable form. You need an analytics engineer first. If you picked MarketBridge, you have conflicting metric definitions. You need someone to build the canonical metrics layer before anyone can trust the numbers. This is the most common hiring mistake in analytics — hiring the analyst before the plumbing is in place.

⏱ Expected: 13:45 (min 15/100)

Talk track: Your candidates are using AI tools — ChatGPT, Copilot, Claude. And they should be. The question is what you are testing. If your work sample tests SQL writing speed, you are testing the AI, not the candidate. Instead, allow AI use and evaluate what they did with the output. Did they validate it? Did they catch the edge case? The meta-skill you are hiring for now is knowing when the AI is wrong — that requires domain knowledge, statistical intuition, and intellectual honesty.

⏱ Expected: 13:45 (min 15/100)

Talk track: These numbers come from Schmidt and Hunter's 1998 meta-analysis. An unstructured interview predicts job performance at 0.38. A structured interview jumps to 0.51. Work samples hit 0.54. And years of experience — the thing most JDs filter on first — is the worst predictor at 0.18. The gap between unstructured and structured is the whole argument for this block.

⏱ Expected: 13:45 (min 15/100)

Talk track: So what do we do with this? The takeaway is that moving from unstructured to structured interviews is the single biggest improvement you can make to your hiring process, and it is free. You do not need to buy software or hire a consultant. You need to write down the questions in advance, ask every candidate the same questions, and score the answers against predefined criteria. That is it.

⏱ Expected: 13:45 (min 15/100)

Talk track: Here is the fairness angle — unstructured interviews amplify affinity bias. When you are just chatting, you gravitate toward people who remind you of yourself. Same school, same hobbies, same communication style. A structured process forces you to evaluate evidence against criteria instead of vibes. Some people hear "structure" and think "bureaucracy." I hear "structure" and think "the only way to find signal in a process that is drowning in noise."

⏱ Expected: 13:55 (min 25/100)

Talk track: The work sample is the centerpiece of your hiring process. For analytics roles, this is not a coding test — it is a simulation of the actual job. You give candidates a realistic dataset, messy with known issues, and a business question. Then you evaluate how they frame the problem, execute the analysis, and communicate findings.

⏱ Expected: 13:55 (min 25/100)

Talk track: What are you really evaluating? First, framing — do they ask clarifying questions and identify scope? Second, execution — is the analysis sound? Third, communication — can they explain to a non-technical audience? Framing is the highest-signal dimension: someone who can operate independently will naturally define what is out of scope before diving in.

⏱ Expected: 13:55 (min 25/100)

Talk track: Respect candidate time — state the expected commitment upfront and mean it. Make it inclusive: do not require a specific tool, share the rubric with the candidate, and allow flexible formats. You want to see their best work on the things that matter, not test whether they happen to know your favorite tool.

⏱ Expected: 13:55 (min 25/100)

Talk track: Set candidates up to succeed. Include a data dictionary, note known data quality issues, and give them a contact for clarifying questions. You want to evaluate their analytical thinking, not their ability to reverse-engineer your data model.

⏱ Expected: 13:55 (min 25/100)

Talk track: Now, when candidates go above and beyond — and some will spend 15 hours on a 4-hour exercise — do not reward that. It actually tells you something concerning about their ability to scope and prioritize. A clear, concise deliverable that nails the core dimensions is more impressive than a 40-slide deck. You are evaluating thinking, not production value.

⏱ Expected: 13:55 (min 25/100)

Talk track: A rubric is not optional. Without one, your debrief becomes a storytelling contest — the most articulate interviewer wins, not the best candidate. The halo effect takes over, and "culture fit" becomes code for "I would get a beer with this person." A rubric forces you to evaluate evidence against criteria, not vibes.

⏱ Expected: 13:55 (min 25/100)

Talk track: With a rubric, the conversation changes. Instead of "I liked them," it becomes "I gave a 2 on Business Context because when I asked how the analysis would inform a decision, they could not connect the dots." That is a specific, actionable observation. And when two interviewers disagree — one gave a 3 on communication, the other gave a 2 — that disagreement becomes productive. You ask "what did you see?" and you learn something. Without a rubric, disagreements are just opinion collisions.

⏱ Expected: 13:55 (min 25/100)

Talk track: For analytics hires, I recommend four dimensions: technical execution, communication clarity, business context awareness, and intellectual curiosity. You can adapt these, but I would not go above five dimensions — more than that and your interviewers lose focus. Each dimension should map to something you can observe in the work sample or the interview.

⏱ Expected: 13:55 (min 25/100)

Talk track: Use a four-point scale. Not three — that lets everyone pick the middle. Not five — nobody can reliably distinguish between a 3 and a 4 in a 45-minute interview. Four points forces a meaningful distinction. Most of your scores should be 2s and 3s. If you are giving many 1s and 4s, your calibration is off or your pipeline needs work.

⏱ Expected: ~13:55 (min 25/100) | SKIP — advance past

Talk track: Four interviewers, same candidate, wildly different scores. The Tech Peer gives mostly 2s, the PM Partner gives a 4 on communication. Is the Tech Peer a tough grader or did they catch a genuine weakness? Without calibration, you cannot tell whether disagreements reflect different standards or different observations.

⏱ Expected: ~13:55 (min 25/100) | SKIP — advance past

Talk track: The fix is a norming session, and it is simpler than you think. Before your interview loop begins, get all interviewers in a room for 20 to 30 minutes. Show them a work sample — either from a past candidate anonymized, or one you fabricated. Have everyone score it independently. Then discuss. You will be amazed at how different the scores are. One person gave a 4 on communication, another gave a 2, and they were looking at the same work. Walk through each disagreement and align on what each score means.

⏱ Expected: ~13:55 (min 25/100) | SKIP — advance past

Talk track: Norming prevents anchoring, halo effect, recency bias, similarity bias, and the contrast effect. Twenty minutes of norming prevents hours of circular debrief arguments. It is the highest-ROI meeting in your entire hiring process.

⏱ Expected: 14:05 (min 35/100)

Talk track: The interview loop is the structure around who evaluates what. Four interviewers, each with a distinct focus. The hiring manager looks at role fit and team dynamics. A technical peer evaluates depth of craft. A cross-functional partner tests collaboration and communication. And a skip-level manager assesses judgment and growth potential over a longer horizon.

⏱ Expected: 14:05 (min 35/100)

Talk track: Four is the right number. Research from Google's People Operations team found that four interviews predict hiring outcomes almost as well as eight or more. Beyond four, you get diminishing returns and a worse candidate experience. The cross-functional partner is often the most revealing interview. A product manager who will work with this analyst daily can tell you things about collaboration and communication that no amount of technical evaluation will surface.

⏱ Expected: 14:05 (min 35/100)

Talk track: The critical rule for the interview loop: everyone submits their scores independently before the debrief. No hallway conversations. No Slack messages saying "what did you think?" before scores are in. Why? Because the first strong opinion in a debrief anchors everyone else. If your VP walks into the room and says "I think she is great," watch how quickly every 2 becomes a 3. Independent scoring ensures you actually get four genuine data points.

⏱ Expected: 14:05 (min 35/100)

Talk track: And here is the nuance — the hiring manager makes the final call. This is not a democracy. Consensus-based hiring selects for "no one objected" rather than "someone was enthusiastic." But the hiring manager should take disagreements seriously. When one interviewer says strong hire and another says no hire, that gap is the most important information in the room. Do not average it away. Dig into it.

⏱ Expected: 14:05 (min 35/100)

Talk track: The debrief is where many structured processes fall apart. All scores are submitted independently beforehand. Start with a round-robin — one key insight per dimension, not a monologue. Then go straight to disagreements. Always separate signal from preference: "I did not like their style" is preference; "They could not connect analysis to a decision" is signal.

⏱ Expected: 14:05 (min 35/100)

Talk track: Watch for the "consistent 3s" pattern — it feels like a hire but often means nobody is excited. When one dimension drags the average, ask if it is coachable. The hiring manager makes the final call — do not hide behind consensus. A "strong hire" from the technical peer is worth more than lukewarm 3s from everyone. When in doubt, pass.

⏱ Expected: ~14:05 (min 35/100) | SKIP — advance past

Talk track: For every person you hire, you reject about 19. Those 19 people will tell their network about their experience. Your employer brand is built in the rejection email, not the offer letter. The analytics community is small — the analyst you reject today might be the hiring manager you need to impress in two years.

⏱ Expected: ~14:05 (min 35/100) | SKIP — advance past

Talk track: Here are the non-negotiables. Acknowledge every application within 48 hours — this can be automated. Provide a clear timeline at every stage — "you will hear from us by Friday" and then actually follow up by Friday. Give feedback with rejections. It does not have to be a detailed performance review. And the big one: never ghost a candidate who completed a work sample. That person invested hours of their life in your process.

⏱ Expected: ~14:05 (min 35/100) | SKIP — advance past

Talk track: Go above the minimum: offer feedback calls, close the loop quickly, and be honest about the role. Even candidates you reject become ambassadors when you treat them well.

⏱ Expected: ~14:05 (min 35/100) | SKIP — advance past

Talk track: The long game of employer brand: open-source contributions, blog posts about your stack, speaking at meetups, and treating rejected candidates well. These compound over time — by year three, candidates apply specifically because they heard your team is well-run.

⏱ Expected: ~14:05 (min 35/100) | SKIP — advance past

Talk track: This is not something you fix in a quarter. It compounds over time. Year one, nobody knows who you are and you compete purely on compensation. Year two, a few blog posts and a meetup talk later, your inbound pipeline improves. Year three, candidates apply specifically because they heard your team is well-run. The best analytics teams I know spend less time recruiting because their brand does the recruiting for them. That compounding effect is worth the investment.

⏱ Expected: 14:10 (min 40/100) | Activity brief — JD + Rubric; activity starts at 14:13

Talk track: Time to put this into practice. For the next 25 minutes, you will focus on two deliverables: your job description and your scoring rubric. These are the foundation of the hiring packet. Work sample design and interview loop design are part of your final portfolio — start them between now and Day 2. Use the templates provided. Work individually but feel free to ask questions. I will circulate. At the end, you will pair up for a role-play exercise using the rubric you built.

⏱ Expected: 14:10 (min 40/100) | Activity brief — JD + Rubric; activity starts at 14:13

Talk track: In class, you have two deliverables: the job description with 90-day outcomes, and the scoring rubric. Focus your 25 minutes here. The work sample exercise and interview loop design are part of your final portfolio — start them between now and Day 2, building on the JD and rubric you draft today. You will have until the portfolio deadline to polish everything.

⏱ Expected: 14:10 (min 40/100) | Activity brief — JD + Rubric; activity starts at 14:13

Talk track: A few tips. Start with the 90-day outcomes because everything else flows from there. Be honest about what is truly required versus merely preferred. Design the rubric so that a stranger could pick it up and reach similar conclusions to you — that is the test of a good rubric. You are about to use this rubric in a live role-play, so make it specific enough to actually score someone. If you are stuck, ask yourself "what would make me confident this person can succeed?" and work backward from there. You have 25 minutes — focus on the JD and rubric. Go.

⏱ Expected: 14:38 (min 68/100) | Role-Play Setup

Talk track: After the 25 minutes of building, we're going to do something uncomfortable and valuable. Pair up — ideally with someone who chose a different case context. One of you will play a candidate, the other will interview them using the rubric you just built. Then you swap. Here's the key: the candidate picks their own profile and reads it silently. The interviewer does not see the profile. You're going in blind, just like a real interview.

⏱ Expected: 14:38 (min 68/100) | Role-Play Setup

Talk track: Here are the four profiles. Each one is an archetype you will encounter in real hiring. Pick one — don't tell your partner which one. Open the file and read it silently. You have two minutes. Play the character authentically, including their weaknesses. If your profile says you struggle to explain things to non-technical people, then struggle. The goal is to stress-test the rubric, not to practice being impressive.

⏱ Expected: 14:38 (min 68/100) | Role-Play Setup

Talk track: I want to be clear about the goal. You are not evaluating whether this person would get hired. You are testing whether your rubric gives you a structured way to evaluate someone. Your rubric was designed for your case context — maybe you wrote it for a senior analyst at FinGuard. The candidate might be playing a junior bootcamp grad. That mismatch is fine. The question is: does your rubric help you see signal, or does it fall apart when a real person is sitting across from you?

⏱ Expected: 15:02 (min 92/100) | Debrief

Talk track: Let's debrief. Interviewers first: what was the hardest signal to evaluate? Where did your rubric give you a clear answer and where did it leave you guessing? This is the most important insight from the exercise — your first rubric will be wrong. Every rubric is wrong the first time. The goal is not perfection, it is iteration.

⏱ Expected: 15:02 (min 92/100) | Debrief

Talk track: Now candidates: what felt fair about the process? What felt arbitrary or unclear? Here's something I didn't tell you before the exercise — each profile had a hidden tell. Profile A proactively suggests improvements beyond the question. Profile B defaults to jargon when asked to simplify. Profile C steers everything toward strategy and delegation. Profile D pivots to storytelling when pressed on methodology. Did your rubric catch the tell? If not, that's your most valuable insight — you now know what dimension to add or sharpen. The rubric is a living document. Your first version will always be wrong. The goal is iteration.

⏱ Expected: 15:02 (min 92/100) | Debrief

Talk track: Quick anti-pattern hits: culture fit as code for "like me," the 20-hour take-home, rockstar JDs, ghosting candidates, consensus hiring, and unstructured "potential" assessments. They all share a root cause — substituting gut feeling for structured evaluation. Structure beats intuition. Every time.

⏱ Expected: 15:02 (min 92/100) | Debrief

Talk track: Five things to remember. Start with outcomes. Structure beats intuition. Work samples are your best tool. Calibrate your scorers. And candidate experience is employer brand. Between now and Day 2, start on your work sample design and interview loop — your JD and rubric from today are the foundation.

⏱ Expected: 15:02 (min 92/100) | Debrief

Talk track: We will take a 20-minute break. When you come back, Block C is about roadmaps, bets, and alignment — how you decide what your analytics team works on, how you communicate priorities to leadership, and how you say no to the 80 percent of requests that do not make the cut. See you at 15:30.