Why Good Data Scientist Candidates Still Get Rejected

EXAMPLE JOB DESCRIPTION PHRASES -> WHAT TO PREPARE

"Partner with product managers to define success metrics"
 -> metric design, guardrail metrics, SQL

"Design, run and analyse online experiments"
 -> sample size, power, significance, pitfalls like peeking

"Build and maintain models in production"
 -> clean Python, tests, versioning, monitoring

"Strong communication with non-technical stakeholders"
 -> one project story told without jargon

Count the phrases in each group. The biggest group is
the must-have this team will test hardest.

SAME PROJECT, PITCHED FOR TWO ROLES

For a product DS role:
"I analysed why some users dropped off after signup,
 defined an activation metric with the product team, and
 proposed an experiment to test a shorter onboarding."

For an applied ML role:
"I built a model predicting early drop-off. I structured
 the feature code as tested functions, wrapped the model
 in a scikit-learn pipeline and saved it with a version
 tag so the same code runs in training and scoring."

Both are true. Each emphasises what that team needs.

# features.py
import pandas as pd


def add_tenure_months(df: pd.DataFrame, ref_date: str) -> pd.DataFrame:
    """Add whole months between signup_date and ref_date.

    Returns a new DataFrame; the input is not modified.
    Signups after ref_date get a tenure of 0.
    """
    out = df.copy()
    ref = pd.Timestamp(ref_date)
    signup = pd.to_datetime(out["signup_date"])
    months = (ref.year - signup.dt.year) * 12 + (ref.month - signup.dt.month)
    out["tenure_months"] = months.clip(lower=0)
    return out

# test_features.py  (run with: pytest -q)
import pandas as pd
from features import add_tenure_months


def test_basic_tenure():
    df = pd.DataFrame({"signup_date": ["2024-01-15", "2024-06-01"]})
    out = add_tenure_months(df, "2024-06-30")
    assert out["tenure_months"].tolist() == [5, 0]


def test_future_signup_is_zero():
    df = pd.DataFrame({"signup_date": ["2024-08-01"]})
    out = add_tenure_months(df, "2024-06-30")
    assert out["tenure_months"].tolist() == [0]


def test_input_not_modified():
    df = pd.DataFrame({"signup_date": ["2024-01-15"]})
    add_tenure_months(df, "2024-06-30")
    assert "tenure_months" not in df.columns

from statsmodels.stats.power import NormalIndPower
from statsmodels.stats.proportion import proportion_effectsize

# Example inputs: current conversion 10%, smallest lift worth
# detecting is to 11%. These are assumptions to agree with the team.
baseline_rate = 0.10
target_rate = 0.11

effect = proportion_effectsize(target_rate, baseline_rate)
n_per_group = NormalIndPower().solve_power(
    effect_size=effect, alpha=0.05, power=0.8,
    ratio=1.0, alternative="two-sided",
)
print(f"users needed per group: {n_per_group:,.0f}")

# Spoken: "Detecting a one-point lift from 10% needs many
# thousands of users per group. If traffic is low, I would
# either run longer or choose a bigger minimum effect."

KEEP ONE ROW PER PROCESS

Company type | Role flavour | Stage reached | Hardest question
_____________|______________|_______________|_________________
_____________|______________|_______________|_________________
_____________|______________|_______________|_________________

After three rows, look for patterns:
- Do rejections cluster at the same stage?
- Is it always the same role flavour?
- Is the hardest question always in one area?

The pattern is your next study topic.

ROLE: ______________________  COMPANY TYPE: ____________

Phrases about experiments/metrics: ______  count: __
Phrases about modelling/ML: _____________  count: __
Phrases about production/code: _________  count: __
Phrases about research/theory: _________  count: __

Role flavour (largest count): ____________________
Must-have skill: ________________________________
My weakest area for this must-have: ______________

FOR THE MUST-HAVE SKILL, PREPARE:

One project story that shows it: ________________
One piece of code or analysis to share: _________
One question I expect, and my answer: ___________
One honest gap, and how I am closing it: ________

FOR EACH ROUND, ONE SENTENCE OF EVIDENCE:
Screen: ____________________________________
Technical: _________________________________
Case / manager: ____________________________

WITHIN 24 HOURS OF EACH ROUND

Questions asked (as exactly as I remember):
1. ______________________
2. ______________________
3. ______________________

Where I felt least confident: ____________________
Which must-have did this round test? _____________
Did I give clear evidence for it? Y / N
What I will practise before the next interview: ____