Data Science Consulting Flashcards

Question 1

What is a data science consulting engagement and who is involved?

Accepted Answer

A data science consulting engagement is a time-limited project in which an analytics specialist helps a client answer a defined business question. The consultant contributes statistical methods, data tools, and structured problem solving, while the client provides domain knowledge, data access, and decision authority. Success depends on both sides agreeing on scope and deliverables.

Question 2

Why do consulting engagements usually begin with a discovery phase?

Accepted Answer

A discovery phase lets the consultant learn the client's goals, constraints, data sources, and decision process before committing to a method. Early interviews and document review reduce the risk of solving the wrong problem. Teams that skip discovery often build accurate models that answer questions nobody asked, wasting time and trust.

Question 3

What is the difference between a deliverable and an insight in a consulting project?

Accepted Answer

A deliverable is a tangible item the client receives, such as a report, dashboard, model, or recommendation memo. An insight is a finding that changes how someone thinks or acts, such as learning that repeat customers cluster around one product. Strong projects produce both, because deliverables carry insights into the organization in usable form.

Question 4

Why should a consultant define success criteria before analysis begins?

Accepted Answer

Success criteria state in measurable terms what result would make the project worth doing, such as a minimum forecast accuracy. Agreeing on them early prevents later disputes about whether the work succeeded. They also guide choices about data and methods.

Question 5

What information usually appears in a consulting statement of work?

Accepted Answer

A statement of work is a written agreement that describes the problem, objectives, scope, deliverables, timeline, responsibilities, and fees. It also lists assumptions and data the client must supply, along with exclusions that clarify what the project will not cover. Clear documentation protects both parties when priorities shift midway through an engagement.

Question 6

How does a consultant determine whether a project is feasible?

Accepted Answer

A feasibility check asks whether the available data, time, budget, and methods can plausibly answer the question. The consultant reviews data availability, confirms the target outcome is recorded, and checks that stakeholders can act on results. If any of these fail, the project should be reframed or stopped early.

Question 7

Why is stakeholder mapping useful at the start of a project?

Accepted Answer

Stakeholder mapping identifies everyone affected by the project, their level of influence, and their stake in the outcome. It shows who must approve decisions, who will use the results, and who may resist change. With that picture, a consultant can plan communication, schedule reviews, and avoid surprises late in the engagement.

Question 8

What is scope creep and how can consultants limit it?

Accepted Answer

Scope creep is the gradual expansion of project requests beyond the original agreement, often without extra time or budget. Consultants limit it by referring to the signed scope, documenting each change request, and estimating its cost before accepting it. Transparent change control keeps the client informed and protects the quality of the core work.

Question 9

What separates a well framed business question from a vague one?

Accepted Answer

A well framed question names a specific decision, the outcome measure, the population or unit of analysis, and the time horizon. A vague question such as how can we grow sales leaves all of these open. Framing turns a broad concern into something data can actually answer.

Question 10

What is a hypothesis tree and when is it useful?

Accepted Answer

A hypothesis tree breaks a broad question into testable statements arranged in branches, from the top goal down to specific causes. It is useful when a team has many possible explanations and needs to decide which to investigate first. Each leaf should be specific enough that someone could find evidence for or against it.

Question 11

What are the main elements of a useful problem statement?

Accepted Answer

A useful problem statement describes the current situation, the gap between that situation and the desired state, the decision it will inform, and the constraints on any solution. It avoids presuming the answer in advance. A concise statement of this kind gives the whole team a shared reference point.

Question 12

What is the difference between a descriptive, predictive, and prescriptive question?

Accepted Answer

A descriptive question asks what happened, such as which regions had the highest returns last quarter. A predictive question asks what is likely to happen, such as which customers will cancel next month. A prescriptive question asks what action should be taken, such as how to split a budget across channels to maximize sales.

Question 13

What is the unit of analysis and how should a consultant choose it?

Accepted Answer

The unit of analysis is the entity being measured, such as a customer, store, transaction, or day. Consultants choose it based on the decision at hand, because results at one level can mislead when applied to another. For example, store-level averages may hide differences among individual customers within those stores.

Question 14

What is a baseline and why does a project need one?

Accepted Answer

A baseline is the reference level of performance against which any change will be judged, such as average delivery time before a new process starts. Without a baseline, a team cannot tell whether an improvement came from its work or from ordinary variation.

Question 15

How does project scoping differ from problem framing?

Accepted Answer

Framing defines the question the project should answer, while scoping defines the boundaries of the work needed to answer it. Scope settles which data sources, populations, time periods, and deliverables are included and which are excluded. A project can have a sharp question and still fail if its scope is left open-ended.

Question 16

What makes a constraint useful when framing a consulting problem?

Accepted Answer

A useful constraint is a real limit, such as a deadline, budget, legal restriction, or data access rule, that shapes which solutions are feasible. Naming constraints early prevents the team from designing an elegant model that the organization cannot deploy. Constraints also sharpen the question by forcing choices about what matters most.

Question 17

What is data provenance and why does it matter to a consultant?

Accepted Answer

Data provenance is the documented history of where data came from, how it was collected, and what transformations it has undergone. It matters because results may reflect how records were captured rather than the real business process. Clear provenance also supports reproducibility and helps the client trust the analysis.

Question 18

How should a consultant handle missing values in a dataset?

Accepted Answer

The right approach depends on why values are missing and how many there are. Consultants first check whether missingness is random or systematic, since patterns can bias results. Options include removing incomplete records, imputing estimates, or adding an indicator for missingness, and each choice should be documented and tested for sensitivity.

Question 19

What does it mean to validate a data source before analysis?

Accepted Answer

Validating a data source means checking that its fields, units, definitions, and totals match what the client expects and what other records show. A consultant might reconcile revenue totals against finance reports. Validation catches errors early, when they are cheaper to fix than after findings have been shared.

Question 20

Why is an exploratory data analysis step valuable before modeling?

Accepted Answer

Exploratory analysis reveals distributions, outliers, relationships, and data quality issues that a model might otherwise hide or amplify. Summary statistics and simple charts help the team choose sensible transformations and candidate predictors. This step also builds intuition about the business, which makes later model results easier to interpret and question.

Question 21

What is a train and test split and why is it used?

Accepted Answer

A train and test split holds back part of the data so a model can be evaluated on cases it did not see during fitting. The purpose is to estimate performance honestly, since a model judged only on its training data often looks better than it really is.

Question 22

What is data leakage and how can it inflate results?

Accepted Answer

Data leakage occurs when information that would not be available at prediction time enters the training process. It inflates results because the model appears to perform well by exploiting information it will not have in real use. Guarding against leakage requires careful timing checks and strict separation of training and evaluation steps.

Question 23

Why should an analysis plan be written down before the data is examined in depth?

Accepted Answer

A written analysis plan records the questions, metrics, methods, and decision rules before results are known, which limits the temptation to search until something looks significant. It also gives reviewers a clear standard to check the work against. Changes made later should be logged with reasons.

Question 24

What is the purpose of a data dictionary in a consulting project?

Accepted Answer

A data dictionary defines each field in a dataset, including its meaning, data type, allowed values, units, and known limitations. It helps the consultant and client share a common understanding of terms such as active customer or completed order, which often differ across departments. Keeping it current reduces misinterpretation and speeds onboarding.

Question 25

Why does correlation alone not establish that one factor causes another?

Accepted Answer

Two variables can move together because one influences the other, because both respond to a third factor, or simply by chance. Establishing causation requires evidence about the mechanism, the timing of changes, and ideally a comparison that isolates the effect, such as a controlled experiment. Consultants therefore present associations carefully.

Question 26

What does a confidence interval communicate to a client?

Accepted Answer

A confidence interval gives a range of plausible values for an estimate, reflecting uncertainty from working with a sample. It does not give the probability that the true value lies inside any single interval. For clients, the useful takeaway is that the decision should account for the whole plausible range.

Question 27

How should a consultant choose between simple and complex models?

Accepted Answer

Start with the simplest model that addresses the decision, then add complexity only when validation shows a meaningful gain. Simpler models are easier to explain, audit, maintain, and trust, which matters when stakeholders must act on the results. Complex models may be justified when accuracy is critical and the gain is substantial.

Question 28

What is overfitting and how can a team detect it?

Accepted Answer

Overfitting happens when a model captures noise in the training data and performs poorly on new cases. A team can detect it by comparing training and held-out performance, since a large gap is a warning sign. Cross-validation, simpler model structures, and more representative data are common ways to reduce the problem.

Question 29

What should a consultant include in a findings summary for executives?

Accepted Answer

An executive summary should open with the answer to the business question, then state the key evidence, the confidence in that evidence, and the recommended action. It should avoid technical detail and keep to a few clear points. Each claim should trace back to analysis the reader can review.

Question 30

Why are clear visualizations so important in consulting communication?

Accepted Answer

A well-chosen chart lets a busy reader see a pattern, comparison, or outlier in seconds, which text tables rarely achieve. Good charts use clear labels, honest scales, and a single message per graphic. Poor design, such as truncated axes or decorative three-dimensional effects, can mislead even when the underlying numbers are correct.

Question 31

What does it mean to report uncertainty honestly?

Accepted Answer

Reporting uncertainty honestly means stating the limits of the data, methods, and estimates instead of presenting a single number as certain. It includes intervals where appropriate, notes on assumptions, and what the analysis cannot show. Honest reporting does not weaken a recommendation; it helps decision makers judge how much risk they are accepting.

Question 32

How should a consultant present a recommendation that the client may resist?

Accepted Answer

Begin by acknowledging the client's concerns and showing that the analysis considered them. Present the evidence, connect each step to the client's stated goals, and be clear about trade-offs and risks. Offer options with different levels of change, so the client can choose a path that fits their tolerance for risk.

Question 33

What does it mean to hand off a model to the client team?

Accepted Answer

A handoff transfers the working model, code, documentation, and monitoring plan so the client team can run and maintain it after the consultant leaves. Good handoffs include training sessions, clear ownership of each task, and a record of known limitations. Without them, models can degrade quietly.

Question 34

What is model drift and why should teams monitor for it?

Accepted Answer

Model drift is the decline in a model's accuracy as incoming data shifts away from the training data. Teams monitor for it by tracking input distributions, prediction quality, and outcomes, and by setting thresholds that trigger review. Early detection allows retraining or redesign before poor predictions affect decisions.

Question 35

What ethical risks should a consultant assess before using personal data?

Accepted Answer

The main risks are using data beyond its original purpose, causing harm to individuals, and producing unfair outcomes for particular groups. Privacy protections such as minimizing fields, limiting access, and removing identifiers should be planned early. Consultants should also flag uses that go beyond what individuals would reasonably expect.

Question 36

How can a consultant build trust with a client over time?

Accepted Answer

Trust grows when consultants deliver on commitments, report bad news early, and admit the limits of their work. Regular check-ins that share progress and open questions keep the client informed and involved. Consistency between what is promised, what is delivered, and how results are described matters more than any single impressive presentation.

Question 37

What does a consultant do when findings contradict a client's preferred conclusion?

Accepted Answer

The consultant should present the findings honestly, explaining the evidence and its limits instead of softening the result. Reviewing the analysis privately first helps catch errors and invites alternative explanations. Once the work is checked, the client decides how to respond, but the conclusion should not change for convenience.

Question 38

How can a team evaluate whether a consulting project delivered its intended impact?

Accepted Answer

Compare the outcome measures against the success criteria and baseline agreed at the start, using the same definitions and data sources. Look for whether the client actually adopted the recommendation, because a correct analysis that goes unused has little impact. Schedule a follow-up review once results have had time to appear.

Question 39

Why is it important to agree on data ownership and access rights?

Accepted Answer

Clear agreements on data ownership and access determine who may use the data, for what purposes, and who is responsible for its security and accuracy. Ambiguity can delay projects and create disputes about results. Writing these terms into the engagement agreement before data is shared prevents most of these problems.

Question 40

What role does communication play in managing client expectations?

Accepted Answer

Communication sets expectations about timelines, uncertainty, and what the project can realistically deliver, so the client does not assume outcomes that the work cannot support. Regular updates, written summaries of decisions, and early warnings about delays keep expectations aligned as conditions change. Well managed expectations make hard findings easier to deliver.

Data Science Consulting Flashcards

1Foundations of Consulting Engagements

2Problem Framing and Scoping

3Data Preparation and Analysis Planning

4Modeling and Communicating Results

5Implementation, Ethics and Client Relationships

Learning Resources