Start free trial
Trending New Genres Authors
EnglishEnglish
EspañolSpanish
简体中文Chinese
繁體中文Chinese (Traditional)
FrançaisFrench
DeutschGerman
日本語Japanese
PortuguêsPortuguese
ItalianoItalian
한국어Korean
РусскийRussian
NederlandsDutch
العربيةArabic
PolskiPolish
हिन्दीHindi
Tiếng ViệtVietnamese
SvenskaSwedish
ΕλληνικάGreek
TürkçeTurkish
ไทยThai
ČeštinaCzech
RomânăRomanian
MagyarHungarian
УкраїнськаUkrainian
IndonesiaIndonesian
DanskDanish
SuomiFinnish
БългарскиBulgarian
עבריתHebrew
NorskNorwegian
HrvatskiCroatian
CatalàCatalan
SlovenčinaSlovak
LietuviųLithuanian
SlovenščinaSlovenian
СрпскиSerbian
EestiEstonian
LatviešuLatvian
فارسیPersian
മലയാളംMalayalam
தமிழ்Tamil
اردوUrdu
Searching...
SoBrief
Build a Career in Data Science

Build a Career in Data Science

Build a data science career around your strengths, from first application to work people use.
by Emily Robinson 2020 250 pages
4.33
192 ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Build a data science career around a specialty, not a mythical all-purpose skill set

Three skills, different mixes. Data science combines statistical reasoning, programming and databases, and business understanding. A good practitioner needs working ability across all three, but rarely expert mastery of each. The balance determines the work: analysts organize data into recurring reports, machine learning engineers build models that run in products, and decision scientists use analysis to recommend business choices. Titles are unreliable guides, so inspect what a role actually produces.

Translate between worlds. A store-location project might combine customer addresses with census information, then turn the result into a recommendation for executives. The hard part is not only the calculation. It is deciding what question the data can answer, checking whether the data means what it appears to mean, and explaining the result in language the business can act on.

Analysis

The useful distinction is between a shared foundation and a single idealized profile. That resembles the division of labor in mature engineering teams: people develop depth in different specialties while coordinating through common interfaces. It also makes career planning less punishing. A person with strong statistical judgment and domain knowledge can complement, rather than imitate, a software-heavy colleague. The caution is that employers sometimes use broad job descriptions to disguise understaffing. A role asking one person to own analytics, experimentation, data engineering, and production ML deserves scrutiny, not automatic admiration.

Choose a learning route that fits your starting point, then publish proof

Four routes in. The book compares graduate study, bootcamps, learning through a current job, and self-teaching. A data science master’s may take two years and cost $70,000 or more, while a bootcamp typically compresses learning into 8 to 15 weeks and costs about $15,000 to $20,000. Work-based learning is lower risk when a team can offer real tasks and mentorship. Self-study is flexible, but requires unusual persistence and a way to get feedback.

Show the work. A portfolio can pair organized code on GitHub with a blog post that explains the question, methods, and result. The authors’ examples range from matching freelancers to jobs to training a small neural network on Arizona license plates rejected as offensive. The playful project built practical confidence, and public documentation let others reuse it.

Analysis

This is a decision about constraints as much as curriculum. Adult-learning research broadly supports practice, feedback, and meaningful tasks, which helps explain why a real project can teach more than passive course consumption. A portfolio also makes skill legible when a résumé cannot. Still, public work is not a universal substitute for credentials: regulated employers and research roles may have formal requirements, and not everyone has spare time or safe access to publish. The strongest route is the one that produces credible evidence without creating unsustainable debt or workload.

Judge employers by their data foundations and working culture, not their glamour

Company size changes the job. A giant technology company may offer experienced colleagues and mature support, but also bureaucracy and specialized teams. A traditional retailer may give a new data scientist influence while saddling them with legacy systems. An early startup can offer rapid learning and modern tools held together by fragile processes. A late-stage startup may balance support and autonomy, while a government contractor may offer security alongside slow change and strict access controls.

Ask about the plumbing. Robert Chang described taking a role expecting visualization and modeling, then discovering that the immediate work was data pipelines and recurring reports. He recommends asking whether a data warehouse and data engineering team exist, how product events become usable tables, and who will mentor the new hire. Those answers reveal whether the job is analysis, infrastructure building, or both.

Analysis

The book treats organizational context as part of technical fit, a point often missed by candidates dazzled by a company’s brand or tools. In organizational design, infrastructure determines which work is feasible, while incentives shape which work gets rewarded. A sophisticated model cannot compensate for absent or inaccessible data. Yet archetypes are only starting hypotheses: a strong team inside a conservative company may be more innovative than a chaotic startup. Candidates should use these patterns to form questions, then test the answers against the specific team.

Treat every application and interview as a two-way fit test

Search by work, not title. Similar jobs may be called analyst, researcher, or machine learning engineer, and the same title can mean very different things. Job descriptions are often wish lists: if a candidate meets roughly 60 percent of the requirements and has relevant transferable experience, applying may be sensible. A broad list of unrelated responsibilities, no explanation of the role, or a mismatch between demanded skills and actual duties can signal trouble.

Make evidence easy to see. A concise résumé should emphasize relevant outcomes, ideally with scope or measurable impact, and use wording that maps to the posting. A tailored cover letter can connect prior experience to the employer’s needs. Referrals and professional conversations can help an application reach a person. Interviews test skills, collaboration, and follow-through, but also give candidates a chance to ask about team priorities, quality checks, and working practices.

Analysis

The two-way frame corrects a common power imbalance in job seeking: applicants behave as though selection is a verdict on their worth, rather than a matching process under incomplete information. Structured interviews can improve consistency, but they can also reward rehearsal or trivia unrelated to the job. Asking how the team handles messy data, project uncertainty, and review processes is therefore not small talk; it probes the actual work system. The advice to apply before feeling perfectly qualified is especially valuable when job descriptions overstate requirements.

Negotiate the whole life of the offer, not just its salary

Slow the decision down. Once an offer arrives, ask for the full terms in writing and reasonable time to review them. The book recommends negotiating respectfully because an employer has already invested time in selecting the candidate. Salary is only one part of compensation: stock, a signing bonus, moving support, retirement contributions, and health costs can change the real value substantially.

Count what shapes daily life. Flexible hours, remote work, conference funding, education benefits, and an earlier performance review may matter more than a small salary increase. When comparing offers, weigh lifestyle, learning, and the organization’s values alongside pay. A higher salary in a more expensive city may leave less disposable income, while a role with better mentorship may strengthen future options. Ask for agreed terms to be documented, and do not negotiate for conditions that would not make the offer acceptable.

Analysis

This approach resembles total-reward analysis in labor economics: compensation is a bundle, and its components differ in certainty, liquidity, and personal value. It also recognizes that career decisions are embedded in lives with caregiving, health, geography, and financial obligations. The book’s confidence about negotiation is useful, though bargaining power is not evenly distributed. Some employers have rigid pay bands, and bias can affect how assertiveness is received. Preparation, market information, and a clear ranking of priorities help make negotiation strategic rather than performative.

Learn the company’s data before promising answers or chasing quick wins

Ramp up deliberately. A large employer may provide formal onboarding but take weeks to grant access. A small company may expect immediate results while offering no standards or documentation. In either setting, early success begins with a manager conversation about priorities, decision-makers, and what good performance means. Simple descriptive questions, such as how customer activity changes each week, can create value while teaching a newcomer how the data works.

Investigate provenance and quirks. A future-dated subscription might represent a paused customer returning, not a new sale. Test accounts, unusual promotions, and manual data entry can distort results. Read existing documentation, inspect samples, ask domain experts, and record discoveries where colleagues can find them. If you are the first data scientist, technical choices also become precedents, so favor maintainable conventions without building elaborate infrastructure for a team that may not arrive soon.

Analysis

This is an argument for epistemic humility: data is not a neutral window onto reality, but a trace left by systems, policies, and people. The same row can mean different things to finance, product, and customer support. Learning those meanings is closer to fieldwork than to simply querying a table. Early descriptive work is also a low-cost way to build trust. The tension is that organizations may demand speed before access or context is ready. Naming that dependency early protects both the analysis and the employee from impossible expectations.

Turn business requests into scoped analyses people can understand and reuse

Start with the decision. A stakeholder may ask why sales fell, but that is not yet an executable data question. Clarify who needs the answer, what decision it will inform, and whether relevant data exists. Then write an analysis plan: a concrete sequence of questions and tasks that can be checked off. Share it with the requester and manager so changes in scope become explicit rather than silently expanding the work.

Prefer useful over impressive. Clean only what the question requires, explore with plots, and use the simplest method that answers the question clearly. A complex model is not an upgrade if it obscures the reasoning or delays a decision. Build the final output for its audience, keep code reproducible, and check in before the end. A well-made analysis answers the question, can be shared on its own, and can be revisited when the question returns.

Analysis

The analysis plan is a lightweight contract between technical work and organizational intent. It limits scope creep while preserving room to revise assumptions when the evidence changes. This resembles iterative design: show a workable version early, gather feedback, and improve the product around real use. The emphasis on simplicity is not anti-technical; it is a demand that complexity earn its cost. Some questions genuinely require sophisticated methods, but complexity should follow from the decision problem, not from a desire to demonstrate expertise.

A model matters only when it survives production and helps its users

Production is a different promise. A model used in a one-time analysis can ignore rare oddities; a live service cannot casually crash when an input is blank, malformed, or unexpected. Production work includes making the model accessible to other systems, often through an API, documenting its inputs and outputs, testing edge cases, and deploying it on infrastructure that stays available.

Maintain the service. Teams need to monitor errors and performance, plan for retraining as data changes, and manage updates safely. Accuracy alone is not enough. The book notes that Netflix paid $1 million for a winning recommendation algorithm but did not use it because its operational complexity outweighed the accuracy gain. A simpler, dependable model can create more value than a more accurate model that nobody can maintain.

Analysis

This is the difference between a laboratory result and an engineered service. Reliability engineering asks not merely whether a component works, but how it behaves under failure, variation, and maintenance. That shift changes what counts as model quality: latency, robustness, observability, and the cost of intervention become part of the product. The Netflix example makes the trade-off memorable, though it should not be read as proof that complex models are generally wasteful. The right complexity depends on measurable user benefit and the organization’s ability to support it.

Discover what stakeholders need before building what they request

Requests conceal problems. Business partners, engineers, executives, and managers rely on data science for different reasons. Ask what decision or workflow the work is meant to change, communicate progress and delays, and adapt the technical explanation to the audience. A stakeholder who rejects a result may lack context, disagree with an assumption, or face incentives the analyst has not understood. Frequent, plain-language updates help surface those issues while there is still time to respond.

Build for actual use. A retention model once produced probabilities that customers might leave, but call-center agents needed guidance about what action to take. The model answered the technical question while missing the users’ practical need. Prioritize work by both business impact and novelty: routine analysis can be valuable, while a clever model without a user or decision may be wasted effort.

Analysis

This connects data science to user-centered design: the person requesting a tool may not be the person using it, and a technically correct artifact can still fail at the point of use. Interviews and workflow observation help expose the gap between stated requests and underlying needs. The framework also guards against mistaking novelty for value. Still, user preference is not the only criterion: stakeholders can request biased or harmful outcomes, and data scientists retain responsibility for sound methods and honest reporting. Collaboration means understanding incentives, not surrendering judgment.

Design projects to fail early, learn cheaply, and stop when evidence says stop

Three common failure points. Projects often collapse because the required data is unavailable or unusable, the data contains no predictive signal, or the intended users do not want the result. More elaborate algorithms cannot manufacture a relationship absent from the data. Start with basic feasibility checks, use early decision points, and define what success looks like before investing months of work.

Recover value from failure. A team may document what happened, salvage useful data work, pivot to a better question, or end the project. A blameless postmortem examines how the process failed without turning the discussion into a search for a scapegoat. The authors compare data science to treasure hunting: the work explores uncertain ground, so not every expedition finds something. Failure is painful, but it is not automatically evidence of poor skill.

Analysis

Treating projects as experiments aligns effort with uncertainty. In research and development, a negative result can be valuable if it rules out an assumption before that assumption becomes an expensive product commitment. Early stopping also reduces sunk-cost pressure, the tendency to keep investing because much has already been spent. Yet failure should not become an all-purpose excuse: teams still need to examine whether poor scoping, weak measurement, or avoidable execution errors contributed. Psychological safety makes that honest diagnosis more possible, not less necessary.

Use the data science community to grow, but keep participation sustainable

Make your work visible and useful. A blog post, conference talk, or open-source contribution can deepen understanding, provide feedback, and help other people find your work. A small number of clear posts can be useful even without a large audience. Beginners can contribute documentation or fixes before attempting major code changes. These activities can create connections and opportunities, but they are optional; a strong career does not require constant public output.

Protect the long game. Community work can build skill and belonging, especially for people without close colleagues in a specialty. It can also consume nights and weekends. The authors warn against comparing one’s pace with prolific peers and recommend reducing activity when it becomes draining. Public work should support a career and a community, not become another impossible productivity contest.

Analysis

Public artifacts function as both learning tools and durable signals: they let others assess reasoning without relying entirely on pedigree or job titles. Open-source participation adds collaborative practice and exposes contributors to review and maintenance. There is also a collective benefit, since documentation and accessible teaching lower barriers for newcomers. But visibility is unevenly rewarded, and online participation can expose people to harassment or pressure. The valuable principle is contribution, not personal branding at any cost; quiet, local, or asynchronous participation can be equally legitimate.

Choose your next career step by the work you want to do each day

Seniority changes the job. A junior data scientist completes defined tasks with guidance; a senior colleague scopes work and handles ambiguity. Beyond that, common paths diverge. Managers set direction, hire, resolve team problems, and coach people, but spend little time doing analysis themselves. Principal data scientists remain technical leaders who solve difficult problems and mentor others. Independent consultants gain autonomy but must also sell, market, and run a business, with less income security.

Keep reassessing fit. If learning stalls, discuss new responsibilities with a manager before assuming departure is the only answer. When leaving, search strategically, communicate notice directly, and document projects, data quirks, and unfinished work so colleagues can take over. A career move is not a permanent verdict. The right path depends on the kind of problems, relationships, and responsibilities that fit a person’s current life.

Analysis

The chapter resists the familiar promotion ladder in which management is treated as the only respectable destination. That matters because technical depth, people leadership, and entrepreneurship reward different temperaments and impose different costs. Career identity is often tied to a title, while satisfaction is shaped more by daily activities and control over work. A useful decision therefore asks what the role requires on an ordinary Tuesday, not merely what status it confers. The same path can also become attractive or unattractive as family, health, and financial circumstances change.

Analysis

The book’s strongest claim is that data science is not primarily a contest in algorithmic sophistication. It is a chain of translation: from a business concern to a tractable question, from imperfect data to a defensible result, and from that result to a decision or functioning product. The career advice follows the same logic. A candidate must translate prior experience into evidence employers can recognize; a new hire must translate local data conventions into reliable analysis; a technical specialist must translate model behavior into a service other teams can operate.

That framing explains why the authors give so much space to communication, expectations, and stakeholder relationships. These are not decorative soft skills added after the technical work. They are mechanisms for reducing uncertainty and coordinating action across people who have different expertise and incentives. The treatment of project failure is especially coherent with this view. If a project is exploratory, then feasibility checks, early feedback, and stopping rules are not signs of pessimism. They are sound ways to allocate limited resources.

The book’s practical limitation is its 2020 setting and its U.S.-centered employment advice. Tool names, job titles, interview norms, and compensation practices change, and advice about education costs or negotiating may not travel unchanged across countries, industries, or economic conditions. Its company archetypes are also deliberately simplified. Their value lies in helping readers ask better questions, not in predicting a particular workplace from its size or sector.

The enduring lesson is to treat a data science career as applied judgment, not credential collection. Technical competence opens the door, but usefulness depends on choosing the right problem, knowing what the data can support, and making the result usable by someone else. The ideal practitioner is not a universal expert. It is a capable specialist who can collaborate across boundaries, recognize uncertainty, and keep learning without confusing constant novelty with impact.

Summary in 30 Seconds
Data science combines statistical reasoning, programming, and business understanding, but practitioners can specialize rather than master all three. Choose training to fit your starting point: a degree may cost $70,000 or more, while boot camps often cost $15,000 to $20,000. Employers need not-hype-ready models, but reliable tools people can use. Check data quality, user needs, and predictive signal early, then stop or redirect projects that lack them. Roughly 60 percent of job requirements may be enough to apply.
Contains spoilers

Last updated:

Report Issue
Want to read the full book?

Glossary

decision scientist

Analyst who guides choices

A data science role focused on turning company data into recommendations for human decision-makers. Decision scientists combine statistical reasoning, programming, business context, and clear communication. Their work often answers a specific question through an analysis or experiment, then explains what the findings imply for a business choice.

analysis plan

A scoped work checklist

A written outline made before analysis begins. It states the objective, breaks the business question into answerable questions, and lists concrete tasks such as data checks, summaries, visualizations, or models. Sharing it with a manager and stakeholder creates agreement about scope and makes changes visible.

stakeholder

Person influencing project outcomes

A person, group, or organization involved in a project, affected by its outcome, or able to influence it. In data science, stakeholders may request an analysis, use a model, provide technical support, or set priorities. Their goals and knowledge differ, so the data scientist must understand the intended use of the work.

blameless postmortem

Review focused on learning

A structured discussion held after a project or system failure to understand what happened, why it happened, and how similar problems might be prevented. The emphasis is on contributing processes and conditions rather than identifying an individual to punish. The approach aims to make candid reporting safer and improve future work.

minimum viable product

Smallest useful test version

An early, limited version of a product or analytical tool that can be shown to its intended users to gather feedback. In data science, a basic model or preliminary analysis can reveal whether the output is useful before a team invests heavily in refinement, automation, or deployment.

principal data scientist

Senior technical leader

A highly experienced individual contributor who guides technical strategy, tackles difficult data science problems, and mentors other practitioners. Unlike a people manager, a principal data scientist generally remains close to technical work, but their role extends beyond completing individual analyses or models to helping a team solve problems well.

CI/CD

Automated build and deployment

Continuous Integration and Continuous Deployment are software practices that automate checks and delivery when code changes. A CI process can run tests after code is committed. A CD process can then move approved code into a production environment. Together, they reduce manual deployment work and help catch errors before users encounter them.

Download PDF

To save this Build a Career in Data Science summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.28 MB     Pages: 17

Download EPUB

To read this Build a Career in Data Science summary on your e-reader device or app, download the free EPUB. The .epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.
Download EPUB
File size: 1.27 MB     Pages: 22
Want to read the full book?
Follow
Listen
Home
Swipe
Library
Get App
Try Full Access for 3 Days
Listen, bookmark, and more
Compare Features Free Pro
📖 Read Summaries
Read unlimited summaries. Free users get 3 per month
🎧 Listen to Summaries
Listen to unlimited summaries in 40 languages
—
❤️ Unlimited Bookmarks
Free users are limited to 4
—
📜 Unlimited History
Free users are limited to 4
—
📥 Unlimited Downloads
Free users are limited to 1
—
Risk-Free Timeline
Today: Get Instant Access
Listen to full summaries of 26,000+ books. That's 12,000+ hours of audio!
Day 2: Trial Reminder
We'll send you a notification that your trial is ending soon.
Day 3: Your subscription begins
You'll be charged on Nov 25,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
1M+ readers
Trustpilot Rating
TrustPilot
4.6 Excellent
This site is a total game-changer. I've been flying through book summaries like never before. Highly, highly recommend.
— Dave G
Worth my money and time, and really well made. I've never seen this quality of summaries on other websites. Very helpful!
— Em
Highly recommended!! Fantastic service. Perfect for those that want a little more than a teaser but not all the intricate details of a full audio book.
— Greg M
Save 62%
Yearly
$119.88 $44.99/year/yr
$3.75/mo
Monthly
$9.99/mo
Start a 3-Day Free Trial
3 days free, then $44.99/year. Cancel anytime.
Unlock a world of fiction & nonfiction books
26,000+ books for the price of 2 books
Read any book in 10 minutes
Discover new books like Tinder
Request any book if it's not summarized
Read more books than anyone you know
#1 app for book lovers
Lifelike & immersive summaries
30-day money-back guarantee
Download summaries in EPUBs or PDFs
Cancel anytime in a few clicks

Why one summary?

Scanner
Find a barcode to scan

⭐
⭐
⭐
We have a special gift for you
Open
38% off
DISCOUNT FOR YOU
⭐
⭐
⭐
$79.99
$49.99/year
Only $4.16 per month
Continue
2 taps to start, super easy to cancel
Settings
General
Widget
Loading...