A timeline honest enough to plan against

Ninety days. That is the number most transition guides quietly promise, and it is the reason so many career switches stall around week seven. A realistic part time move from infrastructure into a first data science role runs closer to nine to twelve months at eight to ten hours a week, and the honest sources agree, with general advice for engineers making this switch landing at six to twelve months of consistent effort before you are ready to apply. That estimate already assumes you write code, which you do, so it applies to you and not to someone starting from nothing. The ninety day figure is not exactly a lie. It is the full time number, the one that holds if you quit and study thirty five hours a week, and almost nobody reading this can do that without a runway of savings.

Put the honest figure on a calendar and it stops being intimidating. Nine to twelve months at eight to ten hours a week is two focused evenings and one weekend morning. It is a season and a half of your life. Against the market it is time well spent, because the field is adding seats faster than almost anything else. The US Bureau of Labor Statistics projects data scientist employment to grow 34 percent between 2024 and 2034, about 82,500 new roles on top of the 245,900 that existed in 2024, with roughly 23,400 openings a year once you count people retiring or moving on. Compare that with 3 percent average growth across all occupations and the picture is plain. You are not sprinting at a closing door, you are pacing yourself toward a field that keeps widening.

Your weekly budget sets your finish date, and the relationship bends because consistency beats intensity. Two protected evenings every single week for a year will carry you past a heroic month followed by three quiet ones, every time. Here is roughly what different weekly budgets buy.

Before you pick a pace, look hard at what the field actually pays, because one number on this chart tends to reverse a bad assumption. A first data science role often pays less than the senior infrastructure or software job you already hold, and pretending otherwise sets you up to quit in month four when the offer lands below your current salary.

Four phases mapped to a calendar

A plan you cannot check is a wish. Every phase below ends in a milestone you can hold up and say, that is done, rather than a vague sense of progress. Keep this roadmap table where you can see it, because it is the reference artifact for this whole part, the thing to return to whenever you lose the thread of where you are. A milestone works because it is binary. Either your telemetry is loaded and cleaned in a notebook or it is not. Either you have a model with an honest cross validated score or you do not.

One project runs the length of this series, and the roadmap tells you exactly when you touch it. Right now that project stands at nothing, a single export you have not pulled yet, the same place Part 3 left it. You do not pull it in phase one for its own sake, you pull it so phase two has real data to load instead of a toy dataset from a tutorial. By phase four that one export has become a served, monitored model, and by phase five it is the centre of your portfolio. Phase two also opens with the boring but load bearing work of a clean environment, and the Data Science Series covers that ground in its part on a reproducible Python setup, so you do not have to invent it here.

flowchart TD
  A[Decide target role and set up] --> G{Employed in infra now}
  G -->|Yes| B[Sideways move, keep salary, shift the work]
  G -->|No| C[Clean break, budget savings first]
  B --> D[Foundation to first model to MLOps, months 1 to 9]
  C --> D
  D --> E[Portfolio and interviews, months 9 to 12]

Notice what the diagram does not include. There is no phase for a masters degree, and that is deliberate. BLS lists a bachelor degree as the typical entry education for data scientists, with a masters preferred by some employers but not required, and a career changer with a working portfolio built on real production data will out interview a fresh graduate with none. Spend the twelve months building things, not sitting in a second degree, unless your target employers specifically gate on it.

Where an infra background moves faster, and where it drags

Your starting point is not the middle of the pack, it is ahead on some axes and behind on others, and knowing which is which lets you skip what you already own. You move faster on the whole back half of the roadmap. Phase four, the MLOps and production work that stops most trained data scientists cold, is the operations you already do, so serving, monitoring, versioning and pipelines will feel like home with new labels. You move faster on data access too, because pulling from SQL, APIs and monitoring systems is a Tuesday for you and a research project for someone who has only touched clean CSV files. Your comfort with the command line, with Git, with reproducibility, all of it counts and none of it has to be learned twice.

Where you drag is phase three, and there is no way around it. Statistics and probability are genuinely new ground for most operators, and the honest budget is that the middle of the roadmap, roughly months three to six, is slower and less fun than the rest. This is the stretch where career changers quit, not because it is impossible but because it is the first time in years they have felt like a beginner. Plan for that feeling instead of being ambushed by it. Your other drag is a mindset tax rather than a knowledge gap, the reflex to perfect a pipeline when a rough answer was the task, which Part 3 walked through in full. Budget extra calendar for phase three specifically, and treat the ease of phase four as the reward waiting on the far side.

Choosing a path, sideways move versus clean break

There are three honest ways to make this move, and the internet only ever sells the loudest one. You can quit and do a full time bootcamp, you can make a sideways move inside your current employer and shift toward data and MLOps while keeping your salary, or you can stay in your role and add data skills with no title change until the opportunity is obvious. They trade speed against risk in ways worth seeing side by side before you commit a year to any of them.

Verdict. For an employed infrastructure engineer the pick is the sideways move, and it is not close. You keep your salary through the slow months, you practise on real production data your bootcamp peers would kill for, and you convert a job you already have into a title you want without a single month of unemployment. Avoid the clean break, the quit and bootcamp route, unless you have both a runway of savings and no current tech job to protect. Leaving the moment you decide to switch throws away your single biggest asset, hands on access to live telemetry, and trades it for generic course datasets everyone else also has. Stay and add is the safe default if you are still unsure, and it converts smoothly into a sideways move the day you are ready. Pick the cheapest route that still moves you, and for most of you that is the middle row of the table.

Commit to the sideways move, not the clean break

If you do one thing after reading this, block two recurring evenings on your calendar for the next year and label them, because a plan without protected time is the ninety day fantasy in a new suit. Then spend the first two weeks on phase one only. Pick one target role and read three real job descriptions for it. Install Python the reproducible way. Identify the single export you will pull from your own systems, a month of one cluster CPU and memory metrics or a quarter of incident tickets, and get permission to use a scrubbed copy. That is phase one, and it is genuinely doable in a fortnight of evenings. Do not touch statistics yet, do not buy a course bundle, do not quit anything. Later, if you find you want to specialise toward language models rather than infra data, that is a separate track that starts with concepts like what retrieval augmented generation is, and it can wait until you have landed the first role. Next part starts the build, Python past scripting, where your one export finally gets loaded. Before you read it, put those two evenings on the calendar and name the export you are going to pull.

References

• US Bureau of Labor Statistics, Occupational Outlook Handbook, Data Scientists, pay and 2024 to 2034 projections
• US Bureau of Labor Statistics, Employment Projections program
• Data Science Series, a reproducible Python setup for data work
• Infra to Data Science, the Complete Guide

About The Author

Dr. Pranay Jha

Dr. Pranay Jha is a Cloud and AI Consultant with 18+ years of experience in hybrid cloud, virtualization, and enterprise infrastructure transformation. He specializes in VMware technologies, multi-cloud strategy, and Generative AI solutions. He holds a PhD in Computer Applications with research focused on Cloud and AI, has published multiple research papers, and has been a VMware vExpert since 2016 and a VMUG Community Leader.

See author's posts