Your First 90 Days as a Data Scientist

I DoorDash about 5 months in the past. That is my first time beginning at a brand new firm as a Knowledge Science Supervisor. DoorDash strikes quick, expectations are excessive, and the area context is deep, which makes onboarding difficult. Nevertheless, it has additionally been one of many fastest-growing durations of my profession.

The primary three months at any new job are essentially a constructing section — constructing connections, area understanding, and information data — and a clean onboarding units the inspiration for later success. Due to this fact, on this article, I’ll share what mattered most to start with months and my guidelines for any information science onboarding.

I. Construct Connections 

Earlier than the rest, let me begin with constructing connections. After I was in school, I pictured information scientists as folks spending all day lengthy heads-down writing code and constructing fashions. Nevertheless, as I grew to become extra senior, I spotted that information scientists make actual impacts by embedding themselves deeply within the enterprise, utilizing information to establish alternatives, and driving enterprise choices. That is very true in the present day with tighter DS headcount and AI automating primary coding and evaluation workflows. 

Due to this fact, constructing connections and incomes a seat on the desk must be a high precedence throughout onboarding. This contains:

• Frequent onboarding classes along with your supervisor and onboarding buddy. These are the individuals who greatest perceive your future scope, expectations, and priorities. In my case, my supervisor was my onboarding buddy, and we met nearly each day in the course of the first two weeks. I at all times got here with a ready listing of questions I encountered throughout onboarding. 
• Arrange meet-and-greet calls with cross-functional companions. Right here is the agenda I normally observe in these calls: 
  • 1. Private introductions
  • 2. Their focus space and high priorities
  • 3. How my staff can greatest help them
  • 4. Any onboarding recommendation or “issues I ought to know”
  • I particularly just like the final query because it constantly offers nice insights. 5 years in the past, once I onboarded at Brex, I requested the identical query and summarised the responses into classes here. One of the best I obtained this time is “Don’t be afraid to ask dumb questions. Play the new-hire card as a lot as doable within the first three months.”
• For these key companions, arrange weekly/bi-weekly 1:1s and get your self added to recurring undertaking conferences. Chances are you’ll not contribute a lot at first, however simply listening in and accumulating the context and questions is useful.
• If you’re onboarding as a supervisor like me, it is best to begin speaking to your direct reviews early. Throughout onboarding, I purpose to study three issues from my direct reviews: 1. Their tasks and challenges, 2. Their expectation of me as a supervisor, 3. Their profession targets. The primary helps me ramp up on the world. The latter two are essential for establishing belief and a collaborative working relationship early on.

II. Construct Area Context

Knowledge scientists succeed after they perceive the enterprise properly sufficient to affect choices — not simply analyze outcomes. Due to this fact, one other precedence throughout onboarding is to construct your area data. Frequent methods embody speaking to folks, studying docs, looking out Slack, and asking a whole lot of questions.

I normally begin with conversations to establish key enterprise context and tasks. Then I dig into related docs in Google Drive or Confluence, and browse Slack messages in undertaking channels. I additionally compile the questions after studying the docs, and ask them in 1:1s.

Nevertheless, one problem I bumped into is digging into the rabbit gap of docs. Every doc results in extra paperwork with quite a few unfamiliar metrics, acronym names, and tasks. That is particularly difficult as a supervisor — if every of your staff members has 3 tasks, then 5 folks means 15 tasks to catch up. At one level, my browser’s “To Learn” tab group had over 30 tabs open.

Fortunately, AI instruments are right here to rescue. Whereas studying all of the docs one after the other is useful to get an in depth understanding, AI instruments are nice to offer a holistic view and join the dots. For instance,

• At DoorDash, Glean has entry to inside docs and Slack. I usually chat with Glean, asking questions like “How is GOV calculated?”, “Present a abstract of the undertaking X, together with the aim, timeline, findings, and conclusion.” It hyperlinks to the doc sources, so I can nonetheless dive deeper shortly if wanted. 
• One other instrument I attempted is NotebookLM. I shared the docs on a particular subject with it, and requested it to generate summaries and thoughts maps for me to gather my ideas in a extra organized approach. It will probably additionally create podcasts, that are typically extra digestible than studying docs. 
• Different AI instruments like ChatGPT also can connect with inside docs and serve an analogous function.

III. Construct Knowledge Data

Constructing information data is as vital as constructing area data for information scientists. As a front-line supervisor, I maintain myself to a easy customary: I ought to be capable to do hands-on information work properly sufficient to offer sensible, credible steerage to my staff. 

Here’s what helped me ramp up shortly:

1. Arrange tech stack in week one: I like to recommend establishing the tech stack and developer surroundings early. Why? Entry points, permissions, and peculiar surroundings issues at all times take longer than anticipated. The sooner you could have the whole lot arrange, the earlier you can begin taking part in with the info. 
2. Make full use of AI-assisted information instruments: Each tech firm is integrating AI into its information workflows. For instance, at DoorDash, we’ve Cursor related to Snowflake with inside information data and context to generate SQL queries and evaluation grounded in our information. Although the generated queries aren’t but 100% correct, the tables, joins, and previous queries it factors me to function wonderful beginning factors. It received’t substitute your technical judgment, but it surely dramatically reduces the time to first perception.
3. Perceive key metrics and their relationships: Knowledge data not solely means with the ability to entry and question the info, however perceive the enterprise from an information lens. I normally begin with weekly enterprise critiques to search out the core metrics and their development. That is additionally an effective way to contextualize the metrics and have an thought of what “regular” appears like. I’ve discovered this extremely useful when gut-checking analyses and experiment outcomes later.
4. Get your arms soiled: Nothing enforces your information understanding greater than doing a little hands-on work. A very good onboarding program normally features a mini starter undertaking. At the same time as a supervisor, I did some IC work throughout my onboarding, together with alternative sizing for the planning cycle, designing and analyzing a number of experiments, and diagnosing and forecasting metrics motion. These tasks accelerated my studying excess of passive studying.

IV. Begin Small and Contribute Early

Whereas onboarding is primarily about studying, I strongly suggest beginning small and contributing early. Early contributions sign possession and construct belief — usually quicker than ready for a “excellent” undertaking. Listed below are some concrete methods:

• Enhance the onboarding documentation: As you undergo the onboarding doc, you’ll run into random technical points, discover damaged hyperlinks, or discover outdated directions. Not simply overcoming them your self, however enhancing the onboarding doc is an effective way to indicate that you’re a staff participant and wish to make onboarding higher for future hires.
• Construct documentation: No firm has excellent documentation — from my very own expertise and chatting with my mates, most information groups face the problem of outdated or lacking documentation. As you might be onboarding and never busy with tasks but, it’s the excellent time to assist fill in these gaps. For instance, I constructed a undertaking listing for my staff to centralize previous and ongoing tasks with key findings and clear factors of contact. I additionally created a group of metrics heuristics, summarising the causal relationship between totally different metrics we discovered from previous experiments and analyses. Notice that each one these paperwork additionally turn into helpful context for AI brokers, enhancing the standard and relevance of AI-generated outputs.
• Counsel course of enhancements: Each information staff operates in a different way, with professionals and cons. Becoming a member of a brand new staff means you convey a contemporary perspective on staff processes and would possibly spot alternatives to enhance effectivity. Considerate ideas primarily based in your previous expertise are tremendous helpful. 

For my part, a profitable onboarding goals to ascertain cross-functional alignment, enterprise fluency, and information instinct.  

Right here is my onboarding guidelines:

1. Week 1–2: Foundations
   – Meet key enterprise companions
   – Get your self added to core cross-functional conferences
   – Perceive staff focus and priorities at a high-level
   – Arrange tech stack, entry, and permissions
   – Write your first line of code
   – Learn documentation and ask questions
2. Week 2–6: Get your arms soiled
   – Deep dive into staff OKR and generally used information tables
   – Deep dive into your focus space (extra docs and questions)
   – Full a starter undertaking end-to-end
   – Make early contributions: Replace outdated data, construct one piece of documentation, or counsel one course of enchancment, and many others.
3. Week 6–12: Possession
   – Have the ability to communicate up in cross-functional conferences and supply your data-informed standpoint
   – Construct belief because the “go-to” individual in your area

Onboarding appears totally different throughout corporations, roles, and seniority ranges. However the ideas keep constant. For those who’re beginning a brand new function quickly, I hope this guidelines helps you ramp up with extra readability and confidence.