Jumping out of Amazon Data Science
Hi blinders, I've finally managed to escape after about a year. Just wanted to post my experience of Amazon (Supply chain/logistics) data science. First off, if you really want to join Amazon as a Data Scientist, apply for an Applied Scientist position. In Amazon, a DS is expected to have the science knowledge of an AS but no programming skills. Only SQL. Put simply, Amazon DS == world Data Analyst++. Amazon AS == world DS. You do actual data science as an Applied Scientist. Because the DS requirements are so weird, you basically get any random project that the bosses need to get done. Which, given the horrible data infra, means spending about 99% of your time writing and editing SQL...and then converting them to excel, because supply chain bosses are dinosaurs who think big data = big excel sheets. The few DS that I saw doing good work on the science side were mostly asked to replicate another researcher's work. Probably the most demotivating part is that you will be 1 of 20 teams trying to solve the same science problem (using SQL) because Amazon is that big and it can afford to do that. The requirements from PMs go along the lines of "build me a prediction model with unlabeled data, that also tells me why it predicted it, what part of the business is causing problems and also auto fixes it. Also, document all the fixes so that we get promoted.". And that request will come from a person whose title would be Head of ML/Data/Research. If it predicts wrong, then model performance == your performance rating. Most DS managers come through non-science backgrounds (because "no programming skills required"). Many are BIEs who convinced their L7s to change their titles to DS to be more marketable to the outside world. Good for them, but you get stuck with DS managers who have never built any models, let alone production level models on large datasets. What you get is a manager who asks you to go from zero to production in 2 weeks on Amazon sized data, forcing you to build very basic models that runs quuckly (think OLS...random forests ≠ Frugality on AWS resources) on millions of data points. You never build upon that knowledge because there isn't any time (unless you aim to get PIPed). Needless to say, this was my observation in the supply chain side. Life could possibly be greener in other orgs. I left the company because the culture was dull. Most of the work was busy work, maintainence overhead, 100s of meaningless document reviews (->this puts gov bureaucracy to shame on time wastage), non-existent engineering culture/norms, etc. As a DS, a big company like Amazon does not need you. It has fulls orgs to deal with any of the more interesting problems. Only way to get into these orgs is by being an AS. Side note: As an AS, there is a slightly higher probability of doing some good work. But it doesn't guarantee that you end up doing SQL. I had a friend who got hired as a Research Scientist. His "research" was to dig into 3000 lined SQL codes to identify bugs. I finally got out (covid slowed my departure). Actually happy to not be doing sql and building shitty recommender systems instead (at least I get to build them ;)). My personal recommendation for new DS would be to join a small company whee you can actually learn to build even basic things. I bet you that you will not improve your knowledge at amazon. New TC: 140k YOE: 3.5yrs #datascience #data
Do you have a masters degree? I’m interviewing for BIE since I didn’t think I could get a DS interview. Same YOE as you but no masters. I’m happy to just do sql for now since I only know the basics of ML
Lol just go for it DS in Amazon. Apply to a few different ones. I do have a masters (econometrics). You'd rather get a masters and get into DS rather than try BIE -> DS. 2 years learning it directly vs hoping to imbibe it via SQL???
I’d like to get an online masters. Georgia tech has masters in comp sci and one in business analytics. Probably one of those two. I just want to land the best job I can before starting though
But that's just like.. your experience. We have data scientist in our team who work along with Research Scientist and Applied scientists. Our manager comes from mangament background but has years of experience managing data science projects and is well versed with the required know-hows. Not everyone applying for a data scientist role is eligible for applying for applied scientist. Specially if it's an external hire.
Agreed that it's my experience. That's what blind is for :)....you get an idea of what bad experiences look like.
How is AS different from DS, or RS at amazon?
Nonsense. Just because your team's data scientists were sql monkeys doesn't mean that's the norm at amazon.
I do point out that it's logistics. I have heard that AWS and some other orgs have it better.
Yes but then you make broad statements about how people should be applying to AS roles in general, which can be interpreted as AS is the solution to a systemic Amazon problem.
My experience at Amazon 1. Team experiences wildly differ 2. The SQL and analytics part is what a typical DS does in most companies. The heavy ML is done by applied scientists everywhere 3. Agree on the PM part- at amazon PMs have an ungodly influence on everything- fetch me data, pull me data etc. PMs should be business partners not dictating the tech work. Titles are also too easy too change at amazon.
On the last point - in my experience it’s very hard to change roles to AS or RS if you’re a DS. Most orgs require full interview loops or will prefer an external hire rather than an internal DS.
Yeah most AS are professors and researchers. I’m not sure people understand how high the bar would be for a non PHD to cut it
AWS has actual data science to be honest. I don't know about CDO but everyone in my org who works as a Data Scientist ends up spending a fair bit on all parts of the process, from extracting with SQL, to modelling, to deploying these models
This is exactly like Data Science at Facebook btw
I would disagree. Talented and ambiguous DSs can do interesting technical stats or ML work (i.e. stand up production ML models, build product simulations, etc) or do technical sophisticated product analytics (i.e. statistical meta-analysis to redesign our ranking objectives). I personally did a lot of programming and drove the development of ML frameworks. My work was impressive enough to get the attention of one of the directors of engineering. While FB DS is sold as a product analytics role, the bottom up culture means that it can be as sophisticated as the DS’s ambition and talent permits.
Maybe depends on the org, but I was there for years and it's mostly product analytics. I fought my leadership hard to do anything else.
congrats for leaving! same boat here
My manager comes from a bie background as well. But his approach is purely scientific. When working with PMs definitely we try to focus on explainable models that could seem rudimentary but our team owns data that we generate with sophisticated models. Scaling and setting up prod deployment pipelines is an expected part of the role.
Dude, first of all thank you for sharing the experience, the community appreciates it! I am on the other side of the fence, where the team was new/ inexperienced that we try everything new in the market- Algorithm/Process wise. You get to learn a lot but it’s chaos all the time! I hope you find what you are looking for in the new role. Btw which company are you heading to ?
Small company. Can't name it as I'll give myself up. As for new tools, everyone is open to trying tools. We worked with spark, dask, tf, everything under aws...that was definitely a learning experience.