Related Posts
More Posts
Prime video catalog specialist job salary?
Anyone into option selling ?
Can anyone shed some light on quant funds please? https://groww.in/mutual-funds/escorts-tax-plan-direct-growth
I am seeing stellar returns and extremely low expense ratios but apparently since they're quantitatively managed algorithms, they don't account for things like corona for example.
Additional Posts in Data & Analytics Consultants
New to Fishbowl?
unlock all discussions on Fishbowl.




Well data science and data engineering are two complete different things. So which is it that you want to pursue? Figure that out and get after it. Personally, I found that I don’t like data science so I side with engineering and building out architecture.
And to add on to this.. the engineering side will make you more money in the long run
Bowl Leader
Yes, it's very common for people in BI roles to want to move toward modeling. You're actually in luck, because you have a lot more data access than other ops analyst folks trying to do the same.
Here's how you do it: be proactive.
Build a model on data you've prepped anyway, then take your potential results to your management. Make a pitch to run a pilot based on the holdout set results signaling value-added. Bonus points if you can find a willing pilot team yourself, though of course you'd have to do this internally. The only thing you need to commission is a few of your personal evenings/weekends (no one said that starting in a new field takes no extra work).
Just know when to move on: not all models will converge easily, and at this point you're looking for a MVP not a research project. I'd say no more than like 24 hours of hands-on time per model, given that you only need some minimal lift and have the data pretty much ready to go. This should be enough for you to try a few algos even if you don't have experience. If nothing's working at that point, find a different modeling question or data source.
Also, get someone with experience to proof your model AND proposed application. The main thing missed by people who've done their modeling homework is how operationalizing the information their tool generates works. People get hung up on metrics, forgetting the users, and as a result miss key metrics. The answer to "will I use the words or numbers that have been added to this report?" should be an emphatic yes. It's never about the model itself, but about the end result. Superior F-scores or AICc or whatever don't always lend themselves to better models, since they can't measure what's not reflected in your sample.
Here’s another angle: now you get to learn just the fun stuff