I have a lot of experience with this problem. The simplest way is via data munging in python/ pandas etc by finding what percent users convert/churn after doing an event N times within the first X days, and all the permutations thereof, using statistical tests around the change point. A more clever way is to use bayesian change point analysis.

The tricky thing is that these insights wind up being kind of obvious from the first analysis. You will find things like "users who use the software more are more likely convert." Other times these types of analysis will confirm what you already know. The tricky thing is making sure you have the right tagging/events and place to make sure you're getting at the right level of detail to get something worthwhile. It's very a much a garbage in garbage out type of thing.

link: http://oppsdaily.com/

I personally have had this problem and tried looking for that insight hidden among our big data.

but we discovered looking for patterns in quantiative data was a waste of time. Instead what we are focusing on is more qualitative data: interviewing customers, interviewing people who cancelled, and finding the "jobs to be done" of your prospective customers. then mapping that to your product features.

once you truly understand your customer needs, knowing what to fix/improve is rather trivial.

Larry Page calls this the "magic moment" at Google...

Anyway, I'm not sure you'll find what you want with data. Usually you understand what the magic moment is by putting yourself in the position of your customers and then using your own product. Or trying to use your product and identifying the moment where your reaction goes from "meh, who cares" to "oh, wait, this is really exciting!". If you're not your target user, watch your users using your product and carefully observe their facial expressions. The magic moment is entirely emotional; that's what makes it magical.

Interesting, if you like to send me some more information I'd be happy to give you some pointers (andreas (ät) 7scientists.com).

Mistake on http://7scientists.com/ueber.html it should be 'Unser' instead of 'Unsere' just an FYI

Thanks, fixed!

@gingerlime Curious to know more about your problem. Can we jump on a quick call to discuss it? (email: [email protected])

A data scientist with experience in causal inference shouldn't cost you more than $150,000 a year.

A better use of that would be to hire a consultant for a couple weeks a year to tweak (or determine, at first) your models. We did this at a market research company I worked at, an oldschool stats guy would come in once a year for a review and tune-up.

I do 'data science' for smaller businesses that would not have the resources, or really the need for a full-time in-house analytics person like me. But, one client does about 8 to 10 hours a week, another does some projects periodically, and I just added a third that I think will end up being something like a day a week.

From my perspective it spreads my risk and from their perspective it means they get access to higher-end analytical help in a more flexible package that meets their needs. Their trade-off is higher rates per hour and my trade-off is a little more variance in income.

Good point. No reason to bring someone in-house before you even know what your needs are.

Moreover, the things they would work on may not really change often enough for an in-house person to ever be worth it.

We don't have this kind of money. And I think lots of small startups face a similar problem. I hope someone would "Productize" this kind of thing (even as a niche consultancy which can serve several smaller businesses)

50k remote from continental Europe

You might look at Azure machine learning.