Hacker News Reader

Comment by 9dev | Hacker News Reader

The article is missing out a huge argument in favour of free plans: the opportunity to generate data. This doesn’t even need to be done in a shady manner - imagine analysing what users usually type in free form boxes, normalise that and build a tag hierarchy, for example. There’s a billion way to make a free plan useful - it’s not just a curtesy to your users!

View on HN

The problem is that free users aren't a representative sample of people who'll pay. I remember a 37signals post that mentioned how something like 90% of their paying users were designers, but that was only 5% of their free users. If you optimize on a sample that's structurally bad, no matter how large it is, you'll end up building the wrong thing.

Seriously!

Quantity data is not a substitute for quality data.

Give an engineer a qualitative data point and they'll find a way to invalidate it.

But give them a million invalid data points...

Nah. I had a million installs of a web widget and was drowning in data doing over a billion impressions a month and we had to pivot or starve to death. Nice idea but not true in my experience unless you already have customers lined up and are intentional about what you launch in order to generate the specific data they are hungry for.

Yeah, I've worked at places with millions of users and product managers are drowning in data. It honestly doesn't really make it much easier to build features that actually drive revenue.

To a large extent, this is why advertising (and related data selling) is so widespread: because it is obvious how to make money from the data. Someone said they'll give you $1,000 for a list of all the users who are male, in Canada, and have typed in stuff about Python.

The parent's example about tag hierarchies... That sounds like exactly the kind of nifty feature I've seen launch a dozen times that has precisely zero impact on revenue.

Fair point, although I wasn't referring to simply hoarding every data point you can get but specifically _using_ free users to improve your data set. Obviously this isn't a viable option for everyone.

Take a look at Google for a great example of this: Their "local guide" feature on Google Maps is using gamification to improve location data by convincing users to work for free; or Recaptcha using user input to train their machine learning models.

Have you (or anyone else here) done that and succeeded at it? I can see why one would assume it could somehow be useful, but in the absence of someone accomplishing it I’m skeptical.

I'm actually doing that currently. We have a large database of company records compiled from different sources. Naturally, some of those records are duplicates. We have cleaned the data as much as it was possible using algorithmic approaches, but that only gets you so far. To go the last mile, we added widgets to the pages showing similar records where users may click "same company" or "other company". Poll results are aggregated and values above a certain threshold create an entry in a verification queue for our staff to look at. This works out surprisingly well for now.

I’m glad it’s working well for your use case, especially with the final expert evaluation.

What it makes me think of is genealogy, part of which can be thought of as the process of assigning records (birth, marriage, death, census, etc) to entities (natural persons). These are complex determinations requiring understanding and correlating multiple records, and analyzing them in terms of information which is not explicit in the records (eg, knowing about relevant place names, name changes, etc). In the case of businesses, imagine trying to figure out the connection or lack of connection between the many Ray’s Pizza shops in New York City, as they open, close, and change names, locations, and owners over time.

Reply on news.ycombinator.com