Show HN: Heap is a new approach to analytics. Just capture everything

If there's something wrong with your DBMS where the alter table to add a row is an expensive or dangerous operation, just add a new table for every attribute, having lots of tables in a database is no worse than having lots of variables in a program.

If your DBA gets angry at you for adding new tables all the time, get a new DBA.

I think it can actually work quite well, provided you can setup a separate service just for analytics. It's not too unreasonable to assume there are sites already doing this since the concept of a central repository that you alone control (and isn't provided by a third party) can be appealing.

Also, you have to worry less about privacy and just concentrate on keeping your own systems and services secure.

But, seriously, everyone rolls this themselves and does it wrong. If you've done it right, and can release it, you'd be doing good for the world.

http://blog.heapanalytics.com/wp-content/uploads/2013/03/blo... for example.

Now can I have an invite? ;) mattl [at] gnu org

Based on the user string, we would serve up a version of the HTML that didn't request the embedded font, and then those users would see Arial instead of, for example, Proxima Nova.

Super-annoying to have to deal with. But it works.

What? Are you asking me to justify having an option of being able to object? That's a bit too meta, with today being Friday and all.

if you object to in store tracking in physical locations, you are going to be able to shop at fewer and fewer locations (anyplace with a security camera).

Maybe, but at least here in the EU you can't use that footage for analytics or share it with any company. Even our government got CCTV project shut down because they didn't comply with Data Protection laws.

I don't advocate for Internet legislation, but I too will block it and wish people were more respectful of others' privacy.

My reaction anymore to consumer "sign-up required" sites or saas "request an invite" products is usually to close the tab, even if my interest is piqued.

I do believe that this used to be a powerful design pattern (Gmail, and as pointed out elsewhere Dropbox) but its prevalence has also become its undoing for me personally.

Start up founders often worry about becoming too popular too quickly -- when usually they should be worried about not becoming popular fast enough.

They may just not have the infrastructure to support a widely available service yet.

Most of their visualizations - while useful in their own right - are pre-built reports that serve a different role than the flexible reporting capabilities in Heap (and Google Analytics et al).

I once created a model that showed very strongly that US/China exchange rates were a driver of revenue. This model was intuitive based on the market dynamics as we understood them, and they helped provide a strong predictive power which helped reduce operational costs. Over time, this model began to be taken as irrefutable truth among quite a few people in the org. Then when the economy collapsed, we very quickly learned that it wasn't US/China exchange rates that was the strong predictor, but rather something else which correlated with those rates.

Not only was the model broken, but it had now breeched trust in the ability of model based forecasts. In other words, if I were a business, I would be bankrupt.

I think the best value-add that you can possibly provide with your service is a way of helping your customers understand what is meaningful and what isn't.

If you have an analytics software that doesn't work in IE8. You lost analytics for 20-30% of your users.

And since array.prototype.slice.call is not supported in IE8, this analytics software is about 80-70% useful. And maybe even less if you have a large IE clientele.

Seems strange to me that one would limit themselves that much.

Oh technology...

Chartbeat has never supported IE, which as you say sounds like cutting off a major part of your market. Even moreso when you consider our customers which include pretty much every major publisher in the US. However what we've learned is that if your product provides enough value, people have no problem installing Chrome or Firefox or whatever.

To this day we receive pretty much no requests for IE support at all, which enabled us to build a product with heavy HTML5 (mostly canvas) and CSS3 usage.

For my past implementation of a similar data-store we had some different requirements that let us cut a few corners. But, if I were to take another stab at it with your requirements then I would try a new approach.

I think I would use a single table to represent the 'event' itself and have that table be essentially owned by the customer. This table could be put on a specific shard set aside for each customer or you could just name it "{customerId}_{eventName}" and have a lot of tables. Then each event coming in would potentially perform an 'alter' on the table in order to make sure all custom properties have a column.

Are there any downsides to this approach I'm not considering?

Any plans for either a free trial period or a very small (<50 users?) free plan?

Anyone who'd use your service can look up their existing active user counts and plan accordingly -- much easier than estimating data points. However, you're going to lose some people if you require mental gymnastics to get an estimate based on arbitrary tiers and pro-rated pricing.

In my app, I actually track client-side actions -- essentially, clicks, but with more context. Anyway, you can treat a single user session like you would the text of an email, where the "actions" are the words.

From there, all you need to do is capture a bunch of sessions and tell the classifier which users are not strong computer users and which know what they're doing, passing the corresponding "documents".

Now you can feed in new documents and determine which of your users know what they're doing, and which aren't really computer users.

(It my experience, people fall into one of those two camps.)

PS: "tchanananam" is a brazilian expression onomatopoeia when there is a punch line revelation...

Even for a "successful" site (>1mm MAU) $2,000 a month feels insanely expensive. The performance difference between this and GA ($0) is not high enough _initially_ to offset the sticker shock.

Look at NewRelic, Tracelytics or even Mixpanel (expensive IMO) for better pricing levels. You've built the equivalent of "hosted snowplow" (https://github.com/snowplow/snowplow) with a nice GUI and tools, you should expect to charge something much closer to cost than your current approach appears to.

How does it affect network performance? If I click a link and it takes me to another page, how do you track that?