DISQUS

Andrew Chen (@andrew_chen): The first 6 steps to homegrowing basic startup metrics

  • dmourati · 1 year ago
    Bangin! Checked out Splunk?

    http://www.splunk.com
  • Andrew Chen · 1 year ago
    yeah, splunk is really rad...
  • Andrew Chen · 1 year ago
    splunk is neat since it's somewhere in-between writing your own log parsing and buying off-the-shelf reporting. Comes with a nice query language, etc.
  • Pierre Henri Clouin · 1 year ago
    Very interesting roadmap to building a robust web analytics capability.

    Regarding step 6 about data overload, have you looked into CEP (complex event processing) to filter out the noise and process relevant data in real time? It is also useful not only for reporting but also to take advantage of a real time feedback loop.
  • Andrew Chen · 1 year ago
    My inclination on this stuff, especially in the early product stages, is to keep things as simple as possible and make sure you're spending the bulk of your attention on product with analytics as a important 2nd priority. So I tend to architect initial analytics systems very simplistically.
  • spanky · 1 year ago
    On top of that, generally the first thing I do from an acquisition standpoint is to set up a landing page system - where you can dynamically create new landing pages that write to a field in the database called acquired_by. For all traffic landing on that page, you can easily calculate CPA on each campaign by querying the database against each unique acquired_by. Then over time you can add reports that will give you the CPA, activation rate, viral rates, retention rates, etc. against each acquired_by source.
  • Andrew Chen · 1 year ago
    Yep, you definitely want to capture both the source as well as potentially referrer URL - great idea.
  • Jeff Lawson · 1 year ago
    Good ideas Andrew. WRT your mysql table, I've found that storing raw data (for later analysis) can potentially become an application bottleneck, as that table will grow large and probably be abused in ad-hoc reporting. That's why we're using Amazon's SimpleDB to throw lots of data at. We can pull it back down later into optimized table formats, or write reports direcly from their API, but we'll never worry about the growth and scale of that database, and we'll be free to over-log, instead of losing data.
  • Andrew Chen · 1 year ago
    yeah, I think it just depends on what phase of the business you're at. Early on, I think that you want to optimize for easy investigation of ideas, which is why I prefer SQL. If the tables get large, you can always just rename the table and start writing into a new one (believe it or not this works fine and we did it with multi-million row tables). Later on, I agree that some of the more exotic technologies can work - but you want to make sure you're not prematurely optimizing.
  • davemc500hats · 1 year ago
    wow, andrew...

    in the infamous words of Wayne & Garth:

    "WE'RE NOT WORTHY! WE'RE NOT WORTHY!"
  • Jesse Farmer · 1 year ago
    I generally invest zero time into cute visualizations and graphs, and prefer to read the key numbers.


    I don't know about "cute," but I think this is a mistake. Good visualizations are hard to make, but they remove unwanted detail and clearly reveal the nature of the underlying data in a way that's easy to understand.

    Doing it for every little report is wasted effort, but not for key metrics -- especially metrics you might want to show to other people some day (say, investors).
  • Ed Baker · 1 year ago
    awesome post, andrew!!!
  • daveschappell · 1 year ago
    Andrew -- this was fantastic -- I got it from a Dave McClure tweet, and it's one of the rare posts that I've actually read top-to-bottom in weeks! Your flow was precise, and has mirrored our progress almost perfectly -- we're trapped right now in a little bit of step 4-5 madness.

    Thanks,

    Dave (Founder and CEO - TeachStreet)
  • Engago team · 1 year ago
    You're quite right: one needs to measure in order to know.
    Still metrics about traffic are not leads. One can have traffic that don't convert to leads.
    It is leads you need to make sales in B2B.
    Get the company names of your website visitors. These "warm" companies you can call.
  • Andrew Chen · 1 year ago
    spam
  • Pete Mauro · 1 year ago
    When I suggested this post I had not idea you were gonna blow it out like this. Awesome work Andrew!

    Regarding #3 User flows: I would love some more detail about collecting and reporting data as it pertains to user flows. GA does a decent job with funnels but I have found it limiting. How do you approach collection and reporting and what tools do you use?
  • Andrew Chen · 1 year ago
    all in house, unfortunately... I agree that GA sucks ;-)
  • andrew korf · 1 year ago
    Thanks for excellent clear concise post. Really helpful and valuable for those of working on starting something.
  • Leila Boujnane · 1 year ago
    Bloody awesome blog post Andrew. I am sitting between step 2 and 3 for one of our products so this was amazingly timely. Thanks!
  • Poornima · 1 year ago
    I just finished implementing an internal tool to help with analytics, and its hella slow, and yes I'm querying slaves, but its the production db. I'm going to think about your suggestion to move it into another data driven db for analytics. Thanks!