News for hackers. Stuff that matters.
Re:Sales of Bingo Card Creator Per Month by patio11 on Tuesday May 18, 2010 @07:49AM (#1357169)
Consolidated conversion rate is a poor metric because it doesn't help me make decisions. If it goes up, is that because of something I did on-page, or is that because of something that one of my competitors did, or is that because of market seasonality, or is it because of a change in the Google algorithms which sent a different mix of people to me? I don't know, so looking at it tells me nothing.

See also: consolidated bounce rate.

The only place I look at conversion rates these days is in the context of either two comparable traffic sources (like two AdWords creatives) or across an A/B test.

Re:Sales of Bingo Card Creator Per Month by paraschopra on Tuesday May 18, 2010 @07:54AM (#1357178)
Well, I think if your conversion rate is going down or up, something has changed and you must investigate what has changed. Agreed that conversion rate, by itself, doesn't yield much information but compared to past or across segments, it can yield a lot of insights.
Re:Sales of Bingo Card Creator Per Month by patio11 on Tuesday May 18, 2010 @08:30AM (#1357245)
Well, I think if your conversion rate is going down or up, something has changed and you must investigate what has changed.

I disagree, strongly.

1) It is highly likely your conversion rate will bounce around without regard to what you're doing.

http://images1.bingocardcreator.com/blog-images/hn/conversio...

Just eyeballing that, can you pick the most critical time for me to have checked conversion rates in the last year? (It might -- and I stress might -- have saved me $3.5k.)

2) I think humans are exceptionally good at retrofitting explanations to data, even when the explanation and the data are unrelated.

3) Rather than continually reacting to minor changes in the conversion rate, deceiving myself as to their cause, and then implementing something that probably doesn't matter, I'd rather schedule this sort of work as part of a process: write A/B test, check conversions (with all extraneous sources of noise filtered out), implement winning alternative as appropriate, get back to work.