4 sales in a thousand visits - should Dave be worried?

Dave needs 10 sales out of every 1,000 visitors to be profitable. He runs a test and gets 12 sales. How can he be 95% sure that this isn't just good luck, and that the site will continue to be profitable?

---


Profitable conversion rate = 10 / 1000
Profitable conversion rate = 1%


Visit this calculator: http://vassarstats.net/prop1.html




---
Using the laws of Expected Value, we would expect:

Number of expected sales = Visit x Per-Unique-User-Conversion-Rate

We know the visits. Dave's profitable conversion rate is 1%.



Number of sales to be profitable  = 1,000 x 1%
Number of  sales to be profitable  = 10


Now this could be good, or it could be bad. Dave gets 4 sales for 1,000 visitors, should he be worried about this profit?

.
.
.
.
.
.
.


No. Dave is cool - 4 could still be profitable. 




---


How to work this out for your numbers? Visit this calculator: http://vassarstats.net/prop1.html

What do you think of this?

Why and how to segment?

Are your users all the same? Really?

Let's take a simple example. "New Visitors" are people who have never visited your site before. People who have come back - "Returning Visitors". Two built in segments.

Do these two groups behave differently with AdWords with some real data? Let's find out.

So looking at the Advertising, AdWords, Campaigns section we see this:

What to make of that data? New Visitors have a goal completion rate of about 2%, but Returning Visitors have 4.2%*.

Are Returning Visitors worth more here? Should we bid more for them? Re-target to them? All fun experiments to do and answers to find.


Answers on a postcard.


* 2% goal completion rate here is 3435 / 171917, rounded.

(Not set) - data missing - why?

You're seeing (Not set) in your Google Analytics. This is a really boring problem to have.

Here's a short list of possible reasons.

The fixable (like having your dog neutered)?

• AdWords and Google Analytics not linked at all
• AdWords and Google Analytics linked incorrectly, or costs not applied.
• Data not sent by clicks (auto-tagging not on in AdWords).

The unfixable (aka Ben Affleck's movie career):

• Data not in that report (e.g. keywords for display network, or dynamic ads).
• Data blocked by browser behaviour (e.g. SSL not sending referers).

If you are truly concerned, and you can't fix yourself, and you can call the A-team. Murdoch has a massive tool for the job, I hear.

Failing that your Google Account Manager, or friendly Google Analytics professional. Ask nicely.

What the feck is attribution?

A common question I get asked is "What's attribution?". Here's one answer - 42. OK Ok ...

.
.
.
the story begins:

Meet Aaron, he's 24, young, full of life, uses too much hair product, and works in MacDonalds.

Aaron is going to Malaga and wants to book a car.

So he searches for "Car Hire Malaga" and sees this ad.

1

He clicks on ad 1 and browses, but doesn't book.

Some time passes. Pay day comes. Horray for the big mac.

Aaron then searches for "Hertz Car Hire" and clicks on this ad 2:

2

He books his car.

Which ad gets paid? 1, 2, both* or neither?

Write your answer down on a scrap of paper.

.

.

.

.

The answer: You're wrong.

There probably isn't a true right answer.

Answering that question is Attribution.

Now go eat some french fries. Subliminal messaging works.

* And if both, split 50:50, 90/10, or other?

P.s. I have more pointless stories using Michael (he's multi channel, geddit?). And Gonzales - always in a hurray. Comment for more or less amusing versions of this.

Season history with seasonality

There's a wonderfully over-used word in statistics, and that's 'maybe'. Only kidding - seasonality. I've heard this mis-applied, never-applied, and over-applied - mostly by me. A Wikipedia definition:

In statistics, many time series exhibit cyclic variation known as seasonality, periodic variation, or periodic fluctuations. This variation can be either regular or semi regular.

Thanks Wikipedia, not!

The short, and my version: history may often repeat itself.

Why is this important? If you are trying to work out why sales went up at 3am on a given Thursday, ask yourself whether that happens EVERY 3am on a given Wednesday. And then ask why again.  You may have better things to do with your time.


Google Analytics has a fantastic 'compare to past' feature as below. 

You can then see if your acclaimed crazy sales spike was real.

Charts like this are 'probably' showing seasonality. Determining scientifically requires effort or some deeper statistical work. 

How is that helpful? Well, if you do a lot of PR activity and see a sales spike, do you normally see a sales spike at that time? If so, perhaps spend your money more wisely. Give it to me - just like a cornetto.

If you'd like more on this, please do Comment. 






Read this? You can claim a prize. First you have to find me.

...

What the frell is CLV !?!?

If you don't know what a customer is worth, you're inefficient. You should be thinking like this:









Average Order Profit
         x
Number of orders per year
         x
Average customer length (years)

         =

Customer Lifetime Value (CLV)


CLVs are important if you are selling anything.

...
CLV is not (here) a reference to how fast CDs work.

The mysterious case of the sick visitor, or why bounce rates make me vomit.

What's a bad bounce rate? If 90% of your shop visitors straight out, is that OK? Choose a number.

To
Firstly, what's your typical bounce rate? This well hidden bit of text tells you (~30%):

So if you have content that has a bounce rate of 100%, against your site's average 30% - panic stations?

But hold on, that's only 1 visitor - maybe he had a serious case of the man flu?

Here's one fix: 'Weighted sort'. Sort by bounce rate, then go to 'Sort Type', then 'Weighted'. Voila!

Here the important sources are direct, the .com site, and then the .co.uk site. Are these different enough from the 30% site average to invest your time in?