I got a good and fairly specific question by email a few days ago that provides a bit of a glimpse into the bot’s inner workings with some colorful charts.
Q: I was just wondering when changes to budget and number of items go into effect. I’d assume that the budget changes when the next billing period hits. However I’d assume that keyword changes go into effect immediately (or daily), and I’m not sure when a change to the number of packages would happen (hopefully not until the budget change happens too). Just curious as to when all of these things happen.
A: All these changes get complicated surprisingly quickly.
Keyword changes get implemented once or twice per week, but obviously there is a delay caused by shipping. Packages you get today will probably be based on last month’s keywords.
On your account you can set how many packages you want the bot to buy for each address. Here you are actually setting the bot’s “package preference” for the address, which is a number from 1 to 10. That number tells the bot how long it should wait before ordering another package. The bot uses the below colorful eye charts to figure this out. For example, if you have the slider set dead in the middle and your budget is $30/month the bot will generate a new order every 2.2 days with a target budget of $2.16 per package. If an item is bought for less than the target budget, the bot rolls over the balance to the next package. The slider on the site will read 9-18 packages per month and the bot will aim for 13.9 packages per month. Changing the number of items increases or decreases the package preference, which in turn changes the time between new orders and target budget.
In generating packages the bot only looks at budget changes that have gone into effect. For example, if you set your budget to $0 for all addresses the bot will stop charging you but will keep generating packages until the funds run out.
Increases in your budget go into effect immediately, but you are only ever charged once per month. A pro-rated amount is added to the next billing cycle after you made the change. So if you signed up for $30/month and increased your budget to $40/month halfway through your first month the bot would tack on a $5 pro-rated charge to your next renewal.
Decreases in your total budget go into effect at the next renewal date. The idea here is that someone adding addresses or budget likely doesn’t want to wait to get packages. Usually decreases are cancellations, and in those cases the bot needs to spend the rest of the funds.
A subscriber to the bot wrote in today…
I have changed a keyword from being positive to negative yet am still getting packages with that keyword. When I set [my account] up originally I selected mystery as one of the key words, then since a lot of the packages had this keyword I removed it. I was still getting packages with the mystery keyword so I then added it as a negative keyword. How quickly do keyword changes take effect? As there have been new packages with the mystery keyword since I made that change.
The keyword ‘mystery’ and ‘random’ are unique. These words essentially refer to anything that you haven’t specifically asked for, which includes everything bought by the bot. So these keywords will match every product the bot has.
New orders that are “pending” where the bot hasn’t yet confirmed the order are also marked as ‘mystery’. So everyone will always see a few ‘mystery’ packages at the top of their account. Eventually those ‘mystery’ packages reveal the keywords they are tagged with and get a better estimated arrival date. Over time the packages become less mysterious.
It is impossible for the bot to buy packages not tagged as ‘random’ or ‘mystery’ so adding them as negative keywords won’t change anything. The ‘mystery’ packages that you see on your account are placeholders and once confirmed will change their keywords. If you really don’t want any mystery, you could have only one, clearly defined, positive keywords on your account. Then you will be guaranteed no mystery. But that is kind of besides the point here. We want some mystery and surprise.
Keyword changes take effect once a week with a few exceptions where they get updated more quickly. The bot runs a program to determine which products match which users’ keywords. These are the stats from that program last week…
Products matching keywords: average is 4,346 products per address
Required positive keywords: average is 26 keywords per address
This afternoon I put together this bar chart of the top 25 keywords by their appearance on packages bought by the bot. Note that it is pretty flat and even the most popular keywords only match 5.5% of all the items the bot buys! We buy a lot of very different sorts of items. That said there are clearly some more popular categories like science, legos and cats.
Take a look at the below lists I put together of the most common keywords (positive and negative) that have been selected by folks using the bot from different countries. The data are quite noisy, but there are some interesting nuggets.
- The US and UK share about half of their top positive and negative keywords. No other countries have nearly as much overlap in the top keywords they select.
- Pretty much everyone likes Legos
- The keywords ‘crafts’ is a top positive in New Zealand and a top negative in Ireland
PS – These are the most common keywords selected through the bot. That is, these are the keywords that participants have told the bot that they want/don’t want. This is not a list of items by the frequency that they have actually been ordered (for example, we have not ordered any ‘nazis’ items for obvious reasons).
Top Positive Keywords (United States)
- Star Wars
Top Negative Keywords (United States)
Top Positive Keywords (Australia)
- brain teaser
Top Negative Keywords (Australia)
- car interior
Top Positive Keywords (Canada)
Top Negative Keywords (Canada)
Top Positive Keywords (Ireland)
Top Negative Keywords (Ireland)
- car interior
Top Positive Keywords (New Zealand)
- usb fun
Top Negative Keywords (New Zealand)
Top Positive Keywords (United Kingdom)
Top Negative Keywords (United Kingdom)