iterable

… but in this world nothing can be said to be certain, except death and taxes decisions.

—Benjamin Franklin, in a letter to Jean-Baptiste Leroy, 1789. (Well, edited by me.)

Should you hire Bob to lead your team that's developing some key company features?

Should you take the higher paying job in the Bay Area, or stick around in the midwest where your friends are?

Should you host your infrastructure for your web application on EC2? Ruby on Rails or Scala with Play? Hadoop or Spark?

Our daily lives are chock full of decisions -- everything from the minutiae of deciding what to eat for lunch to the important decisions that ultimately determine the fate your venture. Whether you're involved in a business, art, or the scientific community, or anything for that matter -- decisions are the unavoidable human certainties that inundate us day in and day out.   

The consequences of the decisions that we make today are felt like ripples into the future, and it’s becoming more clear that  errors in decisions can be extremely costly. For example, Tony Hsieh of Zappos once estimated that bad hires cost the company over $100 million! Imagine if that could be avoided; I’m sure Tony would have leapt at the opportunity.

What can we do to avoid bad decisions and make good ones? Is it possible to make better decisions? The decision making literature over the past few decades is full of indictments of our intuitive judgements. The biases built into our brain cleverly point out the frailty of our evolved hardware.

Luckily, frailty and potential are two sides of the same coin. We happen to live in an age where the torch of science has illuminated at least part of the way, and it’s extremely exciting.

So, without further ado, here are three science backed and battle tested techniques to make better decisions and reduce your errors.

1. Consider the opposite

Let’s say I spin a giant Wheel of Fortune, and it comes up with the number 65. I then ask you: what is the percentage of African countries that are members of the United Nations? What would you say?

Interestingly, scientists have found that when presented with this scenario, the median estimate for people who saw “65” from the wheel was 45%. The median estimate of the percentage of African countries in the UN for subjects who saw “10” on the wheel was 25%. How can these be so different?

The theory here is that when we make estimates, we start off with an anchor and then adjust up or down from that. When people are primed with a random anchor, their estimates are totally different! Now, imagine the implications for things we estimate in daily life, such as negotiating a job offer salary or haggling with a car salesman.

Scientists have found that we can do better on this type of task by simply considering the opposite. In an experiment similar to the wheel of the fortune test, scientists found that just pausing to consider something opposing your first reaction consistently reduces this anchoring bias. What’s more is that this effect is additive -- the more you do it, the better you are at avoiding this bias (and many others! there have been numerous follow up studies testing a wide variety of biases that all succumb to this simple tactic).

This mental habit is very simple and extremely powerful: when faced with a decision, simply considering the opposite of the first thing that pops into your head has been shown to reduce a wide class of errors.

2. Take the view of an outsider

When’s the last time you promised your manager that something would take a week  -- and it actually took 2 weeks? Have you ever been stuck in the quicksand of a problem that it feels like you aren’t making any progress?

If you have, there’s a simple mental habit that can help. It’s called “taking the outside view.” Imagine you’re an alien looking in at what you’re trying to do, and derive your estimates from that. For example, instead of planning out the length of a project by looking at the individual steps, look at how long these types of projects have taken in the past.

In software development, this technique sometimes goes under the alias of "rubber duck debugging": if you’re stuck on a problem, take some time to explain it to a rubber duck. People have reported experiencing that “AHA” moment while doing this. This is also why it’s so valuable to bounce your ideas off of other people (or a rubber duck).

This simple technique is so general and applicable that you can use it with every decision you make. Writers such as Robert Greene even believe that this technique is part of the equation for creativity, and that every great innovator brings an outsider’s mind to a problem.  

3. Change your context

Research suggests that our first response when faced with a decision is very gut based, and that simply changing your context can help make better decisions. This can be something as simple as standing up when you’ve been sitting for hours, or making decisions when the consequences are still far off. Even going for a walk every hour or so can be extremely helpful.

Anecdotally, when I was in school, I used to do the final edits to my humanities papers in the dusty stacks of the campus library, surrounded by the pages of knowledge written by my forebears. This environmental context was a welcome alternative to finishing my papers under the fluorescent lights of the computer lab, where I used to bang out first drafts.

So there you have it. Simply taking a moment to consider the opposite of what pops into your head, or taking the perspective of an outsider to the problem, or just changing contexts can help you make better decisions.

Go forth, ye mighty, and conquer!

… and if you’re interested in delving into the science behind these habits, check the sources below.

-- Girish Sastry, Software Engineer, Iterable

How we sped up Elasticsearch queries by 100x - Part 1: Reindexing

A Little Performance Story

At Iterable, we use Elasticsearch to store all our email campaign events. This includes email sends, opens and clicks.

For each event, we attach the recipient email and display the events nicely in our campaign dashboard

Notice Unique Opens column above. To calculate that, we need to fetch recipient emails from all the open events and do a unique operation on the emails. As you can imagine, this query is quite inefficient. It took upwards of 5 seconds to fetch 500,000 events, despite gigabit connections between our app server and Elasticsearch.

[2013-08-26 20:57:33,856][WARN ][index.search.slowlog.fetch] [][][0] took[4.2s], took_millis[4207], types[emailOpen], stats[], search_type[QUERY_THEN_FETCH], total_shards[5], source[{"size":500000,"facets":{"dateFacet":{"date_histogram":{"field":"createdAt","interval":"minute"}}}}], extra_source[], [2013-08-26 20:57:34,293][WARN ][index.search.slowlog.fetch] [][][4] took[4.6s], took_millis[4659], types[emailOpen], stats[], search_type[QUERY_THEN_FETCH], total_shards[5], source[{"size":500000,"facets":{"dateFacet":{"date_histogram":{"field":"createdAt","interval":"minute"}}}}], extra_source[]

This sluggishness slowed up in our Elasticsearch’s slow log, excerpted above.

To solve this slowness, we had to off load the uniquify queries to the Elasticsearch server and simply return the counts, though this came with a few challenges.

The Power of the Scroll

How tokenization affects your searches

By default, Elasticsearch applies the standard analyzer to fields. For the email address [email protected], the standard analyzer creates three tokens: justin, 9000 and gmail.com. If we’re to search the field with a terms facet to get unique email counts, the result will include an entry for each token, which inflates the unique count.

The Elasticsearch inquisitor plugin has a handy tool that shows tokenizations with different analyzers.

Taming the scroll with Scala

We must set the mapping of the email field to not_analyzed for Elasticsearch to return the email as one token. However, changing a mapping requires reindexing of the data for the new mapping to take effect.

As of Elasticsearch 0.90.3, there is no built in reindexing support. Instead, we rely on the scroll query which essentially maintains the query results in memory, for performance and consistency reasons. There will be some data loss in the next index which can be fixed with another scroll query.

Here’s the code to do it in our favorite language, Scala

def reindexType(srcIndex: String, newIndex: String, searchType: String, client: TransportClient) = { val scrollTime = new TimeValue(600000) val scrollSize = 10000 // Setup the query val query = client.prepareSearch(srcIndex) .setTypes(searchType) .setScroll(scrollTime) .setQuery(QueryBuilders.matchAllQuery()) .setSize(scrollSize) var scrollResp = query.execute().actionGet() // Keep processing results if hits are non empty while(scrollResp.getHits.hits().length > 0) { // Prepare a bulk reindex request val bulkReq = client.prepareBulk() scrollResp.getHits.hits().foreach { hit => bulkReq.add(client.prepareIndex(newIndex, searchType).setSource(hit.source())) } // Reindex the hits bulkReq.execute().actionGet() // Get more entries from the old index scrollResp = client.prepareSearchScroll(scrollResp.getScrollId) .setScroll(scrollTime).execute().actionGet() } }

The code above fetches the _source field from each entry and reindexes to newIndex. Be sure to set up the new mapping properly on the new index first, otherwise all that effort was for naught.

Once the data is in the new index, the last step is to delete the old index and create an alias to the new one

curl -XPOST 'http://localhost:9200/_aliases' -d '{ "actions" : [{ "add" : { "index" : "the_old_index", "alias" : "a_new_index" } }] }'

You can create the alias through the ElasticSearch Head plugin which also nicely displays the aliases on an index.

Coming up next…

I’ll go into how we off loaded the unique queries to the Elasticsearch server, now that the email field is properly treated as one token in the new index.

As customers receive more and more email notifications these days, it's becoming harder to rise up above the noise and pique the customer's attention. As a marketer, how do you get your email to be noticed in a sea of emails, and better yet, be opened?

One strategy is to optimize the first couple of things that a user sees in your message. I'm not talking about the message body itself, but rather things even before that. Basic things like the from name, the from email address, subject line, and preheader text.

What is preheader text?

You’re probably thinking, “I know what the email address, email name, and subject line are, but what’s preheader text?” Preheader text is the first few characters of text from an email that a user sees in their email client, immediately following the subject line. The amount of preheader text shown varies from client to client, but it typically ranges from 80-120 characters, or almost the size of a tweet.

The preheader text is often the difference between a conversion and an archived email, so it’s important to be as clear and concise as possible.

Room for optimization

You'd be amazed at how many companies actually completely ignore this extremely valuable piece of real estate. Extensive testing is often done on things like the from name and subject lines (we used to do this in my days at Google), but hardly anyone thinks to test something like the preheader text. Treating the preheader text like an extension of the subject line (which it is) is the way to go. This is only going to become more apparent, as mobile email client adoption skyrockets over the next couple of years.

Emails that should be improved

Even the biggest companies are guilty of this. Here are just a few examples of emails that could greatly improve their open rates if they customized their preheader text with something more compelling. The worst kinds of preheader text are the ones that invite the user to view the email in their browser (almost no one does this), or ones that tell the user to follow them on Facebook, Twitter, Instagram, Pinterest, etc. The best ones tend to include a call-to-action, promote a time-sensitive sale or event, or simply serve as an extension of the subject line.

"To view this email as a web page," or "Email not displaying properly?" are not compelling enough to get me to open the message.

Implementing your custom preheader text

<head>

<style>

/*--- Preheader declaration in style block in addition to inline for Outlook */

.preheader { display:none !important; visibility:hidden; opacity:0; color:transparent; height:0; width:0; }

</style>

</head>

<body>

<!-- PRE-HEADER TEXT -->

<span class="preheader" style="display: none !important; visibility: hidden; opacity: 0; color: transparent; height: 0; width: 0;">This is the preheader text! It shows up in many email clients like Gmail, iOS mail, Mailbox & more.</span>

Sample code courtesy of Github user kbav

Writing good preheader text is an art, but it’s also something that you can experiment with. Try running different variations of preheader texts to see which one gives you the most opens and conversions.

Here at Iterable, we make it super simple to customize and experiment with all aspects of your emails like the preheader text (and much more), so you don't need to worry about implementing the HTML code yourself. We give you all the tools you need to capture a customer’s attention, increase your sales, and get more users.

Back in May, the Gmail team released a new & exciting feature- inbox action buttons. These buttons allow the user to take a specific action, without even having to open the actual email.

So, for instance, a user can RVSP to an event, check-in to an upcoming flight, track a package, review a restaurant, and more.

Think about how awesome that is. The barriers to getting your users to convert are now lower than ever before.

Not many companies have adopted the new feature yet (here's a preliminary list)- but expect more to soon. Along the same lines, more mail clients than just Gmail will support this feature in the coming months, as it's based on schema.org's open standard.

A few examples in the wild

Spotify uses Gmail actions to allow the user to immediately start listening to Spotify songs and playlists. It links to a page which then launches the Spotify desktop app, if installed.

MailChimp is another company that takes advantage of these actions. They make it easy for people to confirm their subscriptions to new mailing lists.

Even Amazon is getting on board:

If you're a company that sends out emails without adding one of these buttons- you should really think twice, as it can only benefit your campaigns. Users no longer need to open your email, scan the content, and hunt for the call to action- Gmail surfaces it right up to the inbox.