May 28, 2014
Every Minute Counts: Coordinating Heroku’s Incident Response
The hardest thing about ops and incident response isn't designing robust systems, debugging production, or quickly repairing technical issue...
Hi, can you hear me well? Ok, I was just telling everyone how much I hate the first two or three minutes of each talk and it's much easier for me to speak about marketing than to speak about the company and myself, so I'll go over these details quickly.
I'm co-founder and VP of Product and Marketing at Takipi. Takipi is actually my second company. I studied architecture. Just before I graduated, I decided that it's much more interesting to build companies than to build actual buildings.
Then I co-founded my first company, Visual Tao, backed by Sequioa and was acquired by Autodesk about five years ago. We developed an AutoCAD app for mobile devices, helping engineers and architects edit drawings online. I feel the greatest achievement of my first company was getting 10 million users within two and a half years, with hardly any marketing resources.
We did lots of guerrilla marketing over there. The connection between Visual Tao, the first company, and Takipi, which is a very technical company, was actually because of the rapid scale and moving between zero users to one million users within a few months.
Also, all the technical problems that we faced, debugging at scale and moving from one server to 50 servers. We always saw this, the most challenging part with the first company was to bring really complex drawings to the cloud and then we found out that supporting hundreds of thousands of concurrent users is much more challenging. Then we realized that the next company is going to face this challenge.
What we are doing at Takipi is helping developers find really complex bugs in production, and solving them in minutes.
Let's move to the marketing part.
Today, I'm going to show our journey during the last year and how we brought our blog from zero readers to 70,000 unique visitors, in months. The thing that was very interesting about our story is that, when we started doing content marketing, and when we picked content marketing as one of the main channels, we hardly had any experience with content marketing and hardly any experience with blogging.
We took the more scientific way to do it, or maybe the developer's way to solve this problem. I'm going to show you how we handled it in a few minutes. The thing is, most of the motifs can be easily applied by other companies. You don't need to be very talented in your writing or have any blogging experience in order to get to the large numbers.
So, this is part of my team and the reason that we're holding monsters is because we're working on a very complex product. We decided to go with a lighter brand and to illustrate several exceptions and sometimes we take it too far, these are the Takipi hoodies and each developer has their own unique hoodie; and some of our interior design.
So the first question that we asked ourselves was, "Which content should we write?" We knew we had to be laser focused with the content because we had limited resources and we knew that each post should be a hit.
Then we asked ourselves, "Let's see what's working for other companies." So what we've done, like good developers, we wrote a script to do it and we actually developed a script that crawled over different blogs and by using Facebook, Twitter, LinkedIn, and Google+ API, we managed to sort all the posts according to the number of shares.
We were able to scan tech blogs like TechCrunch, Ars Technica, or Wired and the blogs of other companies to see which content is more popular and which content is less popular. That was one of the most fascinating weeks I had at Takipi just taking 70 Excel files and trying to look for patterns and finding very interesting patterns of content that performs better than other types of content. If you want to use this script, just send me an email, I'll be happy to share with you.
Another tool that you can do in order to analyze a blog is QuickSprout. It's a great tool but it has different limitations. Other than timeframe, it doesn't work on sub-domains. So if you're looking for guitar.com/blog, it won't work, but if you're looking for an entire blog which has its own domain, it will work.
The first thing that we found out was through comparing between different products, different tools, different technologies.
This is a number one post on Atlassian's blog which is Git vs Mercurial: Why Git? Their second most popular post is Mercurial vs Git: Why Mercurial? It was very interesting to see posts that compare different products or different technologies always perform very well. We were amazed to see that there wasn't any good resource on Github vs Bitbucket so that's something that we wrote. It's pretty easy to create this content.
What I really like about comparing different tools or different technologies is that you get people in the right state of mind. They're already looking for new tools and they're very open to try them.
Usually, we write about different products and then we add a paragraph about Takipi and how it integrates with these tools. This is a list from Google Analytics from last month of organic traffic and you can see that three out of the ten best performing guest post are a versus. So we have Atom vs. Sublime, Github vs. Bitbucket, and Lambda Expressions in Java 8 vs. Scala.
This is a great tip for vs., but it's also a great tip just to get more ideas for content marketing, I love using Google Autocomplete.
Every time I choose one product, I just type it versus and then I see what is the first suggestion by Google. That's usually what your users want to look for and that's where you will get the most traffic coming from SEO.
I also like using it to get general ideas. So if I'm writing something around Java 8, for example, I'll just do Java 8 and see if Google has any interesting ideas for blog posts.
Research backed by numbers works very well. This is the top performing post for the New Relic blog and you can see that they're measuring different browsers. They had to contain around it and had huge news about which browser performs best, but you can see the difference between the different browsers is really minimal, it's not very interesting. Once you have numbers and once you can say that this tool performs 3x better than other tools, it's always interesting.
This is a post that took us less than one day to produce. We asked one of our developers to take five different frameworks of logs and to measure how long it takes them to write one gigabyte and then we published the results.
It was very interesting. We got over 20,000 unique visitors to this post. It has a very hard correlation to Takipi so it worked very well.
Lists work very well, like five tips, seven ways, and are really great content for your blog.
So that's something that we like to do every month or every other month. We produce a list of tools that has something to do with Takipi. These are the seven new tools for Java developers. We have tools that you should use after deploying a new version.
What I like about this kind of post is that they convert very well because you can write about your product without really promoting it and you can use other companies as well. It's a great way to build a relationship with other companies and they usually share the post and tweet about it, and you get their users to your blog as well.
Great posts come with great headlines and I will spend the same amount of time picking the headlines and writing the post. They say that's what people see on Twitter and that's how they decide whether they're going to read the post or not.
The evolution of our script was to see which headlines are more popular and we actually ran another script where we counted the unique words for popular headlines versus unpopular headlines.
There's the lengths of different headlines and we got some very interesting takeaways that usually work very well.
Once we found that out, it was very easy to see before we even counted, is that negative words perform very well. Kill, dead, nightmare, or bloody will usually make a very good headline.
For example, "Is big data dead?" or "Is x dead?" was always the top performing post for all the large blogs. A lose over win works much better, so if you say, "Apple is losing to Android", it will work much better than, "Android wins over Apple."
This is an example I really like. I think this is one of the five best performing posts for TechCrunch Enterprise and what I like about it is that this topic is really boring and nobody really wants to hear about databases or Oracle, but the headline is amazing, "Oracle is Bleeding at the Hands of their Database Rivals." Another great post and great title from around the same topic, "Oracle makes more moves to kill open source MySQL."
This is how we used it. This is a post about the disadvantages of using the Lambda Expressions in Java 8. No, not a very dark topic or very thrilling one, but I think that the dark side may be used differently and causes a discussion around this and how Java 8 is affecting us and the disadvantages. There were something very interesting about the dark side.
It's pretty well known that using numbers works very well, so using lists like, "Five tips to". What we found is that using large numbers usually works much better than small numbers. So, "Seventy- eight ways to" will perform better than "Seven ways to."
This is an example of the most popular post for Ars Technica around security. What's interesting about it is it's, "Extremely Critical Ruby on Rails Bugs Threatens More than 200,000 Sites," which is over a number of Ruby sites, but it's very dramatic so they just took this huge number.
This is a post that actually started with an anatomy of a great Stack Overflow question. It didn't perform very well and so we added to it after analyzing 10,000 questions. Then it performed much better after adding the large numbers.
Here are some examples from Neil Patel's blog. What he's doing is using very accurate large numbers. It usually works very well because it's more reliable. 225,418 visitors, you can also see he's using losing. I think this was one of his most popular posts during the last couple of months.
Piggybacking cool brands is a very interesting technique.
Everybody wants to read about Twitter, Facebook, or Google. So you can usually say how Facebook was damaged and then just to talk about a certain virus or a bug that has nothing to do with Facebook in particular.
This is an example about how Twitter stores 250 million tweets which is not really about Twitter. This is my most popular post for my personal blog. It's a post about cold emails. I have over 100,000 unique visitors to this post and I'm pretty sure that's what made a huge difference, just adding Twitter, LinkedIn, and GitHub to the title because there are thousands of great posts about how to write cold emails.
We actually started by accident because we didn't have people that were skilled enough to write or had time to write. When we just started the blog, we were about eight people. We decided that the developers should help with writing the content and it's probably easier for them to code the content.
This is an example of a really engineered post. This post was actually very well planned. We chose GitHub as the main topic because we know it's very popular so we piggybacked Github and we chose large numbers. Then for probably five or six hours, we download the most popular GitHub projects. We scanned the content and then we sorted all the popular libraries. Within one day, we were able to create this amazing post with over 50,000 unique visitors to it today and by actually using our developers.
This is a very different example which works very well. It actually came from a moment of deep despair. I remember I tried to get some PR to Takipi and I was unsuccessful doing so. I went to Hacker News and I saw that the first place on Hacker News had about 400 points, but it was a very basic project.
Then I went to Reddit and saw it there and on other blogs as well. I thought to myself, "Why is this project that probably took them two days to develop getting all this traffic and we have this amazing and unique technology and no one wants to write about it?"
Then I decided, "I'll make another, I'll make our own project that will take us two days to write and will be much easier to communicate and market." What we did was, we actually took a small part of Takipi's notes run for code features, which is part of the UI where we take log files and present them in a nicer way.
We launched it as a Stackifier under a different domain with a link to Takipi and we targeted it on the visitors and got 10,000 visitors within the first day.
I really like taking small projects and launching them as different websites. It drives traffic to your website.
People ask me about this project, "Don't you need to market them as well?" "Don't you run into the same problem, because you need to drive traffic to this product as well?" There are a lot of websites, and their entire goal is just to introduce new projects or new startups, so that's how you can drive the initial traffic to these kinds of mini projects or mini websites.
Product Hunt is the most well-known one but there is a great list on GitHub called Places to Post your Startup with over 80 different websites that you can just put your small website over there and to get great traffic.
Creating a distribution channel is super important for content marketing, even if you have some post that you do very well and you get lots of votes on Hacker News or Reddit, it's still very important to create a very solid channel.
I want to share different channels that we use today to distribute posts. The first question that we asked ourselves was of course, "Where do other blogs get their traffic?"
We have lots of great tools to understand how it works and you can check those out at my personal blog. I have a post called Reverse Engineering Marketing with all the tools that you can use in order to understand where other sites get their traffic.
Here's a quick demonstration of SimilarWeb. This is the free version. You can see all the results for Takipi and which websites we republish the content on that actually give us traffic as well as which kind of social media works well for us.
We decided to almost neglect LinkedIn and Facebook, based on research that we've seen from other companies that tell to developers.
I want to share a quick tip about Reddit and how to use it better. You can enter reddit.com/domain/ and enter a domain name. Then you can see all the places that this domain was submitted. This is a really great way to discover new sub-reddits. Some of them might be very small and you don't know them yet but they may be active with lot of users.
Over here I search for kissmetrics and found several 40 new sub-domains in Reddit that they didn't know and used them to publish some of my posts, which is a great way to understand which sub-domains are more active.
With my previous start-up, I used to guest post a lot. I have to say that over here, I changed my perspective on guest posting. I think that if you get a guest post where you can actually write about the product, it's probably worse, but if it's a general guest post where you get this tiny link of Guest Post by Iris Shoor from Takipi, it usually doesn't work, either as a link or as traffic.
With Takipi, we have guest posts at VentureBeat and other large blogs and we only had 200 or 300 visitors to the website, which isn't worth the effort. What we do a lot is republishing. So every time we have a new post, we have about three to five websites that republish our content and this works very well.
If you ask me if you would get penalized for doing so, I'd say only if you republish someone else's content. But if someone else republishes your content, then it's okay by Google.
Newsletters for developers is a great tool. Lots of newsletters with 1000 to 10,000 of subscribers who want different technologies, for us mainly Java and Scala. I know that for Front End there are dozens of interesting newsletters.
What we do today is we have a list of all the newsletters that might be a good fit for our post. Every time we have a new post, we try to think if it can match a certain newsletter and send very short emails, "Maybe you'll be interested in featuring this post," and sometimes they do. Sometimes they don't but when they do, it's really great traffic.
What I like about this traffic is that, unlike Reddit or other websites where people are entering just to read the content, the thing that newsletters do is they're more willing to look for new tools and they have usually more time to look into the product itself.
Wikipedia is an amazing source for traffic and I think of it more like it's Google because it's everywhere. If you're getting a link to a certain article and you get dozens of entries a day, you will have the same number everyday, so it's not like republishing a post where you get a peak and then you lose traffic. What's very important to remember about Wikipedia is, don't spam. The moderators over there will probably remove your links within a day or days. But it's more about finding articles that are right for your posts.
stats.grok.se is kind of the Google Analytics for Wikipedia, so you can see how much traffic each article gets. Before we try to submit a post to different articles, we actually check. Some articles that seem to be a very interesting value only have dozens of entries a day and it isn't worth the effort.
But this is an example of where log4j, where we edit the link for a post on comparing log4j to other posts, which drives very good traffic for us. One thing that I realized doing that and when adding links on articles is adding other links as well so it won't look too commercial. I usually pick another two or three really good articles about the topic and then add all of them. Then it's easier to pass the moderators.
Q&A Websites are mostly Quora and Stack Overflow. I like using their alerts and system of Stack Overflow. When we have a new post, let's say, for example, log4j, we want to look for all the questions that usually don't generate lots of traffic or if we look too commercial, just to push it over there, and we set notifications for questions about log4j.
Then when there are new questions, we get email notifications and we answer if the post fits the question. Again, no spamming, it won't work.
Google+, I know what people think of Google+. I love it as a channel to get traffic. You can see we get around 1,000 visitors a month to our website coming from Google+. What I love about it is two things.
The first one is that the really early adopters are already there so all the answers are really great. I would say that people are willing to use Google+ now and are willing to try new technologies as well.
The other thing that we really like about it is that unlike Twitter or Facebook, when you need to build your own community and you need to work in order to get followers or likes, with Google+ you work in communities. So I can enter the Java community and post one for post over is there and get traffic. You don't need to work for it and within two minutes, you can get hundreds or thousands of visitors to your website.
The last thing that I wanted to show is a different kind of content marketing, which is not about posts. This is a project which we launched two months ago. It's main goal was just to create better relationships with influencers and with blog editors.
So we launched javais.cool where we have a list of the influencers, of the blogs, of the newsletters around Java. Before launching this project, we emailed all the people here and asked if it was okay to include them and if they had any other suggestions.
Seventy five percent of them replied back. It was a great way to start engaging with all the influencers we didn't know before and for them to know about Takipi, and to ask them for favors or to retweet other posts that we write. We also had great traffic.
It was very easy to market it because we sent out all of these people and all of these blogs, and when we launched the website we asked them to tweet about it. This is the first tweet. We got 10,000 visitors to the website on the first week, and just from them tweeting about it.
If you want to use a template and replace it with a different language or technology, just let me know, I'll be happy to share it. Okay, thanks. Check out Startupmoon, my personal blog. I write about marketing and products.
I write some of the content and Tal writes some of the content. It's mostly the developers who write the content and we have someone who started with the support team who writes content. We try to outsource it, but it didn't work very well.
I think the main thing is the content itself is not about writing it very well. It's getting the benchmarks or comparing between the different tools and getting all the data right or writing about new tools. I think that it's more about to getting to interesting content than the quality of the writing.
I prefer quality over quantity. I know that for a really good post we can get maybe 20,000 readers and for not so good posts or for average posts, we get 1,000 to 2,000. We prefer to invest even, so two or three days writing a really amazing post and also to have a post out once every two weeks or so. I would prefer to have two or three posts a week but we don't have enough resources.
Today, we try to post once per week. Where we have two posts a month, those are really technical and good, and two post which are more like fillers and don't perform as well.
Once we had a post where we took all of the Amazon regions and compared what the fastest route is and where the slowest one is. Then we sent it to VentureBeat but it didn't convert very well so we got some traffic to the original post but we got very few signups. I feel that it wasn't worth the effort.
I think it's mostly for developers. It's very different with tools for marketing or any other type of tools. If you were to ask me two or three years ago, I would say we have to do it, this is the right way and what I did when I marketed the previous month.
But the thing is, with developers to lead, it didn't perform very well.
What works best for us is to have the Takipi header on the blog itself. There's a link to the "Features" page, "Getting Started", and so on. That's where we get most of the users that continue from the blog to the website.
On the end of the blog, we direct users to another blog, which is similar and then we direct them to sub-pages within the Takipi.com domain.
For example, you're reading about Lambda Expressions in Java 8, five cool features in Java 8, how to use log files better and then it's actually part of our Features page. So we wrap some of our features as content marketing and suddenly we get more traffic.
No, actually, we haven't. One thing that does work for us is adding the Feedly button. We get around 1,000 visitors and that's coming from Feedly. This number really increased after just integrating the Feedly button. I don't feel that people will tweet more if we have bigger buttons.
One thing that we have tried and didn't work was to have some interesting quotes or interesting lines within the text itself. For example, Amazon AWS in Brazil is ten times slower than in Virginia. We'd say, "Tweet this," but it didn't really work. We hardly got any tweets and we feel it's cutting the flow of the posts.