Welcome to the ATD 2017 conference Summy Pack by Summurai.

The ATD - All Things Data - conference is aimed for marketing and digital professionals who are interested in improving their results by doing a smart use of analytics and data. 

The conference was held in April in Tel-Aviv and included 2 tracks - strategic and technical and 18 fantastic lectures by local and international industry leaders.

Our team from Summurai visited the conference and created this "Summy Pack" to make sure you dont miss anything.

We summarized the lectures and narrated them into 15 summies, that will give you the main insights.

If you find this service valuable, please use the share button and share the love.

We invite you to go ahead and get smarter.
The Summurai team.

We tend to look at data too analytically. Israel is a leader of big data and innovation, However it seems that around big data we tend to “Guesstimate”. Generation Z has a lower attention span than any generation before it and seeks a high return on interaction. Rather than the return on investment, they are looking for experiences. Intelligent big data informs creativity, rather than dictates the process.To understand the demographic today, you should look from a human lance, and not only from the data dashboard lance. Our left brain controls the analytical and logical thought processes, and the right brain is responsible for our creative side. Our daily thought processing brings both sides together. Nowadays we see a lot of Consumer Disinhibition,  people tend to say what they think you want them to say. Data analysts should take that into consideration. In current era, the bigger the story is, the bigger the impact it can make.For the first time for the last 500 years, we are moving towards a changing point. In the next 40 years, big data, VR and AR will become smarter than any person. More and more of our workforce will be replaced by data-driven machines. We will have to connect to people more and more by the storytelling and creativity and less from the logical-machine point of view. According to Elon Musk, we are going to have a digital layer on the limbic system. Gabay suggests that now is the time to work together to understand the new consumer mind. Israel is the leading innovator in this field.

Following Cloud-Computing, we are now in an era of Cloud-Marketing – a big-data concept that is leading the tools. The market uses the term “Marketing automation”, however automation is just one part of the cloud and data-based process. The idea of automation is misleading.

Marketing automation systems are a top system that controls all advertising and marketing tools. It is connected to CRM and is used to manage an automatic flow based on the user customer journey. The user journey flow embeds the worlds of customer acquisition and customer marketing, using the following  elements:

• Digital advertisement



• Personas



• Inbound channels



• Known customer data (CRM)



• Email, mobile and social

 Now that we know this information, what can we do with it?Today we manage different forms of data in 2 main categories. These are Personal known data - such as address, marital status and buying habits and Anonymous cookie based data, such as devices usage, content the user is interested in and social impact.

 The idea of a marketing cloud is to connect the two data sources and use them to improve conversions. From a technical perspective, this is done with a pixel.  An html code that enables us to connect the data worlds into a single layer of data and analyze them together.The improved segmentation allows us to better understand the context, user story and needs. Automation deals with creating the right customer journey. The World’s biggest tech leaders started acquiring marketing tools to create cross platform marketing management suites. Israeli Marketing Cloud is creating a local marketing suite that would fit the local market based on its sizes and financial abilities. The platform will connect to the company’s local CRM, and data warehouse to offer a BI platform with a dashboard and reporting tools. the platform will be called Mapp.  You can learn more about it at mapp-israel.co.il.

"Via" is building an operating system for ride-sharing. It is a software for operating networks of on-demand shuttles. The shuttles are a dynamic route shuttle, and the route is based on the specific demand on the time. It’s had 13 million rides since it’s launch in 2013.The need to measure things quickly comes from the economics of ridesharing: Very rapid growth has  driven the need to update models quickly to fit the market. Other findings are that it’s difficult  to estimate life-time value, since the users are new. It is also a low margin business. The best approach is to work towards repeating business. This means that long-term effects should be measured. The approach the company chose is to use expected life-time value. This is built out from  three main components:

• Life expectancy



• Amount per unit of time



• Correlation between life expectancy and the stream of value.

The company uses machine learning to calculate data on  life expectancy. This process is slow and should be refined and iterated. Understanding the relation between life expectancy and stream of value is the most complicated element.  it is the most difficult and time consuming data point to quantify.

This lecture poses the question - How to get the information,  and what information should you get.James Courier, founder of NFX, worked for a gaming company. During his time there all the companies games were failing. When trying to understand what people are looking for, he identified that most of the names of the games are based on the structure of “X of Y”. For example – “World of Warcraft”, etc. They tested a list of X and Y variables and created ads for all the possible combination. They found that the most successful combination was “Dragons of Atlantis” . This went on to become a 150-Million-dollar game. Avigad calls this type of testing a Minimum viable test. This type of testing leads to Data-led decisions, Reduced risks and costs and Iterates much faster. Some relevant platforms for testing are A/B tests on Facebook, LinkedIn, AdWords and Email with the relevant landing pages. Additional platforms include long surveys or short questionnaires. At Jolt the use questionnaires in the form of product carousels with questions, when the user clicks on the item they are answering.

Jolt also uses Sprint-based Experiments. All the tests needed in the company are mapped into a table. When all of the questions are managed on the same place, tests can be done quickly and answer multiple needs in a single test. The tests have a two week lifecycle are done all at once, and kept to 1 variable. Jolt uses Fibonacci values to give a score, the score is based on the value divided by difficulty.

Avigad Recommends to use the following technologies for test. Similar-Web chrome extension is great for collecting  data about each website. Once you have the data, create an excel file with a list of competitors. The excel file should contain the statistical data and a one-liner about the competitor.

Facebook ads are the easiest way to get to the market and the content should look good on mobile.

AdEspresso allows you to mix and match ads quickly based on all of the elements.

Linked’in is not effective for sending surveys and questionnaires, however you can use it to send messages to a target audience. LinkedIn are also the best at segmentation.

Adwords should be used for the content on landing page, and the  content the user is led to should fit the content on the ad. 

Unomy and Outreach Emails can be used for email marketing and retargeting. These you allow to send an email that looks personal with no “unsubscribe” section, unlike MailChimp for example. Creating quick landing pages – Landing pages are usually not successful. We need the ability to move fast and try a lot. Unbounce/Instapage can help create quick landing pages.
The landing pages should be less creative and test basic ideas quickly. You should only spend time improving the landing pages after they show success.

How much data is needed to be meaningful for each platform? Avigad outlines the following metrics.

• Getting a 1% conversion is a good sign



• 1,000 impressions



• 250 sends for each email version for paid emails



• 50 full surveys



• 400 questionnaires answered



• 7 customer interviews

Sagit Siegal of Thomson Reuters shares her experience from leading the UX re-design process of ‘Open Calais’ – a Reuters platform that offers advanced data analysis. The process started with a user research process, aimed to identify the main user needs. Different users were found to have different data needs. The users were then segmented into those who need basic data and those who need in-depth data. The next stage was prioritizing user needs, the  features were ranked based on these user needs. Additionally the users were ranked from a business point of view. The process starts by finding a common ground – starting with the basic needs that everybody has. After asking multiple users what features they need, the top 3 selections were mapped. These elements were then placed at the top of the page as leading data elements.

The next step was to enhance the interface with more data. A large amount of data on the screen was displayed using a scrollable model.  This way the users are not overwhelmed with too much information. Next was expanding to more features and value islands. Some content pieces became clickable and led to detailed information dashboards. The final step of the process was reaching delight. This was reached by using a high-end design for simplicity. 

The design process was based on research. Different designs were sent to users and they were asked which ones they like more. White space was added to make the content breath.

Siegal outlines 6 suggestions for data-driven products:

• Using user research of some kind



• Prioritize user needs, business and marketing needs to deliver a work map



• Design based on research – test and ask while you design



• Expose data gradually – remove noise, allow users to get the main data first, and reveal more progressively



• Offer a larger universe of information



• Clean data presentation – when design is good, users are ready to accept errors.

Colon cancer is among the most common and most deadly forms of cancer in Israel. The only ways to detect it is by colonoscopy and a stool blood test. Due to the unpleasantness associated with these tests, many people avoid being tested for them. Varda Shalev is a general practitioner who’s patients have died from colon cancer. In her practice she examined data to see if there were early detection signs and noticed that the blood hemoglobin levels were gradually reduced over time, keeping in the normal limits. She believes that by using big data algorithms, the data can be used to learn about patterns to identify early warning signs. By identifying patients in the relevant risk group for that, the patient can be referred to the Maccabi clinic’s test center. This means that the detection process can be initiated proactively, rather than waiting for the patient to initiate the process. The medical services of Maccabi have created a big data analysis for a huge amount of information. Medical treatment evolves from being based on doctor-intuition to a data-driven as doctors can’t stay up to date with all the research and knowledge. Within the next decade, doctors are expected to use more data-oriented services and a data-driven decision making. Morris Kahn’s big data center iis one of the leading data-driven research facilities. Some of the data is structured and other is unstructured (images, photos). The center attempts to turn images and other unstructured data into data that can be embedded in the system.One of the projects dealing with taking all slides that were taken within a process, and create a digital-pathology with no human involvement. In a cooperation with Prof. Kira Radinsky – a machine learning process tries to personalize medicine by identifying that best medicine that fits one’s blood structure.

Gong.io is a conversation intelligence platform for sales. It offers to record, transcribe and analyze sales team calls and demos. The traditional Sales-Development (SDR) process that leads to sales starts with using content, social media or PR to target accounts. The sales team then tries to email or phone the audience and schedule meetings for a sales person. The employees who are good at this and do it successfully are very expensive. The automated process uses SEO, paid ads and automated emails to do this instead. The relevant audience is brought to a website and the click-rate-optimization model moves people to schedule a demo. The user can then schedule a on his own without the middleman.The process aims to identify a self-qualification process, to make sure that the leads that get to the meetings fit the product and have high conversion chances. 

Let’s dive into the development process of the product.  Following a professional summit, the team sent an automated email. The email included valuable content, an eBook. The email included 2 types of links that are referred the user to the same content. one link was meant for the people who browse the email quickly and read only the beginning and was positioned above the fold on mobile. Users who typically read the full content do so on desktop, the call-to-action was repeated, with another link. The email indicates that getting the book doesn’t require any form registration. This removes friction and increases chances that the user will click. Statistics show that 2/3 of the users click the first link, and that 1/3 click the second link.

Udi also learned that P.S. lines are always read. this is another opportunity to  add another call-to-action to request a live demo of the product.

The email was timed for the 1st business day after the conference. It reminded the reader where they met and offered real value. This resulted in an open rate of  71.36% and aClick through rate of 14.09%.

After the user clicks the link, they arrive at a web page where the self-qualification process happens. The website checks IP lookup to identify the lead source. The form then asks what is the size of the sales team and what sales demo platforms they are using. Based on the size of the team, the relevant sales person is allocated.Different thank you pages are shown to unqualified leads and qualified leads. Qualified leads are directed to pick a time for the meeting using a tool called “Calendly” which is synchronized with the sales representative’s calendar.

We are all familiar with Google analytics advantages. Thes are mainly border patrol.  This includes data such as who is visiting our website, how long did they stay and most importantly, who bought what? This lecture will discuss a few other aspects beyond these metrics. Specifically we will focus on Customer grouping of everything, going from Revenue to profit and Organic keywords.

To overcome common Google Analytics limits one should use Googles BigQuery and DataStudio. Google Big Query is based on Google service, it is very easy and very cheap, but we need to rely on other tools, specifically SQL, Javascript and Python.

SQL is the number 1 tool to use, Javascript helps us to work online and Python helps us to automate the process and to analyze the data.

The logic behind the customer grouping method is to try and match the users to be in same groups. The user centric reports that we get from Google analytics for example, won’t be able to give us this data so we need to add an ID to the user with help of Google manager tools.

The basic set-up would be exporting your data into Google BigQuery and writing a few simple SQL queries to read the exports in Google DataStudio. 

With Keywords, there is a big problem because if there are 2 pages in the same website with the same keyword, it will cause a cannibalization to the search results, therefore, you should check which page gets the better results and change the keyword on the other page to be a different one so that it won’t take traffic away from the first page.

Another issue with Google analytics the Time stamp. The problem with time stamp is that you can know what caused the users to convert (for example a TV campaign that was running during those hours) but you won’t know who those users are and which user came from where. Tobias suggests solving this issue by yourself with the data each one has on their website.

Connecting your data into Google analytics.

Google analytics is rich with data but companies often have more information about their users than can be reported by Google analytics. Sometimes it is information that is collected outside the website, by the sales agent, at the physical stores,  by customer support or any other interaction channels with the business.

However, Google analytics has the advantage of being connected with other Google tools

that can enrich your Google analytics, therefore, you should identify three potentially useful sources of data you may already have such as CRM systems, Marketing automation tools, Online ad platform, Social media management tools, E-commerce and Content management system.

It’s important to learn techniques for building connections among existing data sets to understand your customers, and then see concrete examples from your data.

There are two ways of analyzing the data: the first is by bringing data in via Customer dimensions and metrics, Data import, and Measurement protocol

The second,  taking data out with, Google integrations for audience, APIs and BigQuery.

Custom Dimensions allow us to add additional fields on our users or on the event that the user triggers. This can be set on 4 different levels:

1. User Level - This applies to the user, including their current and future session.



2. Sessions - this applies to the current session only



3. Hits - this applies to the page view event only



4. Product - this applies to a specific product for ecommerce tracking

Let’s observe this case study- Teach for America used Custom Dimensions based segments to create audiences that were used to target ad campaigns. They had no problem getting leads and applicants to register at their site but they don’t know how to differentiate the best ones or how to target them later on. They wanted to get more information about their users, so they looked at additional criteria the users have in order to be able to segment their audience better, and later on, to target them for remarketing for payout or search. How did they do it? They can get the info from:

1. Activities the user did on the website such as filling forms or asking support for something specific. These users many times identify themselves.



2. Events that occurred offline and identified the users.



3. Imported data from other sources.This is one of the tools you can use when you want to get more data, first of all you need to give each user an ID number and then you can ask them some questions. Once this is done - you will always know when this user is back on the site.

Measurement protocol. Universal Analytics revolves around a Measurement protocol, you can measure activity in new, non-website environments. This includes activity such as a Salesforce trigger when a prospect is updated, a tracking pixel in an email, via a customer support rep during a phone call, or through a backend system when a purchase is made in a brick-and-mortar store.

Raw data is the only way to go! The deeper and more pure data one can get his hands on, the more valuable it can be for his company. Many times people are making their assumptions even before seeing the data outcomes or even the data itself, and think they know the answer. What happens in those cases is a brutal misrepresentation of the data and numbers in order to achieve the presumably “right” outcome they assumed in the beginning. 

Dan-ya divides a company's lifecycle to 4 stages of data evolutions:

The first, MVP - Asking the first question

The 2nd, Tactic driven- Asking several questions, by several different teams

the 3rd - Integrated analytics- Asking advanced questions,  end to end and across platform

The 4th - Data application- Using data as building blocks

In the MVP stage, is it reasonable and understandable that no one will analyze the data and it will only be collected to one place.In the second stage, tactic driven, there is usually much more data coming in and it is sometimes already being analyzed by Google analytics and as such  is a crucial point to remember and not to be diverted by the biased data. In the third stage, companies usually import data that they have collected off-line. Customized analytics is suitable mainly to companies that are at stage three or four. With Customized analytics the companies can reach the raw (and real) data the company has collected and they can have very strong and meaningful insights from that data.

Here are the Top 5 reasons to go custom:

1. Raw data - Meaningful insights and end-to-end analysis



2. Out of the silo and into the light



3. Agility and flexibility to react to changes and the ability to control all the information the company has.



4. Ownership of the data



5. Cost reduction, the costs to storage data are so minimal that even for a start-up in their first steps it will always be a better decision to store their own raw data.

We need to do better personalization! It seems that everybody is talking about personalization but no one has a real strategy!

There is no strategy for connecting all the tools together, all the data, all the content and all the teams, therefore the personalization we see is at a very shallow level.

Ed presents a 5 Part Framework for personalization and recommends you ask yourself the following questions

1. Who? We must ask ourselves who are we targeting, who are we NOT targeting, and who are we trying to reach with our message? - The way to adress this is by describing your ideal customer profile. When describing it you should be very specific, use adjectives and be quantifiable.
   
   



2. What? What message do you want send to each segment?
   You should convey to people’s identities and interests, write about something they care about.
   
   



3. Where? What tools and channels can you use to send personalized messages to your target segments? Your Goal here should be to be able to coordinate personal message across multiple channels
   
   



4. When? When do you send these messages? You need to figure out the actions a user can accomplish in their journey for example: When would a visitor become a lead? Your Goal should be to create a messaging plan for each segment for every significant event of the lifecycle
   
   



5. Why? What’s the purpose of your personalization? Your Goal should be to deliver the most relevant message at the right time and leverage the power of being immediately relevant to drive action.

Everyone knows that the bread and butter of any online marketer is Optimizing Optimisation. The question is how to change the WHAT, WHEN & HOW of the optimization process in a way that will help you achieve explosive growth.

 When talking about MVP or Minimum viable product, you should put your time and money only in things that have a real value for the company.

The MVP approach in marketing includes Treating marketing campaigns like products development and Eliminating any process/effort that doesn’t dramatically improve performance.When talking about Adwords optimization, you need to only check what makes money and what spends money! And you should handle ONLY those problems. 

Here is an example of creative optimization regarding Facebook ads-
When using Facebook ads, you need to ask if they targeted to the right people?

Moshe found that in his case they did, but still he didn’t get enough audience, So he checked what could be improved. He decided to improve his videos ads because he understood that video is the main feed type Facebook are promoting those days. After the first round of improvements to his video he received a lot of feedback, questions and comments on the video that helped him better understand his audience’s needs, worries, demands and interest. He then decided to go for a “second round” of improvements. The second round raised the conversions, but not as high as he had hoped.  In the third round of improvements to his video, he was able to improve the revenues by 600%!

From this case study we should learn that instead of trying to optimize by a many small factors, we should focus on one aspect and make it the best much like in the Pareto approach.

1. Identify potential game changers & relentlessly pursue them



2. Focus on nothing else. If everything is important- NOTHING really is.



3. Testing isn’t free, optimization isn’t free. If you do something meaningless- you aren’t doing something important.



4. Big results may require a massive effort. Do whatever it takes to create a big change.

If you had gotten the access to your competitor’s Google Analytics, what would have interested you the most? Aminadav claims that all the information you need is out there and is readily available online. You just need to know how to operate the right tool to work with.

 In order to make a good use of your time, a bot should do the data scrolling.

The steps are:

1. To collect the relevant data for you. The bot can go over thousands of pages and scatter the data you are looking for and add it into your system



2. Build the data in a constructive structure- the bot should look for similarities in data, differences in data, and patterns that he can point out.



3. Analyzing the data with Google tools or machine learning. 

In today’s world we need to remind ourselves that the hard work should be done by bots or algorithms in order to maximize our time and exposure. 

One of the most important things that many companies still don’t do enough of is user research. Every company needs to learn about their users. They need to better understand  their needs, their hobbies, which social medium they are using the most and what their fields of interests are. In order to do so, Aminadav suggests testing on an individual user by studying what this user has in common with other users. The tools he suggests to use are:

1. Data scraper, a browser add-on that allows us to get information several pages at the same time.



2. Tool contact that provides you with more information on your users.

Measurement protocol is an essential part of the already rich Google Analytics feature set allowing data collection from various non web or app customer touch points - or offline touch points. The ability to tie user behavior across the entire business ecosystem is pretty much a must these days yet there are some nuances each implementation needs to consider in order to preserve and correctly attribute user interactions. In other words mapping business activities to the Google Analytics user, session, or hit model.

Why or How does this fit into today’s business models?

Because users are everywhere, they interact everywhere and they also purchase anywhere possible, so we want to track all available touch points and attribute.

Zorin suggests that every company should collect additional data on their users and store it manually in Google analytics and My SQL in order to enrich it’s data and to synchronize between the two software they take data from.

The Measurement protocol cases will show:

• How to preserve geo, traffic and device information



• How to and what to send in a recurring transaction case



• How to and what to send in a delayed transaction case

One of the biggest advantages of this system is the ability to learn about a delayed transaction.

A delayed transaction is when an actual conversion happens outside the default session window which is from the last known hit +30 minutes.

This is usually applicable to fast risk assessments processes, 3rd party service dependencies. Cash on Delivery or any delay able payment method.

 A few tips to remember:

1. Always send a customized user agent, never the default server one as Google Analytics may consider you a Bot.



2. Enrich the ‘user data’ inside the payload to preserve attribution.



3. Measurement protocol payloads size, keep it minimal,  below 8KB.

Use data import as an option if you expect payloads to be larger.

1. Send with a non-interaction event (attribution and info stays intact both in standard and MCF reports.)



2. Mind the view filters- whatever you set up mimic in payload.

Each company should come up with their own way of customizing Google Analytics to serve it’s needs.

In this lecture Shuki will go over a few creative examples:

1. Custom dimensions- In Google Analytics we can all see the “regular” data, but in order to expand our knowledge, we should expand to the dimensions that are taken in consideration.In order to do so, we add more and more layers to the same user. This is how we collect more information about him.
   
   



2. Because there are many obligations in Analytics, there are many things we should try to understand further than what is  provided to us by the Analytics.  For example, a lead source.  Each company should check if they have any information about it and if not, they should save it in their local storage and later on they should send it to the Analytics if they recognize it is a “quality user”.
   
   



3. Custom metrics- with Google Analytics one can never know the conversion rate in an accurate manner. Shuki suggests a new way to calculate the conversion by adding more metrics on that specific user. This can help businesses that are B2B and have many forms they want their customers to fill. By choosing their own metrics, they will be able to know how many forms were opened and how many of the customers eventually filled them out.
   
   



4. Calculated metrics- a good example is when trying to understand which questions out of the FAQ questions that were published at the conference’s website were the most relevant ones.In order to calculate this , Shuki took the ratio between how many times a question was touched or “hovered”  over and how many times people opened and spend time on the question itself. 

The most important takeaway is to not be satisfied with the data that Google Analytics can provides and to try to come up with metrics and dimensions that best suit your users.