Looking Forward to 2018 and Beyond: Where the Data is Taking Us
Thanks so much to ObservePoint for having me back here. This is such a wonderful summit and I’m really excited to be a part. As mentioned, I’m going to be talking about looking forward to where the data is taking us.
To do that, I wanted to cover a few of the key market trends that we are seeing that are really influencing our roadmap.
First, we’re seeing an explosion of data and information. Individual touchpoints, channels and devices are all generating more and more rich customer data and signals that are critical to your business. Next, data is complex and varied. Finding the right insights and acting on them is increasingly difficult because there’s just too much data. Last, visualizing data across systems has become difficult. it’s a broad ecosystem and it’s fragmented and finding tools that are seamlessly integrated and easy to use together is increasingly rare. Let’s take a look at each of these trends one by one and look at the ways we’re tackling them, both today, and where we’re going in the future.
First, as I said, more data is being generated than ever before. As a great example of this, in Google Analytics, we process hundreds of billions of digital moments across devices, and we do that every single day. Huge amount of data is coming into Google Analytics alone, and that’s just one of your tools, one of your sources of data. We see that one in four internet users say that they start an activity on one device, but continue or finish it on another at least once per day. All of those different touch points may have impacts on your business goals and it’s important to understand how each of those touch points can impact on those goals.
Often skilled analyst or data scientists will use many different tools and a lot of their time to be involved in getting you’re the answers in terms of which sources are driving the most impact for your business.
That’s where attribution comes in.
Let’s take a moment here and think of your favorite sports team. Regardless of the sport, the person who scores a goal or touchdown, or basket, typically gets their name on the scoresheet as having scored. All of us have heard the phrase that there’s no “I” in “Team,” but how doesn’t that work in practice? A good example of this, I think, is the 2006 World Cup. There was a goal scored by Argentina. Leading up to that goal, seven team members were involved in touching the ball and passing it and keeping it going down the field before Esteban Cambiasso ultimately scored.
Would this goal have actually happened if it weren’t for the other team members? Were there one of the other team members who were perhaps more important than Esteban in actually scoring the goal and the overall successful outcome of that match? Or was this a team effort? Similarly, how do we attribute the appropriate credit in a digital world?
Think about your marketing activity and how many different channels you’re spending money on to drive success for your business. There’s paid search, display advertising, affiliate marketing, email activities, social media, and those are just some of the major paid media channels that most large advertisers are using. We know that in the majority of cases, it’s not just one channel that is involved in driving conversion or ecommerce transaction of your website, it’s several. How do you assign the appropriate touch points so that you have a good idea of what touchpoints your consumers engaged in before converting with you?
In the majority of tools today, we lack the civility to do this effective measurement across all of these channels. Most times it’s last-click attribution as the default model. Even though there are multiple touch points—in this case, six touch points—leading to conversion. In this case, with the slide you see on your screen. This conversion is attributed to this last touchpoint, which is my home computer, from a direct source. As you can see, I had many different interactions across three or four different devices that lead to this conversion. This is actually highly undervaluing all of these other channels that were impactful in me ultimately making this conversion on your site.
The idea of attribution is that we want to move beyond last-click. There are several different ways of doing attribution. We have our simple rules-based models. These are last-event or last-click, linear, and position-based, and these allow you to look at different rules, different ways of calculating these out in structured manners. So, 100 percent is credited to last-click. Linear is that event leading up to the conversion is given a fair amount or an exact amount of that conversion. Position-based is looking at different positions and weighing them in different ways. Ultimately these are all subjective in nature. How do we know which one is the right one to use for our business? How do we know that that decision isn’t biased by our own thinking and isn’t in fact that right model for our business?
Data driven attribution is the answer that we believe is the right way forward. This methodology not only considers user journeys that ended in a conversion, but also those journeys that did not convert. Then we use machine learning to truly understand the value of each of these touchpoints.
When we look at Google Attribution, it provides the single view to calculate attribution credit for each media touchpoint. In this case, we are looking across Google search, display, and video, all of the clicks on your website or app from Google Analytics, and all of your DoubleClick stack; DoubleClick, Search, DBM, and DCM, Served and Tracked. All coming into Google Attribution to be able to properly attribute that credit back.
I want to give you a real example that you can look at of how this actually works. In this case, we’re looking at two conversion paths. They’re the same conversion path. There’s four ads or four touch points leading up to the ultimate conversion. The first one is the true path, and that one has a 10 percent conversion rate. In the second case, we’ve actually removed the impact of the very first touchpoint, that very first ad that a consumer was seeing. What we see here is that there’s very little impact on the overall conversion rate. It’s only .2 percent different.
However, if we look at that same scenarios, and in this case, we’re removing the second ad or the second touchpoint in that conversion path, we see that there’s a much more substantial impact on the overall conversion rate. 3.5 percent difference. I think this is a great basic example to help you start to understand how we’re attributing different credits to different touch points in our attribution model. We’re doing a lot of this analysis to understand which are the most impactful and which are less impactful touch points in that journey.
Let’s go ahead and look at the future of Google Attribution and where we’re really taking this product.
At the very foundation of Google Attribution, we unify the data. I already walked you through that a couple of slides ago where I showed you all of the different data sources coming into Google Attribution. The next step is to analyze the performance. We run these attribution models across devices and channels and we use Google’s device graph to understand touchpoints across devices. Finally, we’ll take action on it by sending those results back into AdWords so you can immediately act on the results that you’re seeing.
I think it’s an important call out here to mention that in Google Analytics 360, in the data-driven model that is there, we are analyzing the last four touch points to go into that model. With Google Attribution though, we’re actually looking at unlimited touch points. There are many more things that can come into this model and make it a much truer understanding of the impact of these various touch points.
For an attribution solution, I think that you need four things. I’m going to walk you through these four things. The first is easy setup, it needs to be easy to implement. The next is that it should work across channels. It should work across devices. And finally, you need to be able to take action on the results that you’re seeing.
Let’s start with setup. You should not need a team of data analysts or data scientists to implement a better attribution model and you shouldn’t have to wait months to get results. We’ve seen clients in previous version of attribution tools take months to setup or five months before they’re seeing any data. It’s a very manual process. That’s why you prioritize building a frictionless onboarding flow. You can have you Google Attribution account set up in minutes with just a few clicks. If we look at that, it really is very simple to setup.
You login, we automatically identify your Google AdWords and Google Analytics accounts. You select the analytics property that you want to link and the right view that you want to use in that property, and then we’ll automatically pull in the AdWords account that are already linked to that property. The result of this is that within a few clicks, you’re setup to go. In about 48 hours, you’re going to see some data start to come into this attribution model. Within 30 days, you should have a very actionable attribution model. We’re really taking that timeline that used to be months and scaling it down to make this a quick and easy setup for you.
The next is that attribution should be comprehensive and it should work across all of your channels that you’re marketing on.
That’s why Google Attribution will look at all of the available interactions across all of you channels. It will also include impressions, which is particularly relevant when it comes to display. Over time, we’ll improve this even further by adding coverage for apps and also integrating other DoubleClick products, like DoubleClick Search.
Third, we want to look across devices, because again, we want this to be a truly comprehensive solution. With Google Attribution, we are using Google’s device graph. Here’s how that works. You probably know that Google has seven services with over a billion users. We have a strong sign-in rate across all of the devices, platforms, operating systems, and browsers using the services. We use sophisticated algorithms to extrapolate for all of the users who are not signed into a Google service and using that information, we’re able to calculate out a device graph to be able to de-dup these users and find a true user across devices.
Now that you have a unified view of your performance across channels and across devices, next you need to choose a new attribution model, one that can accurately value all of your customer touchpoints. With so many marketing moments, it doesn’t make sense to give all of the credit to the last ad a person experienced. As I mentioned earlier in that example, there were five other touchpoints before that direct visit that lead to y conversion. In that case, it really did not make any sense to attribute my conversion to that last click, to that last visit solely. Because last-click attribution has many of these clicks in the conversion paths, it just doesn’t make sense.
Historically, attribution has been considered a subject for PHDs, econometrics, but we’re really determined to change that. I just showed you how easy it was to setup and that you don’t actually need a team of data scientists to do that. That’s one way that we’re trying to make this really accessible for everyone. Another way, is that we’ve introduced this data-driven attribution model so you can let your data decide how much credit to award to each touchpoint. So, we’ve talked about the easy setup, cross-channel, cross-device, now let’s look a little bit more at the model. We’ve already talked about what DDA means and that last-click ignores many of the clicks in the conversion paths and with Google Attribution, you can compare multiple models against the DDA model to better understand which channels you might have been undervaluing or overvaluing.
Here we are in the product and we want to look at this model comparison report to understand the difference between our DDA model, or data-driven models, and the other models that we have at our disposal. In this case, we are doing a side-by-side comparison of our last interaction model, that last-click model, and the data-driven model. You can notice that in this example, some channels like display and generic search are significantly undervalued using the last-click model. But with the data-driven attribution model, we have a more accurate and realistic view of the impact of our marketing.
Finally, the insights from your attribution model are only as useful as your ability to take action on them. That’s where I think Google Attribution really stands out. It’s where we really take this to the next level by allowing you to immediately act on these results.
With just a few clicks, you can make your results available for reporting and bidding in AdWords. In short, this is a powerful feature for optimizing your performance. The numbers that you see in AdWords now affect how your ads actually work together with your other channels, like email, organic search, etcetera, to drive your conversions.
I’d like to just sum up this part of the presentation with a quote: “Understanding the customer journey is a must have for marketers. You can only define the right marketing channel strategy when you know how many touch points customers have with your brand and the role of each interaction.” In short: “No data on customer touchpoints, no party.”
To review this section, we want to easily gain insight with Google technology and machine learning when it comes to attribution, doing this with quick time to value, intuitive platforms and insights, and action ability via budgeting recommendations and programmatic bidding connectors.
Our next topic or key trend that we mentioned at the beginning is that data is complex and varied. It’s not only an explosion of data, but it’s also coming from so many different places. In fact, the typical marketer uses 13 channels to drive his or her marketing objectives. As a result, a typical analyst or marketer has an overwhelming amount of data to sort, understand, and draw insights from. This leaves a lot of room for error or oversight. That’s what we want to help prevent or what we want to make better for everyone within your organization.
Even with all of the right data, our customers still face many challenges. How can I quickly identify critical changes and growth opportunities in my analytics data? How can we empower all users to get answers to basic analytics questions? I know how users already behave on my site or app, but what are they going to do next? I know this is a virtual summit so I can really ask you to raise your hands, but I bet, as I was saying each one of these things, you’re probably nodding along and feeling the pain of these types of challenges that each of your businesses face.
Our vision for this, is Analytics Intelligence. We want to use machine learning to help customers understand and act on their data. We’re actually going to do this in three ways. We’re doubling down in three areas or investment to make this vision a reality. The first, is that we want to surface actionable insights. We want to automatically identify major changes in growth opportunities along with recommended actions to save analysts time and make sure that they don’t miss out on important opportunities.
The second area of investment here is that we want to democratize data. Our goal is that anyone can understand their data by asking questions in plain English. Analytics Intelligence is a part of this vision to make analytics data more accessible and more useful for everyone, regardless of your business size, regardless of your analytics expertise or background. We want this data to really be actionable for you.
Finally, we want to detect and predict user behavior. This is why we’re bringing machine learning into the mix, to help our customers maximize the value of their data. This also includes another predictive model.
A key thing that I wanted to note here is that there’s a bit of a differentiator when it comes to Google’s existing technology stack. We have all of these world-class products and they’re already using these things like TensorFlow, DeepMind, and Google Brain. When I talk about Analytics Intelligence and building on machine learning and predictive modeling, we’re not starting from scratch. We’re starting at a very good place in terms of using all of the different work that has gone on in Google already to forward machine learning in this field.
Let’s look at surfacing actionable insights.
This is a screenshot of the Analytics Intelligence feature on mobile app for Google Analytics. Some of the things that we see here are top landing pages that might be underperforming and we’re going to give you improvement opportunities, suggestions. Another thing we might see is an important page is loading too slowly. In this case, we’ll have diagnostic tools and speed suggestions for you. We also might see that there are high performing keywords or landing page or products that are missing from your media campaigns, and we would recommend that you consider adding these.
Common customer concerns. What are the statistically significant changes in the metrics and dimension slices that I care about.
I’m excited to say that we’ve recently launched anomaly detection insights for key metrics and dimension slices. If you’ve used Google Analytics for a long time, you are probably thinking to yourself, “Krista, we know that there’s been this anomaly detection alert thing in Google Analytics for quite some time and it really wasn’t very good.” And I would say to you, “You’re right. We did have alerts and they would take you to some report that didn’t have a lot of data and certainly didn’t point out what the problem was for you. And I’m sorry for that, but going forward, I’m super excited about this new anomaly detection feature that we are using because I do thing that it’s actually insightful and actionable.”
I want to walk you through an example of this. Here we are looking an anomaly in daily revenue for women’s sneakers. We have a four-step process for classifying this as an anomaly.
The first step is to get the background data. In this case, we’re looking at the previous 90 days’ worth of data because this is a daily anomaly. If it was a weekly anomaly, we’d be looking at the previous 32 weeks of data. An important note here is that we only run this on profiles that have over 500 sessions per month and that consistent data as a metric.
The next step is that we’re going to use time series hard testing to predict the forecasted value for the current day. For this, we’re using a Bayesian state space model just for trends and weekly patterns from the previous time period. Next, we are going to do significance testing on the forecasted value versus the actual value. If we classify or we see that the value is outside of our forecasted bounds, step four is that we classify this as an anomaly. If we do classify it as an anomaly, we will give you a reason why or we will try to give you a reason why.
In this case, the drop is likely due to a new sneaker creative. You might have just launched a new creative in market. It is not performing as well as your old creative. You might want to consider revisiting the old creative or switching out the new creative or AB testing something else because this one doesn’t seem to be performing for you.
Our next area, when it comes to analytics intelligence, was to democratize data.
This quote here, “I spend about half of my time answering relatively basic analytics questions for other people in my organization.” This comes from a GA 360 client, but I bet when I read this to you—I can’t ask for a show of hands, but if I didn’t, I think everybody in the audience would be raising their hands for this one. We spend so much time as analysts answering basic questions. Somebody asking me, “How many people came to my site last month?” or “Where did they come from? How many people converted?”
These are not the questions, as an analyst, that I want to be spending my time answering. I want to be spending my time doing much more exciting and more in-depth pieces of analysis for my organization. But I still have all of these basic questions that I have to answer. That’s where democratizing data really comes into play.
We want to get better with machine learning, we are getting better with machine learning, to really ask and answer these questions for you. You see a little video playing on your screen right now of an example of the analytics intelligence feature on both desktop and web. I’m going to go ahead and switch into a quick demo because I want to show you this feature in real life here.
Demo screen 1:
Here we are in the Google Merchandise Store. This is our demo account, so anybody can come to the Google Merchandise Store and buy Google branded gear and we make this account available for all of our end users to be able to play with the data. In the top right corner, we see a feature called “Intelligence.” If I click that, my Intelligence bar pops out here and I can start to ask questions. My very basic questions that I mentioned that I hate answering: “How many people came last month?” Type that in.
Then I get an account for my users for the month of October, I had 88 thousand users. Notice that I didn’t type something like: “How many users came to my website in October?” I typed: How many people came last month? You can start to see some of that natural language practicing that we’re getting at with this tool. Another example would be: where did they come from? In this case, I’m going to get a report of my top 10 countries by users to see where those users came from. This is great. These are pretty easy metrics to be putting out here, but I think this is only the first step. I think the really important thing is getting passed the simple questions to the why. That’s why people really want to know is why did they happen.
I noticed we had a change in users from the last week to this one. I’m going to ask: “Why did users change?” it is going to give me a week over week comparison. It’s going to say your users decreased almost by 10 percent week over week. It’s going to give me a couple of segments or places I can look further for the reason why my user count might have changed. I had a significant drop in new users, 11 percent less new users this past week. I can go to the full report and analyze this in more detail if I wanted to, but I wanted to quickly highlight this and show an example of how we’re starting to take this to the next level with that analysis piece of this intelligence feature. You can expect many more things to be coming here in the foreseeable future.
I also wanted to note that this is not something that is only going to be living in Google Analytics. Google is going to be bringing natural language integration to the enterprise. You’ve probably seen this with the explore feature in Google Sheet, but we really believe that semantic intelligence-powered voice interfaces are the next logical step in how we interact with technology.
The last part of analytics intelligence that I mentioned was to detect and predict user behavior.
Another quote for you: “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” This comes from the early 20th century. I’m sure everybody in this stream has heard this quote before. It’s not new, but what is new is we don’t want this to be the case. We want to help you predict how your advertising dollars are being spent, so you know exactly which half is effective and which half you need to take action to change.
To do that, we have introduced a few new features for conversion modeling over the past year. First, we have Smart Goals. These identify sessions that are highly likely to contain a conversion. Second is Session Quality. This is a quality score for sessions based on their likelihood of conversion. And the last one is Smart Lists. These are auto created remarketing lists for users who are highly likely to convert. There are only three on this list, but we have a long roadmap of other things coming to this list. So, stay tuned for a lot more in this area on how we’re really helping to detect and predict user behavior.
To sum up this section, we looked at natural language search and automated insights, we looked at predictive metrics, and smart budget recommendations through that predicted conversion modeling.
Our last key trend to analyze is visualizing data across systems. 61 percent of marketing decision makers said that they struggle to access or integrate the data that they needed.
This is where Google Data Studio comes in. our vision for Google Data Studio is to help users use data to make better decisions.
How do we communicate data?
Data Studio allows you to tell a story with your data. You can connect with many different data sources to bring data in from multiple analytics accounts, from spreadsheets, from other sources of information, and visualize it in a really beautiful way. I think one of the things that sets Data Studio apart in terms of dashboarding in particular is ease of sharing that dashboard across your organization. It’s a really beautiful way to visualize data, bring it together, and share it throughout your organization.
We want to bring all of your data into one place with Data Studio. Here you can see that we have data from Google Analytics, BigQuery, Google Sheets, all coming into Data Studio. One of the things that we’re really excited about that we’ve just launched is called Community Connectors. This allows you to bring in that much more data. We know that our native connectors have access to a good amount of data to bring into Data Studio, but as businesses, you have data in your CRM system, from your social tools, from email. It’s really important that you’re able to break down those data siloes and bring it all together to visualize and analyze in one place.
You want a complete story. As I mentioned, we have a lot of integrations here, but we need more data.
That’s where the Community Connectors have finally arrived. A new connector ecosystem, is what this is that allows you to access all of your data from so many different places to build your own connector using app script so developers and partners can now build connectors to any first or third-party data sources. We already have—my slide says 200—but I think it’s actually over 600 at this point new data sources and connector that have been built since we launched this just a few weeks ago. There are many more on the way. I think this is super exciting.
To give you a little bit more insight into that, these are the connectors that we had available at our developer one, which again was only a few weeks ago. As I mentioned, we’ve already had so many more submissions for these connectors to come into our set. So, a lot of different data sources that you can connect to through Data Studio, bring that data in right alongside your Google Analytics, BigQuery, AdWords data and visualize all of that data together in one place.
To sum up this section of the talk, simplified visualization and open platforms for data analysis are really important and we’re doing that simplified visualization using Data Studio and the data integrations you can use in that platform. Deeper data integrations with a lot of Google product, and community data connectors with all of your other various data sets.
If we go back to those key trends that we were talking about in the very beginning, we have an explosion of data and information, our data is very complex and varied, and visualizing data across systems is tough.
But, we’ve found ways now, through this presentation and through products that we are working on at Google, to solve some of these or to make some of these pain points easier for you. We’re bringing multiple touch points together in Google Attribution. We’re incorporating machine learning to more product so users can easily gain insight. And we’re investing in simple visualization and open platforms for data analysis.
Thank you very much, and again, I want to thank ObservePoint for having me back at this virtual summit. I hope you guys can see some of these key trends that we are focused on and how we are looking at the market going forward into 2018. Thank you.