Drive better results for your business with machine learning: Part III

(Part 1, Part 2)

Consumers are more empowered than ever before and expect brands to provide fast and helpful experiences. That’s why leading marketers are 50% more likely to increase investments in capabilities like machine learning to help them win.1 In fact, brands like Rappi and AutoGravity are already using machine learning in AdWords to reach their most valuable app users and grow their businesses. In our final installment of this series, we explore how machine learning is being applied to bid optimization to help businesses make sense of the data around them and get better results at scale.

It’s more than a bid

The days of predictable web sessions are over, replaced by bursts of digital activity throughout the day on multiple devices. Your bids now have to take into consideration a wide range of contextual signals that impact ad performance, including a user’s device, location and time of day. That’s where machine learning can help.

AdWords Smart Bidding uses Google’s machine learning to help you set the right bid for every auction through three core capabilities:

  • Auction-time bidding: Smart Bidding sets bids for each individual auction, not just a few times per day. AdWords Smart Bidding evaluates the relevant contextual signals present at each of those auctions—such as time of day, specific ad creative being shown, or user’s device, and browser—to identify the conversion opportunity, and set an optimized bid tailored to each auction. This allows Smart Bidding to set millions of bids per second, something even an army of marketers wouldn’t be able to do.
  • Cross-signal analysis: Smart Bidding understands how signal combinations impact conversion rate. For example, a retailer might notice their mobile conversion rates are 20% higher than their desktop conversion rates, and set a mobile bid adjustment of +20%. However, this doesn’t account for the times of day where mobile conversion rates are even stronger, like in the mornings, when people are researching during their commute. Smart Bidding analyzes billions of these types of signals to identify meaningful correlations, and calculates bids based on how likely a conversion will occur.
  • Query-level learning: Smart Bidding maximizes performance for new and low-volume keywords. By looking at performance data across similar auctions in your account, Google’s machine learning platform makes informed bidding decisions and helps reduce performance fluctuations even when data is scarce. For example, let’s say you just added a new keyword “cheap flights to NYC.” If that query was already matching to another part of your account and similar auctions, Smart Bidding simply applies what it’s learned about that query to set the best possible bid.

Focus on the next big opportunity

Brands around the world are using Smart Bidding to unlock growth for their business and reinvesting their time and money into new opportunities.

Harmoney, a peer-to-peer lending service in New Zealand, teamed up with its agency, First Digital, to find more, qualified applicants while still hitting an aggressive ROAS goal. They used Target ROAS across their non-brand Search campaigns to reach customers who were most likely to apply and be approved for a personal loan. As a result, Harmoney saw a 219% growth in high-value accounts at a 37% lower cost-per-acquisition (CPA). Importantly, Smart Bidding freed up 5 hours per week for the team to focus on high-value tasks like testing ad copy and learning more about their best customers.

FirstPoint is a Swiss-based digital agency that wanted to maximize its client’s Search budget while driving more conversions. After testing Smart Bidding, the agency moved away from manual bidding in favor of Maximize conversions. It increased conversions by 2.4x, increased conversion rates by 12%, and decreased CPA by 59%.

Put machine learning to the test

Moving to Smart Bidding and enabling machine learning to do the heavy lifting for you doesn’t have to happen overnight. Set up a campaign draft and experiment to run a 50/50 split test and see how your old bidding strategy stacks up against one powered by Google’s machine learning. With a little time, you may find yourself delivering better results. For our own media team at Google, Smart Bidding is now a best practice and is enabled across 98% of eligible campaigns.



Check out our updated Smart Bidding guide for best practices on picking the right bid strategy for your business goals.


1. Econsultancy and Google, Marketing and Measurement Survey, 2017

Grow your business faster with machine learning: Part II

(Part 1, Part 3)

Last week at the Consumer Electronics Show, we learned about today’s more empowered consumer. They’re more curious, demanding and impatient than ever before, and expect assistive experiences everywhere–like checking in and unlocking their hotel room using their smartphone.

Meeting these rising consumer expectations is critical. Over the next two weeks, we’ll explore some of our favorite AdWords products and show how machine learning is enabling brands to meet those expectations, while saving time and improving performance.

Applying machine learning in AdWords

Campaign management involves time-consuming tasks. Rather than manually adding thousands of keywords or individually testing headlines to see which ones work best, you can train Google’s machine learning platform to do it for you.

For example, you might’ve had new products added to your inventory or more content added to your website recently. Dynamic Search Ads would see this and automatically fill gaps in keyword coverage to help you reach people who are searching for those new products and services.

Or to show relevant ads that fit anywhere across the millions of sites in the Google Display Network, you can upload more creative assets to your Smart display campaign and automatically show relevant ads to the right people. Machine learning makes all of this possible.

Changing the app game

For app developers and marketers, we know competition is fierce: the number of developers with more than 1 million monthly installs grew by 35% year over year.1 There are more apps and experiences competing for your users’ attention and dollars than ever before. This is another area where machine learning is changing the game.

Universal App campaigns (UAC) enable brands like Rappi, a delivery service in Latin America, to reach their most valuable users across Google Play, Search, Display Network, and YouTube with a single campaign.

Rappi uploaded as many creative assets as it had, allowing Google’s machine learning platform to rotate each asset, understand which ones perform best across each channel, and show the ads that users are most likely to engage with. After only one month, Rappi’s conversion rates grew by 10X, and the brand expanded into Brazil, Mexico and                                                         Argentina.

AutoGravity, an auto financing company, reached tens of thousands of car buyers and increased user engagement by 120% in only 5 weeks. The brand plans on increasing UAC investment by 200% to reach more of its highest-value users, people who are most likely to receive credit ‘approval’.

How does UAC reach these types of high-value users? Google’s machine learning platform uses insights from Google.com and Google Play, web data and other signals, in addition to information about your app. This data is analyzed across each channel where AdWords shows your ads and is updated in real time. That’s how AdWords can quickly pick up on trending keywords, like events and holidays, and ensure your ads show to the right users.

AdWords then looks at people who have completed your selected action, like ‘approvals’, and those who haven’t, as well as user signals that are unique to each auction. Device type, operating system, network, apps they already have, and other signals create patterns that help identify high-value users. These patterns are used to predict future auctions, where and how to bid, and what ads to show and to whom.

Using machine learning, brands are not only delivering better performance at scale, but they’re also unlocking their most precious resource: time.

Paul Teresi, Growth Executive at Skyscanner, a travel app, says he’s been able to save a lot more time thanks to UAC. “Now, I can focus on truly understanding our users, metrics, and discovering growth and expansion opportunities necessary to keep us ahead of the curve.”



To learn more about how Universal App campaigns can help you reach your most valuable users, take our new education course.

Next week, we’ll conclude our journey with a look at how machine learning is being applied to bid optimization, including an interesting case study from Google’s in-house media team.


1. Google Internal Data, May 2017

Grow your app business with Google’s new education program for Universal App campaigns

Today, we’re launching a new interactive education program for Universal App campaigns (UAC). UAC makes it easy for you to reach users and grow your app business at scale. It uses Google’s machine learning technology to help find the customers that matter most to you, based on your business goals—across Google Play, Google.com, YouTube and the millions of sites and apps in the Display Network.

UAC is a shift in the way you market your mobile apps, so we designed the program’s first course to help you learn how to get the best results from UAC. Here are a few reasons we encourage you take the course:

  • Learn from industry experts. The course was created by marketers who’ve been in your shoes and vetted by the team who built the Universal App campaign.
  • Learn on your schedule. Watch snackable videos at your own pace. The course is made up of short 3-minute videos to help you master the content faster.

  • Practice what you learn. Complete interactive activities based on real life scenarios like using UAC to help launch a new app or release an update for your app.

    So, take the course today and let us know what you think. You can also read more about UAC best practices here and here.

    Happy New Year and hope to see you in class!

    A New Approach to YouTube Monetization

    There’s no denying 2017 was a difficult year, with several issues affecting our community and our advertising partners. We are passionate about protecting our users, advertisers and creators and making sure YouTube is not a place that can be co-opted by bad actors. While we took several steps last year to protect advertisers from inappropriate content, we know we need to do more to ensure that their ads run alongside content that reflects their values. As we mentioned in December, we needed a fresh approach to advertising on YouTube. Today, we are announcing three significant changes.

    Stricter criteria for monetization on YouTube

    After careful consideration and extended conversations with advertisers and creators, we’re making big changes to the process that determines which channels can run ads on YouTube. Previously, channels had to reach 10,000 total views to be eligible for the YouTube Partner Program (YPP). It’s been clear over the last few months that we need the right requirements and better signals to identify the channels that have earned the right to run ads. Instead of basing acceptance purely on views, we want to take channel size, audience engagement, and creator behavior into consideration to determine eligibility for ads.

    That’s why starting today, new channels will need to have 1,000 subscribers and 4,000 hours of watch time within the past 12 months to be eligible for ads. We will begin enforcing these new requirements for existing channels in YPP beginning February 20th, 2018.

    Of course, size alone is not enough to determine whether a channel is suitable for advertising. We will closely monitor signals like community strikes, spam, and other abuse flags to ensure they comply with our policies. Both new and existing YPP channels will be automatically evaluated under this strict criteria and if we find a channel repeatedly or egregiously violates our community guidelines, we will remove that channel from YPP. As always, if the account has been issued three community guidelines strikes, we will remove that user’s accounts and channels from YouTube.

    This combination of hard-to-game user signals and improved abuse indicators will help us reward the creators who make engaging content while preventing bad actors and spammers from gaming the system in order to monetize unsuitable content. While this new approach will affect a significant number of channels eligible to run ads, the creators who will remain part of YPP represent more than 95% of YouTube’s reach for advertisers.

    Those of you who want more details, can find additional information in our Help Center.

    Manually reviewing Google Preferred

    We’re changing Google Preferred so that it not only offers the most popular content on YouTube, but also the most vetted. We created Google Preferred to surface YouTube’s most engaging channels and to help our customers easily reach our most passionate audiences. Moving forward, the channels included in Google Preferred will be manually reviewed and ads will only run on videos that have been verified to meet our ad-friendly guidelines. We expect to complete manual reviews of Google Preferred channels and videos by mid-February in the U.S. and by the end of March in all other markets where Google Preferred is offered.

    Greater transparency and simpler controls over where ads appear

    We know advertisers want simpler and more transparent controls. In the coming months, we will introduce a three-tier suitability system that allows advertisers to reflect their view of appropriate placements for their brand, while understanding potential reach trade offs.

    We also know we need to offer advertisers transparency regarding where their ads run. We’ve begun working with trusted vendors to provide third-party brand safety reporting on YouTube. We’re currently in a beta with Integral Ad Science (IAS) and we’re planning to launch a beta with DoubleVerify soon. We are also exploring partnerships with OpenSlate, comScore and Moat and look forward to scaling our third-party measurement offerings over the course of the year.

    The challenges we faced in 2017 have helped us make tough but necessary changes in 2018. These changes will help us better fulfill the promise YouTube holds for advertisers: the chance to reach over 1.5 billion people around the world who are truly engaged with content they love. We value the partnership and patience of all our advertisers to date and look forward to strengthening those ties throughout 2018.

    Grow your business faster with machine learning: Part I

    (Part 2, Part 3)

    At the start of the new year, we take time to look at what’s ahead, from eating healthier to spending more time outdoors. This week at the Consumer Electronics Show, we get to take a similar look ahead, at the future of technology. Thanks to innovations like smartphones and voice-activated speakers, consumers are now super-empowered and expect more from their favorite brands. This is redefining the consumer experience and reshaping what’s required of marketers.

    To help you meet rising consumer expectations, over the next three weeks we’ll share insights and best practices from brands that have made machine learning an enabler for new opportunities in this “age of assistance”–instead of another challenge to figure out.

    Solving problems with machine learning

    At its core, machine learning is a new way of problem solving. Rather than spending hundreds of hours manually coding computers to answer specific questions, we can save time by teaching them to learn on their own. To do that, we give the computer examples until it starts to learn from them–identifying patterns, like the difference between a cat and a dog.

    To illustrate how machine learning can help solve some of the most complex problems in the world, take the latest advances in medicine. In the US, doctors know survival rates for skin cancer increase dramatically with early detection.1 That’s why researchers at Stanford University used Google’s machine learning platform, TensorFlow, to train a model that can identify cancerous skin conditions from healthy ones with 91% accuracy–on par with 21 board-certified physicians.

    New opportunities to accelerate growth

    As marketers, you don’t wake up everyday expecting to save lives. But we do ask ourselves a very different question: how can I grow my business faster? This is where Google’s machine learning technology can help.

    We know that choosing where your ads show and manually adjusting bids is time consuming, leaving less time for strategic tasks, like capturing the latest trends or entering new markets. Google’s machine learning considers billions of consumer data points everyday, from color and tone preference on mobile screens, to purchase history, device and location. With products like Universal App Campaigns and Smart Bidding, it’s now possible to use this data to help deliver millions of ads customized for your customers, and set the right bid for each of those ads–in real time.

    Even if you’re not using these AdWords innovations, you’re still seeing the benefits of machine learning. Google uses information about search queries, historical ad performance and other contextual signals combined with machine learning, to predict whether or not someone will click on your ad. This predicted click-through rate helps determine the selection, ranking and pricing of your ads–meaning machine learning is already working to show the right ads to the right customers.

    Over the next three weeks, we’ll continue exploring how you can use machine learning to reach your marketing goals and grow your business faster. To get the latest updates on this series, follow along on the Inside AdWords blog or subscribe to our Best Practices newsletter.


    1. Stanford News, 2017