How Artificial Intelligence and Machine Learning trends have evolved in Real Estate Digital MarketingAugust 1, 2022 3:11 pm
Artificial Intelligence (AI) and Machine Learning (ML) have come a long way in the past 10 years. For a time, the ad platforms were pushing clients to use their AI, however, the results were sub-par compared to human managed campaigns with manual bidding. Eventually several of the largest platforms weaved AI into their core products and as the AI grew larger it also became much better.
It is important to distinguish the difference between AI and ML. AI is used more broadly to describe the machine’s ability to identify opportunities and solve problems, whereas ML is the application of executing the tasks related to solving the problem.
We’ve been following the results of human-managed versus AI-managed campaigns for many years to see which generates the highest performance. Another factor to consider is time. Many AI and ML programs take time to learn things that are already intuitive to humans. Some faster than others.
While there are many AI programs out there, we decided to highlight several of the larger and broader AI programs running at a core level for ad platforms and digital marketers focused on real estate:
Google AI – Focus on Bid Strategy and A/B testing
Google uses a variety of active AI programs in order to serve users the best ad, at the best time, at the best price. To do this these programs use machine learning to conduct complex multivariate tests across a variety of platforms to determine which ad should be shown at any given second. Google’s AI analyzes variables such as ad copy and imagery, web page functionality, and users online behavior in order to make its determinations. Below we’ve highlighted some of the major core AI tools, and their benefits, to be used on a practical level in Google Ads advertising platform.
Responsive Search Ads: This new ad type uses ML to test combinations of different headlines and descriptions to create ads which are responsive to users search queries and intent. With just 15 headlines and 4 descriptions Google can generate up to 45K different ad combinations! After enough data has been collected the combinations and their individual assets are graded. Advertisers can then use this data to tailor their messaging and switch out underperforming assets with more productive ones.
Smart Bidding: Google’s new bidding AI offers an alternative to manually selecting bids for keywords on CPC ad campaigns. Smart Bidding asks users to select a goal, such as Target CPA, Target ROAS, Maximize Conversions, or Maximize Conversion Value, and allow Google’s AI to dynamically bid in each ad auction in order to achieve your stated business goal. Humans can make bid changes based on past performance whereas Google can make real time bid adjustments using factors such as Device type, Location, Hour Of Day, Ad Characteristics, Operating System, Search Query, Placement and our favorite, User Experience. By creating campaigns with different goals you can compare results to see which AI based approach worked best for you.
Google Optimize: Allows advertisers to A/B test different facets of their website, such as creative imagery, pop ups, and site copy. Google’s AI funnels users through the experiment using complex data modeling strategies to ensure the purity of the results. For reporting Google Optimize primarily uses the Bayesian Inference method, which is a statistical analysis that continually re-adjusts results as more data is gathered.
The advantages are this method of data collection are 4-fold:
First, it allows more accurate answers to fundamental questions such as ‘which has a probability of doing better – the original or the variant?.’
Secondly, it allows separation of the notion that p-values are the same as probability and so allows for more actionable items after data has been collected. The p-value is a number, calculated from a statistical test such an A/B test, that describes how likely you are to have found a particular set of observations if the null hypothesis were true and the experiment was never actually carried out.
Thirdly, it avoids hypothesis-based decision making since it determines definitively that one variant will be the best performing out of a number of others.
Lastly, it allows the experiment to be ended when there is no more to be learned from the test.
In a more general way, all the features of Optimize use some of the principles of AI – it automates a previously manual process, it learns from itself, and it is computationally heavy in nature.
Facebook AI – Focus on your ads and custom audiences
Meta-AI is the latest model of machine learning for the umbrella that is all of Meta, Facebook and Instagram included. Their machine learning is primarily focused on serving the best performing ads automatically and then learning more and pushing selected ads. The ML if focused closely on the ad quality scores for generating bids, user behavior on and off Facebook, ad content, time of day, your interactions on Facebook, including with other people. The models get better at predicting over time.
The learning time frame has improved over the years from our observations. We work in tandem with the AI by pushing daily budgets into ad sets the Meta Algorithm selects. They are also working on a new section regarding the creative side and promoting stories and voices from users. Meta-AI we have seen the most success is the advanced matching to build custom audiences using their pixel tracking. We understand over 200 parameters are used in the custom audience results.
Both Google and Facebook have several open-source AI programs available including TensorFlow and PyTorch
Programmatic AI – Focus on your CTR metrics for quality awareness
Display networks use data from over 30 parameters, at the ad platform core, to create dynamic models that are optimizing towards CPC goals on millions of websites. These models are designed to be constantly reconfiguring towards those goals. The algorithmic optimization tools can make automated big adjustments in real time for the placement in campaigns and ads. In many cases results are why we see the same opportunities consistently here. The machine learning is reanalyzing from initial point on every 12 hours.
Third-Party AI Tools – Consider Growth Quadrants for Strategic Planning
There are many available AI algorithms at the core as programs. They analyze multiple parameters such as audience interests, website categories, number of backlinks, common organic keywords, and SERP positions. Based on the competition for online visibility, they allow us to determine similar websites within a market and make an industry. One such market concentration model we use is based on the Herfindahl–Hirschman Index. What we like most about the types of tools are how they can be used to identify who’s growing with a similar target market as you online and the opportunities identified. We use those as one lens into driving more demand.
Element Insights – Get the highest efficacy predictions
Google and Facebook spent decades developing their proprietary AI and ML. And yes! We are building our own proprietary custom AI and Machine Learning program with enhanced behavioral attributes based upon findings further through the funnel, differentiating our insights! We have developed our own tracking and optimization processes off this new data. This is next level customer experience analytics insight. We look forward to sharing more soon!
Conclusion and Next Steps – Never Stop Learning!
The AI is getting better. The most effective ways AI is being used are different for different platforms. It’s like AI is picking up where people leave off in a broad fashion online. We would be remiss if we didn’t also highlight the cyber security AI happening and progression with Google’s Powerdrill. Processing is getting a lot faster in the cyber security world.
AI insights may change your perspective. Discover what you are learning from the AI. Understand AI has a learning curve process of its own too. And never stop testing!