123 - The Role of Data in Approaching AI Marketing

Share:

Listens: 0

B2B Marketing and More With Pam Didner

Business


Hello, everyone. In the past several episodes, I’ve been discussing AI’s role in Marketing. As I’ve been talking about all the wonderful magic of AI, you might ask, “What’s the catch?” Well, there is a dark side. If you use machine learning as a way to implement AI, you need to make sure you have high-quality data first.   Whenever I mention data, marketers are like “WHAT?” I was like “WHAAAAT? I know, right? Data is really the marketers’ nemesis. So the quality of data: you need to define in terms of the quality of your data and also your data sources.   You need to know the quality of your data. And here’s an analogy from Wilson Peng, CTO of Appen (and I love the analogy!). He said, “If you train a computer vision system for autonomous vehicles with images of sidewalks mislabeled as streets, the results could be disastrous. In order to develop accurate algorithms or predictions, you will need high-quality training data.” And I 100% agree with that.   Amazon famously scrapped its AI recruiting tool because it shows bias against women, given they based their datasets from resumes submitted to the company over a ten-year period. And guess who submitted the majority of resumes to Amazon? Mostly men, a reflection of male dominance across the tech industry.[1] Basically, the way that they used the dataset to train the AI—or Artificial Intelligence—is biased, therefore the recommendations that come down from the AI recruiting screening is going to go against women. So they basically said, “that’s not gonna work” and I 100% agree with that.   As a marketer we often have this love/hate (mostly hate) relationship with data. Most data tends to be scattered in different areas, sources or over different platforms. And please don’t ask me where the data source of my Google Analytics is! I don’t even know.   To work with the data and understand the data sources, it’s very important to work with IT to pull data from various different places. It’s also important to work with vendors or data scientists who actually know how to build the model and have experience to prepare and measure the quality of datasets.   Yes, AI is a lot of work. I never said it was easy, right?. So if you are responsible for AI initiatives, you need to work with IT and different stakeholders and vendors to create in-depth quality control process for data. Can you imagine marketing and quality control? That just doesn’t go together, such as creating multiple quality metrics, and actually having weekly data deep-dives, regular testing and auditing. There is no shortcut when it comes down to the quality control of your data.   Here are some suggestions to mitigate bias from the Inside Big Data site:[2] (yes, there is a site called Inside Big Data!)   1) In addition having the data, you also need to label the data; you’ll need skilled annotators to carefully label the information you plan to use with your algorithm.”[3] You need to be sure they are labeled correctly.   When internal team members label data, they will always add some bias because they have expectations about what their system should conclude. If you decide to use an internal team to label your data, make sure you also consult an outside source to help foster an objective annotations environment. So labeling is very, very important.   2) When you train your datasets, try to find the most comprehensive data and experiment with different datasets, metrics and segmentation to ensure you’ve covered the bases and also tested and see the outcomes of recommendations and compare them.   3) Check for implicit bias in the data as part of your quality-assurance process. Once the product is live, monitor performance using the data it generates to determine whether it’s delivering equitable opportunities and outcomes for all users.   So the training data is critical for a successful AI initiative. Although the quality-assurance processes are tedious (I mean I went through it and I’m pretty sure I put you to sleep!), “Quality training data not only begets algorithms that work in the real world, it also helps mitigate some of the bias inherent in manual data annotation.”[4]   That is the role of data in AI marketing.   If you enjoy the podcast, subscribe to the show on your favorite podcast platform or visit my website at PamDidner.com/podcast.    Again, if you prefer watching video, simply type Pam Didner on YouTube and subscribe. One new video every week.   Be well, and let’s talk again next week. See ya!   [1] Jeffrey Dastin, Amazon scraps secret AI recruiting tool that showed bias against women, Reuters, Oct 8, 2018, https://www.reuters.com/article/us-amazon-com-jobs-automation-insight/amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK08G [2] Editorial Team, How to ensure data quality for AI, Inside BigData, Nov. 17, 2019, https://insidebigdata.com/2019/11/17/how-to-ensure-data-quality-for-ai/ [3] Editorial Team, How to ensure data quality for AI, Inside BigData, Nov. 17, 2019, https://insidebigdata.com/2019/11/17/how-to-ensure-data-quality-for-ai/   [4] Editorial Team, How to ensure data quality for AI, Inside BigData, Nov. 17, 2019, https://insidebigdata.com/2019/11/17/how-to-ensure-data-quality-for-ai/