Annika asks, “Can you give me some examples of digital Marketing projects that use AI in some way that failed and succeeded and reasons behind The success or failure?”
Some of the most prominent failures are those in which people didn’t keep their eyes on the system after deployment. In marketing, that’s systems like sentiment analysis, which has a small raft of problems. Microsoft had its very visible Tay failure. Successes are legion and very tactical, along what we call the 5Us. Watch the video for full details.
Can’t see anything? Watch it on YouTube here.
Listen to the audio here:
- Got a question for You Ask, I’ll Answer? Submit it here!
- Subscribe to my weekly newsletter for more useful marketing tips.
- Find older episodes of You Ask, I Answer on my YouTube channel.
- Need help with your company’s data and analytics? Let me know!
- Join my free Slack group for marketers interested in analytics!
Machine-Generated Transcript
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.
In today’s episode, Monica asks, Can you give some examples of digital marketing projects that use AI in some way that failed and succeeded, and the reasons behind the successes and failures.
Some of the the more prominent failures in the space of those in which people didn’t keep their eyes on the systems after deployment, one of the major faults, or reasons for failure and many AI projects is that after you deploy your model, you don’t keep tuning it, you don’t keep adjusting it in the model drifts, which means that it begins to behave unpredictably.
It receives data that it doesn’t know what to do with or forecasts poorly.
So that is sort of the root reason why many AI projects kind of go off the else.
The most prominent example This was having Microsoft’s a Twitter chat bot.
I think it was from 2017, which in which Microsoft deployed this chat bot on Twitter, and it was unsupervised.
They wanted to see how it would interact with users.
And within 24 hours, users had turned it into a racist porn bot.
And that is a function of model drift, where they didn’t anticipate the the lesser aspects of human nature.
And the model drifted and eventually became unusable.
What they should have done that in that particular instance, is had a constant cycle of tuning a constant cycle of reinforcement learning with active learning, which means that a human should have been in the loop at all times.
Continuing to tune the model and saying yeah, that you just got their reject that when Microsoft and IBM had it, some project debater AI, this past February, one of the 10 ai is under the hood was an ethics module that had, in addition to being trained on a bunch of what not to say, actually had some rules hard coded into it that it was never allowed to do to do was never, for example, allowed at any point to reference Hitler.
So that is a really good example of one of the core things, one of the many, many things that can go wrong with any kind of AI is model drift.
And if you want to read more about the ways AI projects failed, is actually a blog series I’m doing over we’re at the trust insights blog, you can see it there on ways AI projects fail.
Another way that is very marketing specific, is that a lot of companies have done a relatively poor job with sentiment analysis, they tend to rely on very, very old approaches for sentiment analysis bag of words, approaches, you know, this word is generally bad.
This word is generally good and of course, fails to take into account any kind of context.
And the reason for that is not AI based in a lot of cases, that is just computational cost based.
The the cost to do more advanced types of machine learning on language, in many cases outstrip what that company is able to spend and still be profitable from a compute perspective, or they don’t have the code and talent to implement modern sentiment analysis.
In terms of successes in AI, in marketing, there are they are legion because many of them are very tactical, and address very solvable, manageable problems for marketing.
Again, this isn’t that we call the five use untapped data, being able to use machine learning to classify and categorize unstructured data like call center calls, emails, etc, and be able to extract information classified very, very quickly.
dimension reduction, which is a huge problem for most marketing data sets.
They have many, many dimensions, many columns in the spreadsheet, if you will.
And using techniques to do feature selection and dimension reduction, understand what’s really important.
There are many, many examples of being be able to do driver analysis, multiple regression in which you are trying to figure out what is at the heart of a problem.
So if you have all this data, say, posts on social media, and you have all the reactions likes, wow, and whatever in posts, views, and shares and comments, which one of those metrics or which combination of those metrics is really impactful? would be a an example of that.
You have techniques like network graphic, which is the type of machine learning, being able to identify the relationships between entities very useful for sentiment and outlet for social media monitoring, for influencer analysis, very, very powerful stuff.
And of course, you have predictive analytics time series forecasting, being able to forecast forward, when is a specific search term going to trend, and that that helps you then set your editorial calendar.
So there are many, many tactical successes for using AI and machine learning in marketing? They are they are legion, I can refer you to a number of talks that we’ve done over again on the trust insights website, detailing this in much greater depth.
But in terms of successes and failures, what sets apart successes and failures in AI and marketing more than anything else, this is something that my CEO constantly says is all about the planning.
Do you have a clear business goal? Do you have a well defined approach? Have you done your technical requirements? Well, at the end of the day, when you’re doing AI, and you’re doing machine learning, the output is software.
So if you are not treating it like a software development project, you’re setting yourself up to fail.
On the other hand, if you follow decades long best practices for software development, knowing that you know a portion of the prod of the development will be done by a machine you will be much more likely to succeed.
So those are some of the basic basics.
As always, if you have fault constantly one of the comments box below please subscribe to the YouTube channel and to the newsletter.
I’ll talk to you soon What help solving your company’s data analytics and digital marketing problems.
This is trusted insights.ai today and let us know how we can help you
You might also enjoy:
- You Ask, I Answer: Reliability of LLMs vs Other Software?
- Almost Timely News, January 28, 2024: Copyright Must NEVER Apply to AI-Made Works
- Mind Readings: What Makes A Good Conference/Event?
- Mind Readings: You Need Passwords for Life in the Age of Generative AI Fraud
- Almost Timely News: Recipes vs. Principles in Generative AI (2024-03-03)
Want to read more like this from Christopher Penn? Get updates here:
 Take my Generative AI for Marketers course! |
Leave a Reply