The Future Of Marketing Is Here : AI & Machine Learning
Machine learning and AI has been around for many years.
We are now at a tipping point where marketers for companies of all sizes have access to this technology and are able to put it to incredible use.
- A concise explanation of what AI & Machine Learning is
- How AI & Machine Learning can help you in a marketing role
- Examples of successful applications of Machine Learning
- Advice for marketers who want to make a start with AI & Machine Learning
James: Hi, I’m James Rostance and welcome, to The 414. Each week with some of the greatest minds in marketing.
And joining me today is a man with 20 years experience in digital and publishing, who’s made a name for himself in the successful study of audiences, why people do things, and how they behave.
He’s now one of the UK’s foremost experts in AI and machine learning in marketing.
And it’s my pleasure to introduce Steve Masters.
So Steve, What would be your most concise answer for what AI and machine learning is, along with its potential for marketeers?
Steve: It’s a replication of, or an alternative to what we consider to be intelligence.
So it’s artificial, which can be robots replacing us, but it’s not true intelligence, because it’s not thinking for itself.
Machine learning is about applying a learning structure to that.
So you would teach a machine to do something, and then you allow it to learn from what it’s doing so it can adapt its behaviour.
There’s also deep learning, which is even deeper than that where instead of telling it what to learn and what to do, you’re allowing it to make its own decisions and find out its own information and probably adapt its behaviour in ways that even we haven’t planned for.
James: So could you give us some really cool examples of AI already in use in a marketing?
Steve: Well, one really famous example happened a few years ago. I think it was 2004, where Walmart in America decided to start using data analysis to predict what products they would sell.
So, before a hurricane they analysed the behaviour of what happened in a previous hurricane to see what people most wanted to buy.
And they discovered that strawberry Pop Tarts increased by 7x normal sales volume
before a hurricane for some reason. And after as well.
Another one is Netflix, Netflix is a really interesting company because it’s effectively data driven television.
If you’re a customer of Netflix then you will know that when ever you look at the interface you will see a different thing, because it serves different choices of movies and television for each person based on their interests.
And that comes from tracking data on what they watch, how long they watch it for.
And Netflix is building a picture of what we like, and it’s actually building its programmes around the data that it has.
Let’s take a classic example of a client of ours, which is a gardening company.
Seasonality is important for any business, but especially for gardening where people buy certain products at different times of the year, we can not only track seasonality but we can also predict behaviour for different circumstances to work out when is the most likely time to invest money for advertising, or how to predict things like the Pop Tart story, unexpected events that you might think well if I was going to just base it on assumption, I could be wrong and the data might prove otherwise.
James: So to what extent can AI technology set about doing your job for you in marketing?
Steve: Well, two particular things that spring to mind first is the ability to audit the website.
There are thousands of data points that can affect a website’s ability to perform, either in search or with the user, it would take an age to analyse all those data points manually.
The other area is that when you have information about the relationship between pages, words, content, and users you then have to ask a million questions about what permutations are going to make a difference.
You can apply robotic technology, software algorithms to do the analysis for you in a multitude of ways simultaneously at the same time.
Whereas a human can only do things in a linear fashion.
James: So what would you advise for marketers who want to make a start with machine learning and AI?
Steve: Quite simply, to make sense of it I would say think about all the things you do repetitively that are manual that could be replicated by a robot.
So if you for example use spreadsheets all the time to do an analysis of data, and you’re doing the same thing every week or every month and sharing data with your bosses, why can’t that be done automatically?
You can think of all the things that are repetitive tasks that following the same process every time and think about ways to automate those processes using tools that may be available, or even services that might be available in the cloud.
And you can start using AI technology that is already out in the marketplace.
James: Steve thank you very much.
Steve: Thank you.
The 414 EXTRA
This is where we get the chance to look a little deeper into the most interesting elements of the content that we’ve just covered in the main show
Steve shares with us some detailed examples of successful project he’s delivered recently, as well as a deep dive into what Regression Analysis is.
Click HERE to watch Episode 2.
James: Hello and welcome to The 414 Extra.
Now, this is where we get the chance to look a little deeper into the most interesting elements of the content that we’ve just covered in the main show.
So, I’d like to find out a little bit more about your current most favourite successful applications of AI in projects that you’ve delivered.
Steve: Ok, I can talk about two particular examples that are current favourites of ours.
The first one is an organic project. And when I say organic, I mean organic search, so it’s SEO related.
So we have a customer that is a garden client, the name is Sarah Raven.
A very famous flower and gardening retailer, and we’ve been working with them for quite some time.
Now, what we did for them is we had to work out how to use SEO to increase their traffic but also their sales, because they’re very sales driven being a retailer.
Now the thing about SEO is it’s very much about visibility and traffic and things like that and it’s not always directly related to actual sales, but what we did to make this work is we analysed their pages.
So we sat down and we said first of all, “What kind of page on the website is most likely to drive a sale?”
And we looked at product pages, and category pages and information pages and we ran them through a regression analysis model where we’re effectively looking at past behaviour to identify whether a particular type of page is more likely to get a conversion.
Using that analysis we then identified that category pages work better.
So in a category page being a page for example if you’re selling dahlias or tulips a category page contains lots of different products of the same type, right.
We found that people are more likely to convert if they land on a category page than if they land on a single product page.
So we focused on those pages.
Once we had the analysis which we had done using the regression model.
We then started to look at how do we increase sales on those pages using organic search?
And it actually was really simple what we did, we looked at the product one of the products in search results and we had a look at how the pages appear in search results.
We identified that there are two pages that appear side by side in Google.
For example one of them is a blog post, and one of them is a category page.
So if you’re searching for a particular flower such as dahlias
you would get two results, one is how to look after dahlias or how to grow dahlias.
The other one is a category page to buy dahlias, but it wasn’t well written, it wasn’t well optimised to signal to the reader or to the searcher that one was for purchasing on one was for information.
So we simply just rewrote the information so that the signalling and the sign posting on the pages was more obvious, as a result of that more of the people who were searching to buy the product could see which of the pages to click on and they went through to the purchase page.
People who were looking for information went to the blog post and then of course they can read the information.
So, simply by changing the way the pages appear in search results we increased the activity on the page, the right people went to the page, the revenue went up considerably more than we predicted it would have done naturally, but all of that started by us analysing how different pages perform.
Now if you bear in mind a retail site has thousands of pages thousands of products it would take you until the cows come home to really look through that and prioritise things, so it’s really important to apply some kind of machine learning model or regression analysis or AI technology to kind of do that work for you to then give you the answers you need so that you can then act on that.
What is regression analysis?
Regression analysis, effectively it’s about regressing back through past activity, so you can take multiple data points.
So in this example we looked at daily sales, for I think, a year so we could then think right, on one particular day if you analyse a page
you might think “oh okay three people went to that page and bought some products”.
That doesn’t really tell you much because the next day maybe nobody did.
So we look at the behaviour over time to build a picture of how that page is likely to perform.
And in the real world it would be like for example a retailer measuring footfall.
So if you have for example Marks and Spencers food hall and they have two doors they could do some analysis that says if more people come in through the front door, are they more likely to buy bread and chocolate?
So if they come in through the back door, are they likely to just walk straight through and not see anything?
So you can do all kinds of analysis based on that.
But looking at one person’s behaviour on one particular occasion doesn’t give you the information.
What regression analysis does is it looks at a period of time and looks it looks for trends and patterns and that kind of stuff.
So it gives you a much more coherent picture of how a particular event or page or product and would perform for a multitude of circumstances.
And what was the other example?
So the other example is paid search, so using AdWords. Google Adwords.
Paid search is interesting because it’s very different from SEO.
With SEO you were trying to create a scenario where Google would recommend your page to people.
With Adwords you’re choosing when you should appear, who you should appear to and what they click on when they get there so you have more control.
But the interesting thing about that is it’s very easy to waste your money.
It’s very easy for example to decide on the wrong keywords and your ads appear at the wrong time.
It’s very easy for people to click on your ads
thinking your one thing when you’re not and then you’ve wasted your money.
The other thing that’s really interesting and Google’s done a lot of work in this area using AI technology and and really clever deep learning techniques in its algorithm.
Is it’s able to automate really clever stuff.
So for example if you had a hundred people clicking on your ad and visiting your website, they’re not all going to behave the same way, let’s say 50% of them are female and 50% male, if 10% of them over 65 and 30% of them are under 25 they’re not all going to be behave the same way.
So you can identify breaking down your audiences into categories, which ones are the most profitable.
So now what we’ve done at Vertical Leap is we’ve created a way of automating a bid structure that looks at seven or eight different data points
so what device is the person on? where are they? what keyword have they typed in? what age are they? what gender are they?and what interest group are they in based what Google knows about them.
So we can break this down and think right this particular person with these circumstances is worth spending two pounds for a click.
This particular person we don’t want to pay more than 10p because they’re very unlikely to convert.
So we can actually really cleverly adjust our bidding on an automated basis instead of having to manually look up every data point.
We can use a basically machine learning model that tells us every single day.
What we need to adjust to try to maximise our spend.
By doing that our profitability just increases rapidly because we either can use the same budget to get more clicks which of course converts into more revenue.
Or, we can spend less budget to get the same number of clicks depending on what the client wants us to do.
Steve thank you so much for your time today.