In today’s episode of In-Ear Insights, we’re pleased to present the audio from Katie Robbert’s lunchtime keynote at the HELLO Conference. In this dynamic and humorous talk, Katie covers predictive analytics for social media marketing. You’ll learn common use cases of predictive analytics, the two major branches of predictive analytics, and how marketers can get started using predictive analytics and forecasting. Enjoy the episode, and if you’d like to watch the full video or obtain a copy of the slides, visit this page to get the materials.

Listen to the audio here:

Download the MP3 audio here.

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

So the next speaker I’m so excited to introduce this is one of my really great friends. But she also happens to be a forefront diner MLM. So you wonder, okay, data scientists came up here and CEO of trust and faith, and she is also the queen of high stakes, no mistakes data. So we’re about to get into the scary stuff now. So strap in, and everybody give it up for Katie.

Well, I

don’t know how scary I’m going to be because I will be following along by Chris Penn, my co founder. So before we get into predictive analytics, which is you know, AI and machine learning driven, I firmly have one foot in the human side, and for only one foot in the machine side. So I also have a love of Venn diagrams, I actually mentioned on the wedding vows,

which

God

got the way. So,

you know, if you think of the two human and machine, I’m like, right in the middle, because I very firmly believe that you need the data in order to be more human. And so that’s a bit of what we’re going to talk about today. So,

little bit of housekeeping,

will the slides be available,

be available calm.

And that

is marketing, ladies, it

is a commonly asked questions, and you try to make it as easy as possible.

So many of you may have seen in the most recent cmo survey published in February of 2019. The top uses of and marketing were

customized content and predictive analytics, which is a really exciting thing for social media marketers, because social media marketers want to be creating that customer content to be reaching the audience and what they care about. And with predictive analytics, you want to know what’s the most important and when to be sharing that content. This is really exciting for us to see that CMOS, we’re caring most about predictive analytics, and this customized content, this personalization.

However, when we took a look at the company’s use of marketing analytics to make decisions, while you see that it is an upswing, it is only 43.5%. Less than 50% of companies

are using the data to make decisions which

I struggled with that number, I shouldn’t be surprised, given where I work and what I do. But it’s really disheartening to think that if you’re using data, then you can’t be reaching your audience because you’re just looking at them as data points. And that’s not true. So less than 50% of companies are using data to make decisions.

So what does that mean for social media marketers? Well, we spent a lot of money on social media, however, we’re not seeing social media make big contributions to company performance. And why is that was because we’re not using the right information to plan ahead to measure to

stop those metrics.

We used to work with a very large weight loss company based overseas, and their audiences primarily on Facebook, social media, and they would spend six figures a month trying to reach these people, but not using any data to sort of make those decisions with and it was just really hard. As the ones giving the consultation and guidance to watch have, you could be doing something better, if you were using the data that we’re providing you, then you could be making better decisions, you could be reaching them more efficiently, you could be getting better conversions. Because ultimately, at the end of the day, you want to reach the right people, and you want them to buy the things that you’re selling. So one of the problems that you as social media marketers are facing right now, you’re probably being asked to create content that has no plan, you don’t really sure why you’re creating it. The timing of when you’re sending it out. You have client demands or manager demands with no clear path to success. You can’t measure what you’re doing. And you’re feeling like it constantly behind you’re constantly reacting today. And you’re not being proactive with the information that you have. Because it’s, you know, this crisis columns or something came up or someone had a negative comment on Facebook. And so you’re reacting to everything you can never get ahead of the game. So how do we solve these problems? And that’s what we’re going to talk about today is, how do we start to solve some of these problems in a very practical way, a very tangible, reachable, approachable way. And that’s where predictive analytics comes in. So we’re going to talk we’re going to give you a better understanding of predictive analytics, we’re going to talk through some practical applications and ideas to use. So predictive analytics right now. And then we’re going to help you ditch this, this is how we’ve always done it mantra, which we’ve all been there is terrible.

Because I am the more practical, more responsible one of the company.

I feel like it’s important to sort of go through a little bit, a high level, sort of were picked and analytics

falls into the data analytics hierarchy. So you can’t necessarily just jump into doing predictive analytics, if you don’t have the foundation underneath you to make sure that you can support it. So the very bottom layer is your descriptive analytics, or what happened. And this is your Google Analytics, your Adobe analytics, your CRM, your quantitative data, that helps you understand what decisions were made by your customers. A lot of companies get stuck at this bottom rung, I used to work for health IT company. And they were really strong in the research and academics. We worked pharmaceutical companies. And they were really great with data, except when it came to collecting data about what was happening with the

customers

to the point where we were so lost as a team that they let go of the sales and marketing departments because they didn’t think that they were necessary.

And then there was this sort of outrage of why are we selling anything?

Well, there was nobody

to figure out what had happened. And we were just stuck there. I sat in six months worth of meetings, talking about Google Analytics goals, how to set them up, what to set up, what are we measuring, because there was no plan. And unfortunately, this is very common with a lot of companies today. They don’t know what to measure. And so they get stuck at this bar. And well, they don’t know what happened. That said, when you do have that foundation really solid and squared away, and you’re measuring the right things, oops, spoiler, you can move on to the diagnostic, why did it happen? This is your qualitative data. This is your customer research surveys. This is the feedback forms. This is where people are really explaining to you why they made the decisions that they made. And this is so valuable, but you can’t get there until you know what happened. So once you know what happened and why it happens, and those are often bundled together, then you can move on to predictive analytics, which is what will happen, what is likely to happen. And then that is bundled with prescriptive is what should we do about it? Because it’s great to know what happened and what might happen. But then you need to come up with a plan of action of what are we going to do about it? And a lot so we put this on here, just to sort of I mentioned, the the fifth wrong in the hierarchy is proactive. And this is where machine learning and deep learning come in really strong. So we’re a few years away from this, maybe five, maybe 10. According to Chris, we’re already there. And this is when you can take all of those wrongs, the descriptive, the diagnostic, the predictive and prescriptive, feeding into a machine. And then literally to show it to work it hands you here’s your plan based on all of the data that you’ve input. We’re not there, a lot of this is still vendor, and manually. A lot of it is plugging in math equations.

But at some point that will happen. And Chris will talk more about that later. So what is predictive analytics?

So predictive analytics is now it’s statistics. It’s an algorithm that helps you understand what is the most important thing, and when is it likely to happen? predictive analytics has been around as long as people have been around, if you think of the old, Nostradamus trying to predict the end of the world. But as I’m, again, coming, he was using data that he had available. And that might have been the astrological data, or the weather data, or things that he had known to happen with humanity to say this is when the world was likely to end. Unfortunately, for us, we’ve lived through over again, probably about a dozen times now that fiction was pretty well, he didn’t upgrade his algorithms. But I mean, how many people today check the weather or check their horoscope? That’s all predictive analytics.

But yeah, if you go back to that number, that 43.5%, you’re more willing to check your horoscope, which is based on the astrological like where the stars are, but you’re not willing to use your own data to make decisions of where you’re spending your money. So that’s the problem.

So within marketing, there are two main types of predictive analytics, you have driver analytics, which is the what is the most important driver analytics in a nutshell basically takes a large series of variables. If you think about if you’re running Facebook ads, or you’re running anything on social media, when you export the data from those systems, you’re likely to get like this 14 tab unwieldy report when data is all over the place, and you’re not really sure where you should be looking. And so driver analysis helps you understand what are the most important variables that you should be looking at, within a whole set of numbers. And so if you want to understand, you know, what’s driving engagement, or what am I most likely to get more followers than a driver analysis is the type that you would use. The other one, which is a little bit more common, and a little bit easier to understand, because time is your Isabel Asus. And that’s really when something is likely to happen. And that’s, you know, your wedding for it has or, you know, when someone is likely to follow you based on the type of content that you’re posting

it back to this whole

responsibility thing.

We need to talk

about the data set itself, because you can’t do these predictions

without having data. So we need to spend just a little bit of time talking about what makes good data, and the types of data that are available to you right now to use.

Somebody used car salesman right now.

So there are 66

sees easy for me to say updated quality, you have clean data, complete data, comprehensive data, chosen data, credible, incalculable

data, and data quality. I mean, as Rick mentioned, it’s like high stakes, no mistakes. Yeah, I take that very seriously. I take data very seriously. Because I want to make sure that whatever we’re putting out there is accurate. So there are certain types of data you can’t use, which we’ll get into that. You need a plan, you need to know what you’re doing, you need to understand what is the question you’re trying to answer? Why are you doing this predictive analysis? Because it’s not a small undertaking to do what type of analysis like this, she don’t just want to like be grabbing at straws trying to understand Okay, maybe it’s this me business help formulate that question, first hypothesis, because this all very much follows the scientific method. So what is the question that you’re trying to answer?

Once you know what that is

everything else that you do after that should tie back into what’s the question you’re trying to answer.

So then you need to get the data to extract the data from where it lives. Now, interestingly, insights, we’re big on using data to do to make decisions, but we’ve only been around for about a year. So we actually don’t have enough historical data, or for ourselves in order to do these predictions. So we have to start to look at outside third party data sets to help us start to inform, okay, what should our decision starts to look like? So we’ve been using a lot of Google Trends data, if you’re not using Google Trends, as a marketer, you’re definitely missing out something really special Google Trends has been collecting data on search since 2004.

Yes,

I said

it wrong the first time, and I was I didn’t want to have it.

And so you can get a lot of really good consumer behavioral data. from Google Trends, we like to attract about five years worth of data because it gives us a nice broad data set. And you can do it by search term by topic. You can extract YouTube search data, you can do e commerce, shopping data. There’s a lot of different kinds. I’m very much in love with Google Trends. But you can also use other publicly available data sources, such as data, taco

statistics, the Bureau of Labor Statistics,

a lot of healthcare data gets a little bit older, but we can definitely do the job.

But if you do have good data, if you do have clean data, pristine data, data that you have collected consistently, and you can use your own data, and this includes Google Analytics, your CRM, your financials, anything that you’ve collected consistently, and that you feel like it’s not, it doesn’t have a lot of those anomalies in it, which you need to clean your data set. So even as you’re extracting pastoral data sets, you need to just make sure that it’s the right kind of data set. To answer the question, you can’t just use anything to answer any questions, there really needs to be some thought put into

it.

I am, do it this way person.

And then you identify which variables to predict. So this is your driver analysis. So as you have that large data set with all of the hundreds of millions of variables, that’s where running a driver analysis can really help you start to narrow down on what is the most important thing to your audience. So if your question is what is getting us the most engagement, you might be surprised to find out that it’s not time of day of posting, or the length of the post, maybe it’s sentiment around it. And so this is where you might need to do a little bit of feature engineering with your data set. And basically, feature engineering is just adding more additional variables to your data set to make it a bit more granular. So again, if you think back to that data that you can extract from social media is probably telling you time of day, well, time of day is great, but maybe you want to know what month of the year, or what week of the year, or what day of the week. And so you can start to engineer those additional variables into your data set to really get a more granular approach to figuring out what is the most important thing to your customers.

You know, Twitter, for example, you’re restricted to a certain number of characters, maybe you want to know is a 10 character post, or 140 characters or the number of hashtags, or you know how long the link is like all of those things matter to people because there’s such a short attention span. And so including a bit of that feature engineering into your data set, before you run the analysis is really going to help you dig down into what’s the most important.

So once you know that important variable, then you can run a time series analysis to figure out when it’s going to be the most important to people. So you create the prediction. And then you build a plan of action. And this is where you get to really reach your audience and understand, okay, these are the things that are most important. These are the times when the things are most important, what are we as a company going to do about it, because they have just given us all of this information to say, this is what matters to me the most.

So

I’m going to go through now now that we’ve sort of gotten all of that housekeeping on the way it was like good data, what you can use, we can now sort of talk about some of the actual practical examples of driver analysis and time series analysis. So driver analysis, some of the big examples, and these are not the only examples. But these are the ones that in social media really sort of making more sense. You have attribution analysis, obviously, social media, and then you have customer reviews. So that’s your vision analysis. What it really means is, what are the combination of things that are working to drive conversions. So Google Analytics, for example, has an out of the box attribution analysis feature, but it’s not great, because it doesn’t really get to that level of detail. That’s really I help you make decisions. And so we like to use something called a Markov chain. And a Markov chain model is essentially like a game of Jenga with Jenga, you know, if you pull one piece out, everything comes top of the ground. And so Markov chain model is very much the same way where you need all the pieces working in conjunction together in order to not topple everything to the ground. So in this example, what we saw with one of our clients was, they were spending a lot of money on Facebook. Now, again, you can get some level of detail out of Google Analytics, but it’s not going to it’s going to tell you social, but it’s not going to tell you specifically the platforms within its attribution model. So this client was spending a lot of money, you know, a few thousand dollars 10s of thousand dollars a month on Facebook as a platform to drive conversions. And once we did the Markov chain attribution model, what we found was that it was Twitter and LinkedIn that was really driving conversions for them. So they were able to start to pivot their strategy and say, okay, Facebook’s important, but we’re ignoring these other social channels over here that organically are just kind of doing their own thing. So imagine how much better it would be. And we actually started to pay attention to it and do more.

And so with social media, once you know what channels are working for you what platforms then you can do the social media driver analysis, which we talked about a little bit, but in this example, we had a company that had multiple social media accounts, their original company, and they had a bunch of regions like hundreds of regions all over the United States. And they just they couldn’t nail down what was working, what was driving

growth of their accounts.

And so I’m surprisingly, it was the size of the audience at hosting. So the interesting thing about a lot of this predictive analysis is that it’s not necessarily telling you something you don’t know. It’s validating your assumptions. It’s making sure that you’re not just making decisions based on anecdotal data is saying, Okay, I think I know what the outcomes likely to be. Let me just double check, run analysis. Okay, great. Now can back up my decision, the data, so if someone comes and questions that you could say, this is what the data is saying. So this is the action that

I’m going to be taking.

And then this was a really interesting case study that we did. So we actually work with a very large staffing company, it was a trucking company. And what they wanted to know, was what is driving those top rated reviews on the review sites? I believe it was Glassdoor. And they wanted to know, what are people saying? what keywords? Are they using what is driving people to leave those top rave reviews, versus the lower number of the lower star reviews or something. And so unsurprisingly, again, we could have guessed this, but by running the data, we know for sure, people care the most about pay, and people care the most about the culture. And in a trucking company culture is an interesting thing that they may not have thought about, because each one is essentially an independent contractor. So they need to make sure that they’re providing the type of culture that someone who works at a trucking company would want, which would then allow them to leave that for Star Media. So you can sort of see how they were trying to figure out what is driving this thing. And then we were able to sort of help them start to understand this is what’s driving those top reviews.

So in a nutshell, that’s driver analysis is really helping you understand what is the most important thing to your audience? What do they care about the most that is helping you reach your goal. If your goal is those top reviews, if your goal is more revenue, more followers, this is the type of analysis that you want to run in order to understand what to do more of and what to do less of

an hour time series analysis. Once you sort of know that, when you compare the timing, the thing that I like about a time series analysis is you can run the analysis once and use it multiple different ways. So it’s very scalable. If you’re a smaller shop, and you’re looking to do something a little bit more data driven. So a time series, if you think about SEO, content creation, again, social media and advertising.

So this calendar, here is the exact calendar that we use a trusted site. So we did our own keyword analysis we are we wanted to know

when people would be looking forward searching for the topics that we care about that we want to be known for, and that they have questions on. So we were able to pull this together using Google Trends data to forecast out

52 weeks when these particular topics would be searched the most. And so in the SEO and content creation context, it’s when you’re creating the content for the things that people care the most about. So in September, for example, people in our audience care the most about Google Data Studio. Okay, great. Well, there’s only July. So we need to start creating more content, answering those questions, so that when they’re searching, defining us, or if we have existing data, then we are optimized using it and making sure that they’re able to find that data when September rolls around. And people are back from vacation. They’re trying to put together their end of quarter end of year dashboards and

reporting.

So let us be the company let us be the resource that they find to answer those questions. So we created quite a few different types of content, video podcasts, blogs, social media, and made sure that it was available by the time that people are going to be searching for it.

So speaking of social media, a calendar like this helps you understand the timing of posts of when to post on social media, there’s so much information going on social media, that it’s really easy to get lost, you can I mean, goodness, you can post 10 posts a day, and someone might only see one of them. So you want to make sure whatever you’re posting is the most valuable in the event that the algorithm is going to show your content and you want to make sure that it’s on topic, and timely. So again, using the same calendar, if you’ve created all of your Google Data Studio content, and you know, the week of September 5 is when you start promote pushing out organically, your webinars about Google Data Studio, your trainings, your blog posts, wherever it is that you’ve created, this is when you start to push those things out so that people are finding you.

And then lastly, as I guess sort of the biggest one is advertising spend. So it takes time and effort to create this content, but that you’re actually going to spend out of pocket money on promoting these things, you want to make sure that you’re getting the timing right. And so again, using the Google Data Studio, if you have the content you’ve been posted, it will now you can promote it, because you know, people are looking for it. And then they’re not necessarily looking for something like business intelligence. So you can sort of wind down those campaigns and spend a bit less on those during the time that other topics are more important to them.

To break

it down even more simply, we’ve created basically the four p framework and this is something that you have, you’re a manager, you can hand up to your you know, whoever your report, so you can hand it to your clients, you can hand it to the person who’s actually executing, to sort of say, this is the plan that you need to follow. And then you have something that’s trackable. And so programmatic marketing, for example, the first P is to plan for weeks out, do we have any content on programmatic marketing? No, okay, we need to start to delegate that out and you start to outline your research, do we know what it is? Is this something we want to be known for. So that’s what you do for weeks out? Three weeks out is when you start to prep, that’s when you’re creating it, you’re refining it, you’re making sure that you have all the questions answered that someone might have about programmatic marketing, so that they’re finding you in a couple of weeks when they’re looking for this thing. Two weeks out your publishing it pretty straightforward, you making sure it’s going across all of your different social handles, you’re making sure that it’s going on your website so that the Google search engine can start to index it, you’re pushing your videos to YouTube to whatever platform you’re using, you’re making sure that your podcasts are going live, and that their SEO optimized with your keywords. And then lastly, the week that it’s the most important, you’re promoting it. So you’re putting the money behind it, you’re making sure that you absolutely can reach the people who care about the most who are looking for it.

So this messy graph actually tells you a lot. This is something that we actually did as an experiment last year,

what when, in the next 52 weeks, will people be reading their email the least, or the most. And so what we did was, again, we use Google Search trend data, in order to understand when people were checking email. And when people weren’t, you can use your own email marketing data, if you haven’t a robust enough data set for this, but we didn’t. So we try to using Google Trends, we what we actually found was, on surprisingly, the week of July 4, and the week of Christmas was on people reading their email the least. And this data was able to confirm our assumptions. So what we did, there’s two different ways you can approach it, you can do nothing, you can send no email at all. Or you can send a scale back email, knowing that people aren’t necessarily

reading it, but you still

want to stay relevant to them. And that was the tactic that we chose, we send what we actually called in the subject line, worst performing email of the year. So we actually got really good reaction, because people were there were still people checking their email, it was able to rise above the noise of all of the other clutter that they’re getting in their inbox during the week of July 4. But we scaled back the amount of content that we were sending, it was our weekly newsletter. And so rather than putting all of the effort that we normally put into the newsletter, we took about a third of the content and send that out, just sort of sent it out the highlights, knowing that was people were going to be receiving it that we get was actually one of our better performing newsletters at the time. So what does this messy, messy graph look like to you, you’re probably saying, hey, it would be great if I had data.

Well, guess what? We’re going to give it to you Yay.

So if you want to take a picture on the next slide, we actually break down the dates for you. Over the next three quarters, the best time to send email, and the worst time to send email.

So the best weeks, q2 is April 21, which is coming up pretty quickly. So start getting those campaigns together. He’s three is the week of September 15. And q4 is the week of October 21. The worst weeks just on you know by quarter is in q2, it’s June 23. q3 is July 28. And q4 is December 23. On surprisingly, people are going on vacation. They don’t want to hear from you.

Thank you. So predictive analytics.

It’s just math. It’s not super scary. It’s accessible. It’s something you can get started with right away in your daily routine is something you can get your team using to make better decisions, more data driven decisions to reach your audience in a more effective way. So you’re not just you know, putting your message out to anybody who might be listening but to the right people at the right time.


Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, Data in the Headlights. Subscribe now for free; new issues every Wednesday!

Click here to subscribe now »

Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new 10-minute or less episodes every week.

Pin It on Pinterest

Share This