Using marketing signals to accurately predict brand health:
Brian Mossop and Methods+Mastery

Topics

Senior Vice President and Director of Strategic Intelligence, Brian Mossop, joins NewsWhip’s CEO, Paul Quigley, to discuss some of the ways that prediction can be used in advanced brand reputation forecasting.

Guests

Although Brian Mossop has a background in biomedical engineering and neuroscience, Brian currently leads a marketing and communications intelligence team to supply data-driven insights to some of the world’s leading brands. You may have also spotted Brian’s work as a journalist writing for Wired, The New Yorker, or Scientific American. Follow Brian on LinkedIn.

Talking Points

• About Methods+Mastery | Jump to text
• Measuring brand reputation | Jump to text
• Real-time brand health metrics | Jump to text
• Identifying predictors of brand health | Jump to text
• The importance of data for clients | Jump to text
• Incorporating emergent data | Jump to text
• Sharing predictive insights with clients | Jump to text
• Building credibility with clients | Jump to text
• Changing communications strategy | Jump to text
• The longevity of brand predictions | Jump to text
• Seasonality of brand health | Jump to text
• Sentiment in the brand health equation | Jump to text

Paul Quigley: [Starts at 1:39] Okay, Brian, let’s get right to it. Maybe we set the stage first, if you can tell us a bit about Methods+Mastery, which is within the FleishmanHillard global network, but an independent agency. Can you tell us about why it was created and what you do?

Brian Mossop: Yeah, absolutely. You’re exactly right. Methods+Mastery is part of the FleishmanHillard network. We were created to kind of handle some of the business that FleishmanHillard couldn’t. Over the years we’ve really developed our focus into marketing analytics and marketing social strategy and things like that. The creative elements that you mentioned, and that’s our world. We’re comprised of roughly 20+ analysts. So really, intelligence forms the basis for everything that we do. I know a lot of agencies say that, but that’s a real commitment of ours and part of the north star that we align to.

Paul: And does that mean that you were like a small agency or a part of a big agency? Or how does that interact and being kind of a specialist super analytical unit, but within the wider network?

Brian: Yeah, I mean, at its very basic level, we just work on different clients, on different kinds of goals for those clients. So, to that extent we are independent, but at the same time, we always are sharing methodologies and findings and learnings kind of across the network as we can. Because there’s always great work being done on both sides. And so there’s always opportunities for us to kind of come together and really put the best of both of our worlds forward for the clients we work with.

Paul: Right. One of the places where you’re doing, I think, some really innovative work is around brand health and brand reputation. And that was definitely where we’re thinking about today’s Pulse, we wanted to focus. Can you tell us a bit about how you’re thinking about measuring brand reputation today, metrics that you’re paying attention to?

Brian: Yeah, absolutely. I mean, brand health is top of mind for a lot of comms and marketing professionals right now, both on an agency side and on client side. What we’re particularly interested in, again, being kind of firmly entrenched in the marketing intelligence and marketing analytics world is figuring out how these different kind of marketing signals, and there’s thousands of them that exist, we want to understand how each of those kinds of pushes and pulls on the lever, so to speak, on brand health scores. This is kind of brand health is really about perception of a brand to some extent and the reputation of that brand amongst its followers and its fans. So, it’s an interesting space for us because the way I like to think about it, a lot of marketing efforts that look at last click attribution, they’re trying to come in and tie a specific post to a single sale.

And there’s a lot of value in that, but our kind of belief is when you’re looking at something a little bit broader than that. So, you get into the brand health arena, there’s so many different aspects that can kind of contribute to that, but if you get it right, whereas attribution is going to get you a single sale from a post. If you get the brand health and brand perception and really drive that point home with consumers, you’re going to enable a lifetime of sales, because the people come and become fans of the brand and loyal to that brand. And so that’s why we think it’s really important, especially for marketing firms and marketing organizations within larger companies, really need to be focused on this. Because there are tangible and actionable things that can come out of this. Once we understand how this is all connected and how it all ties to ultimately to brand health and reputation.

Brian: Yeah, absolutely. I mean, I think, no matter which way you get there, brand health is usually assessed by some sort of survey. It gets to consumer perception and understanding of and expectations of what that brand should do. But fortunately, there’s a lot of tools out there and technology vendors that are doing surveys at scale for across different brands. And so they’re syndicating this survey research to people like us.

So, it’s kind of taking the onus off of us to conduct that primary research to get those end points that we’re looking at, because there’s all this syndicated survey research that’s out there. And that’s an amazing thing for us, because it’s a lot of information about a number of brands and they track these things weekly. It’s really hard to get that sort of granular look at brand health for any company if you’re doing it yourself. You kind of would have to do this repeatedly week after week to really get the same amount of data. We’re fortunate to be able to have that sort of syndicated data to really build these complicated models around.

Paul: So, if the health is being measured here by the surveys and the syndicated data, you’re also trying to find the signals and the KPIs that are going to feed into and drive that outcome and measure those two, right, Brian?

Brian: That’s right. That’s right. I mean, that’s what this big project that we’re talking about was really about. We wanted to understand, we knew that there’s hundreds, probably thousands of different marketing signals that might affect brand health overall. And so what we wanted to do was build a model where we can look at those inputs as granular as possible. So, if we’re just looking at earned news coverage, we might want to break out that coverage by topic. We may want to break it out by publication. We may even want to go as granular as specific reporters so that we can understand how each of these individual signals might contribute to positive or negative changes in brand health. So, it’s important to go as granular as possible so that you have a really good understanding of what is actually at play here and not some kind of correlation with another variable.

That really would be driving the effect. You want to get as precise as possible, because that’s where the real information gets translated to clients. We want to be able to take this information and say, here are the things that you should really dial up next month. And here’s the things you really need to kind of address to steer the ship in a positive direction. And we do that because, again, we’re looking at these very granular marketing inputs. But that’s very complicated to do, especially for a human to do. How could we possibly kind of assess thousands of signals, look not only for correlations, but look, figure out which of those signals are actually predictive of brand health.

That’s just not possible without machines. So, I feel fortunate that we have invested a lot in bringing people on who are experts in data science and machine learning and data engineering and advanced analytics. And I feel fortunate to work with some of the best in the industry who, I like to say, make my job a lot easier to kind of take great work and show it to clients. That’s what really excites me. So, we’ve made that investment and that has enabled us to create these models that can crunch those numbers in those ways and look for those signals, again, that are not just correlated to brand health scores, but are actually predictive of those scores two to four weeks out.

Paul: So, as you can imagine, we get very excited when we’re talking about prediction, Brian, given our own interest in that. And one of the beauties of good models is we can tell ourselves a story about any one of these 1,000 data points and why it might drive brand health and why it might drive, even answer to those questions a month or six weeks down the line. But in the end, only some of those stories will become true. And often the data con can demonstrate which one will come true. I know that there may be sensitivities with naming clients, but can you walk us through at whatever level of detail you’re comfortable with, an example of how this has worked for one of your clients?

Brian: Sure, absolutely. I mean, we work with a number of Fortune 500 companies in the technology industry and other industries as well. This kind of topic around understanding how marketing is driving brand health. I would say it’s becoming of greater importance to the industry. We’re under tremendous scrutiny to show return on investment for marketing dollars spent. And we just really feel that kind of going after those lofty goals of figuring out how marketing initiatives are, again, not just driving sales, but driving perception of the brand, that’s valuable information for any brand. And so we’ve kind of rolled this out for some specific clients, but this type of model scales across our entire business, we could do this for any client in any industry, as long as there’s enough data there to form the models around. And that’s the biggest challenge that I think that we experience in comms and marketing is the availability of data to kind of build out the models, these predictive models that we’re talking about for brand health.

We needed about two years of historical data, and that’s both on the input side and on the brand health side, on the output side. It’s a lot of data for companies to have access to. So, traditionally that’s done in one of two ways. Either you’ve purchased tools that have kind of the ability to historically crawl content and bring that information in, or I think what you’re starting to see more and more companies do is taking their own approach to data warehousing and things like that, collecting and storing their own data. And I think that’s a good thing. Because that enables, down the line, if these companies do want to build predictive models around things, they have a good base, they have the information that’s needed in order to do that. Because a lot of times it will be when they come in and a company has to say, well, we don’t have two years of X, Y, and Z data.

And then we have to kind of figure out, okay, how do we start collecting these things? How do we get started with what we have? So, I think that’s the opportunity for brands is to really do something with the data, to have a strategy on how they’re going to archive data, how they’re going to make it available, what type of data they’re storing. And that, I think, is going to allow us to translate this into more and more clients. For instance, a lot of these syndicated brand health scores are only for the largest companies, but what if we want to look at brand health for a company that’s not in that index? We might have to rely on their own data.

You mentioned NPS scores and survey data. Maybe they have that. So, it’s complicated. We want to roll this out to as many brands as possible, because we really think that it can help. We’re really seeing actionable insights down to do this action with this specific reporter or dial up your own marketing efforts on these topics, because those are what translates to be predictive of brand health scores. We know that can help a lot of brands, but we just need to think through the data strategy on how we’re going to actually bring that to life.

Paul: Being concrete, you are seeing then that a brand engaging either with particular communities or on particular topics is much more of a driver than 10 other things that could be drivers. And you’re saying, okay, if we want this to move in a month’s time, we need to engage with these communities or on these topics, and even with these publications, even with this journalist.

Brian: Yeah, that’s right. I mean, that’s what this model does. And the first stage of the model is look at these 1,000 inputs and just tell us which ones we need to worry about. What are the five or 10 that are really driving the most. Which are the most predictive of changes in brand health? And so that just uses machines to do what they do best, which is kind of collect and collate and sort data and kind of filter that down. So, now we can bring an analyst in and kind of look at those specific inputs, because we’ve kind of taken it from thousands to maybe five or six or 10. So, it becomes a much more tractable solution in that way. And then of course the models have additional needs, where there are additional functions, where they are forecasting brand health scores in the future.

We kind of compare that to what’s happened in the past. And with very high statistical certainty, we can say, hey, these are the 10 marketing inputs that you really need to focus on. But not only that, when our analysts kind of get a hold of that data, they’re able to be much more prescriptive. And that’s what I was kind of alluding to before was we can say, hey, you need to dial up media relations with these specific outlets or reporters. These are the things you need to do from an own social perspective. These are the things you need to get a hold of from the earned social conversation. Because this is what we’re kind of seeing come in from the community and appear in that conversation.

So, it’s really the best. I think that’s what’s so exciting about it is it’s really bringing the two things together, letting machines do what they do best, which is collect and kind of collate and sort that data, filter it down and then bring in humans to do what they do best, which is draw disparate connections in data. And this is an area where Methods+Mastery has really doubled down bringing these two things together. Because that’s really where you’re delivering the most precise insights to your clients.

Paul: And when I imagine the models at work here, the inputs, the outputs, one of the kind of worries I’d have is the media ecosystem and culture is not a closed system. There’s new issues emerging all the time, there’s new political perspectives, new groups emerging and competition, perceptions are constantly moving in a way that could render a model potentially, or one of these drivers much less relevant, pretty quickly. Or a new one come on to the landscape pretty quickly. How do you deal with that? This is maybe something that’s initially well in polls that are too structured and don’t capture new things through emergent, but this is the [NLP] and big data side of it.

Brian: Yeah. It’s a great question. And I think that’s one of the hardest things in prediction. The predictions are only as good as the data coming into that. And so you need to understand that data coming in. Are there inherent biases in that data? Is it really saying what you think it’s saying? So, it’s difficult. I mean, we really want to kind of take a look at and not bias our own selves into thinking like these are the data sources that we really need to focus on. The model itself, it’s not just about clean data coming in and useful data coming in, but in any mathematical models, there’s assumptions that are made to produce that model. So, we have to get comfortable with uncertainty. And that’s really difficult for us to do in the comms and marketing industry. People who come from kind of technical and statistical backgrounds, this is second nature to them.

They understand that anything that you build computationally, it has assumptions built into it, and those will dictate the results. Those will really drive the accuracy of the results. So, it’s a lot of education in that domain as well, because clients don’t like uncertainty, but we have to kind of talk about it and be comfortable with it. And that’s a skill as well. And that’s a skill that we’re building out as an agency is being able to interpret those results into actionable insights for a client while not overselling it, while not over committing and not saying that the model is something that it isn’t. Making sure that the assumptions are very clear so that we understand how to interpret those results in a meaningful way.

Paul: Talking about turning this into actions, is there a new workflow here? Because you may be recommending quite specific options based on longer term brand health goals. How does that workflow work between Methods+Mastery and Fleishman and clients? How do you bring these insights, which all have a sell-by date, because we’ve developed something that’s got quite powerful insight if we can put it to work now? How does that happen, as we’re changing workflow and relationship with clients that comes with that?

Brian: Yeah. I mean, that’s a really interesting point too. Ultimately we deliver all of our insights to our client. But oftentimes, as we all know, a lot of companies are kind of highly matrixed, marketing is often separate from comms and PR and things like that. So, it’s a challenge when we present kind of these cohesive and complete findings to our client. We also have to figure out, well, how do we get that activated internally for them? How do we take the insights that we’ve derived from earned news coverage and get that in front of the PR and comms team so that they can incorporate that into their media relations strategy. Sometimes even within marketing, like social production is split from social analytics.

And so we have to kind of figure out how we get this really powerful and actual information in the right hands, because ultimately you can only tell clients these are the things that, from a statistical standpoint, we’re fairly certain are driving changes in your brand health. We just need to kind of start having that conversation so that it’s actionable. And we actually do something with the data that we’re driving from the models. Because there is a lot of actionable stuff there. So, it’s tough, but that’s why we have good relationships with our clients and we help them kind of think through these challenges and come up with a strategy on how they might kind of socialize this within their organization as well.

Paul: Yeah, it’s kind of a perennial problem, I would say. Especially people who are the audience here at the Pulse, because you’ve got people who are at the more analytical end of the communications industry and come from more analytical backgrounds and often have developed insights and methods, but it’s a question of how to get the behavior moving around those and it’s different every time. But you said working closely with clients being key, I suppose, is earning trust. Does how you present the information make a big difference as well? Or are there any other tips and tricks from Methods+Mastery for making sure that the insights are put into action?

Brian: Yeah, absolutely. I mean, I think that’s why storytelling is so important, even when you’re talking about analytics, because it’s not just about the numbers. It’s how do you communicate what those numbers mean. And this is an area that Methods+Mastery has really excelled at. Kind of telling the story behind data. I kind of often draw parallels to what we do to kind of what a data journalist might do. How do you explain something really, really complicated in ways that are understandable by the general public? And that’s not meant to be like a disparaging comment at all. It’s just like we come from different worlds. And so storytelling becomes crucial to what we do, but it’s also, I consider that, that’s really fun work for us as well.

Because I think for a lot of us at Methods+Mastery, there’s that point where you see kind of a client’s eyes light up because you tell them something that’s really actionable, understandable from data. And again, that kind of, you were talking about trust, that builds trust as well. We need to kind of make clients comfortable, like I said, with uncertainty, but with quantitative information overall. We need to make them comfortable with that by telling stories on how we might act upon this, what these numbers really mean. That’s another way for us to really build sound trust with our clients, because they know that what we’re doing is accurate and meaningful.

Paul: Yeah. And when you look at the future behind the successes you’re having there, do you think this advanced benchmarking that you’re working on or the predictive benchmarking almost … There’s a better phrase, but you know what I mean, I think. Is that drastically going to change communication strategy and how people work in five years down the line?

Brian: I think it will. I think that communications and marketing is always going to draw from both sides of the spectrum. There’s going to be a need for quantitative information and data and understanding that, but there’s also a place for experience, intuition, and instinct as well. And it’s just figuring out how to make those work together. I don’t like to think of them as kind of competing things. They’re actually, they both come together to make something that’s greater than the sum of their parts. And so that’s the space that we play in.

But I think that we’re seeing overall, we’re seeing our industry getting more and more comfortable with data-driven insights. And so, I mean, I think that there’s kind of this paradigm shift that’s going on in the industry. And I think that’s why at Methods+Mastery, we’re so passionate and bang the drum about the importance of data, because it’s really powerful not to counteract instinct or intuition, but how to inform a strategy where you can bring that experience in and make something really great happen from a creative standpoint, from a social strategy standpoint. I think that’s it for me, it’s showing that these two things are not competing. They can actually be used in tandem to do something really amazing.

Paul: Yep. Yep. We agree with that. And with our kind of real-time predictions, which I think we’re still winning over people as they see these kinds of things come to pass, or as they see the predictions bear out. And what you’re doing like, our software’s predicting 24 hours down the line. When it comes to brand health metrics, you’re looking how far down the line ahead? Essentially, on a piece of coverage on an event affecting brand health. How many weeks out would that be?

Brian: Yeah, our predictions are accurate two to four weeks out, and that’s largely due to that, again, that two years of historical data. The more historical data you have, the further out, in general, you can predict. So, that’s kind of where we’re at now, which is, I think, it’s a good space for clients. I think changing social strategy in under a month’s time is really tough. So, getting kind of monthly insights on what is changing and what’s not, that really right now is kind of serving as the best cadence for us. So, monthly to quarterly, I would say, is really kind of that sweet spot for clients bringing out that information.

Paul: You’re bringing the client’s attention too. And there’s a good bit of thinking and work that goes in every two to four weeks about directionally, how the key metrics and predictions are tracking.

Brian: Yeah. Well, I would say, just to clarify, I mean our predictions are accurate two to four weeks out. We’re delivering recommendations either monthly or quarterly. So, we’re kind of taking all of this in and not only looking for okay, this week, these are the marketing signals you need to worry about. Week two, these are the different set of signals you need to worry about. We’re kind of looking at them together and looking for the trends. And so, I think monthly to quarterly insights are what are what’s going to be the most impactful for clients.

Paul: We’re just getting to our last few minutes already. So, if you’ve got a question please prop it in, Danielle’s already got one that’s come through from the audience. “If brand health is relative, do you weigh brand dimensions differently during slower news cycles?” This is a great question… we just did a piece on vaccines this week about the drop-off in media engagement generally and in particular on political topics, we’re seeing kind of a 27 to 30, even 40% drops off in traffic and social engagement and a lot of the hot political things that people were losing their minds about last year. But just generally people are, I guess, going to the beach after 18 months of being glued to the phone. So, we’re going to see a lot of lines trending downwards. So, how do you account for this new model that doesn’t make it look like interest in the brand is declining, when in fact there’s just less people generating the numbers that are turning into KPIs.

Brian: Yeah. That’s a great question. That’s why we’ve really taken a step to incorporate seasonality in our data or accounting for seasonality and data. Not to say that this is particularly seasonal, but I think, you know what I mean, like this is kind of a cycle of events that might happen. And again, that’s why having two years of historical data, you can see that happen over the months. You can see interest wax and wane. And so, I think that’s why it becomes particularly important in that realm.

Paul: Right. Yeah, the historical data helps a lot there. And when you’re thinking about sentiment to be part of what you’re tracking how is your evaluation of the quality of that today? Do you feel you’re able to do that well, or just need to be a bit of human nudging and oversight to ensure that it’s not a misunderstanding, trolling or sarcasm?

Brian: Yeah. I mean sentiment is tough. We all know that in our industry, especially when you’re talking about news articles. How do you take a 1,000 word article? Is it positive or negative or neutral? Well, in relation to what? And so that’s why I think there’s been a lot of emphasis placed on natural language processing driven sentiment analysis. And so that’s one of the things that we’ve also been working on is we’ve created brand-specific NLP models that are much more accurate than kind of out of the box sentiment scoring capabilities. And so that’s right. Kind of factoring that into the model. You want to make sure it is accurate. So, these kind of custom NLP models help us get there.

Paul: I think eventually we’ll reach a kind of end point of accuracy that’s possible there, because even people frequently misunderstand one another when they’re being sarcastic, especially across cultures, right? So, we won’t be able to get it right every time, I suppose.

Brian: That’s right. That’s right. Yeah. Sarcasm is always a tough one, but the trained NLP models can help. It can help understand that. But it’s a constant challenge for us in the industry.

Paul: Yeah. That will be. And so, Brian, it’s been brilliant to have you on for our conversation today. I’m really fascinated of how you’re able to develop or how you’ve been able to isolate these reputation drivers weeks in advance.

And I would also invite everyone who was here this week. We’re going to be having another very special Pulse in not exactly two weeks. In two weeks and two days, we’re going to have Michael Young, who’s the Global Insights Manager at Ford Motor Company. And they’ve been developing in-house analytics at a very high standard for 15 years. Have had a real background in doing analytics and the intersection of analytics and communications. Going to be a really interesting conversation focused on the Ford F-150, which I’ve been learning a lot about in the last few days. Brian, I hope you will be joining us for that as well. I think you’ll enjoy it.

Brian: It sounds amazing. I’ll be there.

Paul: Brian, Thanks so much. It’s been brilliant to learn about Methods+Mastery. Do you have anything  for people who’d like to learn more should visit your website and reach out to you?

Brian: Yeah, absolutely. I would love to have those conversations.

Paul: Brilliant. Okay. Thanks very much, Brian. And thanks everyone.