Ever feel like automating your marketing takes more time than it saves? Machine learning should free you from manual tasks, not add to the list.
Will Scully-Power, CEO of customer experience automation platform PASCAL51 argues there’s one very simple reason why you aren’t making any progress towards full automation – because most machines aren’t actually learning anything.
“All of these platforms that exist today, there’s no machine learning,” Will says.
“And if it is, their machine learning is, ‘Did someone open an email? Yes, give them a point.’ Machine learning. That’s not real machine learning.”
Which is a problem, because effective automation is fast becoming a prerequisite of effective marketing. From chatbots to real-time contextual geographic marketing, modern marketing solutions demand insight-driven automation to deploy the right message quickly, at scale.
Will Scully-Power anticipates marketing automation will eventually free marketers from manual work which comprises ‘98% of their eight hours a day’, empowering them to spend their time more productively tackling the creative jobs that machines aren’t well suited to.
Will outlines three key problems AI providers need to solve:
1. The creation of effective, scalable machine learning which can optimise a campaign without human input.
2. Ensuring that decision-making system’s logic is transparent and easily comprehensible by marketers seeking to analyze and augment those automated insights.
3. Designing a prescriptive system which can not only predict future actions – but understand why the user would make those actions.
On this week’s episode of The CMO Show, Mark and Nicole explore the truth behind machine learning, scalable marketing insights and why ‘drag and drop’ usually doesn’t mean anything of the sort.
- How to Use Machine Learning to Navigate the Big Data Deluge
- How Machine Learning Will Be Used For Marketing In 2017
- Machine learning is marketing’s future
The CMO Show production team
Got an idea for an upcoming episode or want to be a guest on The CMO Show? We’d love to hear from you: firstname.lastname@example.org.
MJ: Thanks for joining us on the CMO Show. My name is Mark Jones.
NM: I’m Nicole Manktelow.
MJ: And our very special guest today is Will Scully-Power. He’s the CEO at Pascal51. Thanks for joining us, Will.
WSP: Thanks, guys.
NM: We’re so excited to have you here today.
WSP: Great to be here.
MJ: You are a really interesting guy. A serial entrepreneur, if that’s not an unfair way of describing you, because we can, I guess, go into all the different aspects of your career maybe during the conversation, but I’m particularly interested in this idea of customer experience automation.
WSP: Yeah. So, you know, my whole career has really been around tech and specifically automation, both on the marketing side and also the CRM side. You know, when I set out in my career about 15 years ago, it was with building or using technology to automate basic campaigns. So back then, it was, you know, email marketing was new and the industry was sort of just wrapping their heads around digital. We believed there is a big opportunity to go beyond just email marketing, but actually start to measure the impact.
What a novel idea.
MJ: It’s only the nirvana that marketers have been dreaming about.
WSP: That’s right. That’s right.
WSP: The fundamental problem with all of this is that all of the technology that’s available in the market today require marketers to build linear-based customer journeys. When we all know and agree violently that that’s not how customers behave…
MJ: I would actually just moderate that to say many marketers are still trying to maintain the silo because it’s simpler and easier, the linear approach, right? It’s the classic funnel. The number of times I keep hearing it, even though we all agree, but it’s like, “How else do I do my budgets?” It’s this resistance for complexity. “I want to just keep it simple despite reality.”
WSP: Absolutely, and one of the things for us was, how do we help solve this? We need to actually have a global step change in the world of automation, and when we started to look at some of the technologies that were up and coming and also market timing, we’ve been very specific to the market timing for this, is look at what we could build that could literally be that next step change.
What we simply mean by that is we’re ultimately solving the top four problems in automation for marketers.
That is, first and foremost, is bringing together both structured and unstructured data into one platform. So traditionally in today’s world, in today’s organisations, they’re very point solutions. In the old days, it was called a data warehouse and a few years ago, they were called data lakes. In today’s world, they’re called customer data platforms. It’s all the same thing. It’s just a big data warehouse for structured and unstructured data. The first thing is what we’re doing is bringing all that data together in one location. The second thing, which is what we’re doing, is we’re helping marketers learn how their customers are behaving at specific customer lifecycle stages.
So what does that mean? That means that we’re listening to both historical and real-time data around how customers are behaving throughout their customer life cycle. Now it sounds like a pretty simple, novice idea and that everyone’s doing it, but they’re not. All of these platforms that exist today, there is no machine learning. And if it is, their machine learning is, “Did someone open an email? Yes, give them a point.” Machine learning. That’s not real machine learning.
NM: Are you serious? Is this an open rate disguised as machine learning?
WSP: That’s what the traditional industry has been.
WSP: Everyone’s tagging this term machine learning to everything, but when you look under the covers, it’s fluff.
MJ: Well, it’s just rules. It’s just setting up, you know, activity rules based on some kind of action taken by somebody, right?
WSP: If I was to say, “Okay. Telstra, you have a whole bunch of pre-set business rules to determine that if this, then that, send that message,” show me where in that platform it’s produced learning and insight that you can then action. Doesn’t happen.
MJ: How do you iterate? How do you refine the system?
WSP: A hundred percent. We’re using what they call evolutionary algorithms. This is based on Darwin’s theory of evolution, meaning survival of the fittest. What does all that mean? That means that campaigns that are running at specific customer lifecycle stages, it’s using millions, if not hundreds of millions of different permutations to decide which different variables are engaging customers and different segments for different products at different times, and then it optimises that piece of content and campaign and channel in real time.
NM: Is this the point where something will say, “People who search for this also search for something like that.” Then over the course of other customers being at that same point, the system will then start to offer similar things to them?
WSP: Correct, so it’s learning, right? It’s learning in real time and it doesn’t take a human analyst to pull out data, put it into a predictive analytic software and then make an [insight]. By the time they’ve done all that today, it’s too late. The customer’s already purchased from potentially another competitor.
NM: This would seem like a pretty obvious recipe for an activity for a piece of software to do.
WSP: It’s that, but the reality is in the industry, they’ve all been point solutions. There hasn’t been one platform that’s provided all of this. That’s the second problem that we’re solving is how do we understand how customers are behaving and optimise the experience within that customer lifecycle stage.
MJ: Is the big picture planned that you would dump all the different martech solutions that you’re using and use this, or that you plug in? Because I was at MarTech in San Francisco maybe two months ago now, and everybody’s a platform, right? That’s problem number one. Problem number two is everybody’s got this patchwork quilt of 30, 40 solutions that they’re all tying together. Presumably some of those are working. Some of them aren’t, but they’re not going to dump them all en masse, are they?
WSP: A hundred percent, and I’ll come to that because that’s our last part of the problem.
MJ: I’m jumping ahead, am I?
WSP: Yes. That’s the fourth area that we’re solving. The first is bringing the data together. The second is machine learning based on customer life cycle stages, so that campaigns and content can be fully optimised. The third stage is what I’m most excited about, which is automated insights. Over the history of my career, the term insights has been a very loose term. Traditionally, from my time at M&C, the service that we provide was campaign reporting with a little bit of insight. Human-led insight. Now, firstly, that’s not scalable. Secondly, what’s to say that my insight is the right insight? And what’s to say, next, that the recommendation that I’m providing to the customer is the right one? There’s no data attached to it other than a hypothesis or an assumption.
So what we’re doing is we’re building a series of algorithms that will build narration around how customers are not only, you know, not only what they did, not only what they’re likely to do next, but why they’re likely to do something next. It goes from what they call descriptive analytics to predictive analytics to prescriptive analytics, which is the last component, which is, “What should I do next?” That’s where you start getting into the traditional martech stacks, which is send an email, send an SMS, push a display ad, push this social.
For us, we recognise that in the enterprise space, the likes of Salesforce, Oracle, Microsoft, et cetera, have big footprints in a lot of these organisations. Our goal is not to dislodge them or replace them. Our goal is to partner with them to connect our platform so that if the customer is using Oracle or Salesforce, for example, we can just point to it to send what we call the last mile, which is the message. Right? And why that makes sense is because all of the martech stacks make money. Their business models are built around messaging.
Now in today’s world, the manual linear-based campaign logic requires, obviously, humans. Now the vendors are only making money if the humans are building lots of campaigns. What we’re doing, because we’re optimising millions of different variables in real time, there’s going to be a lot of different, more relevant messaging happening. From the partnership with the vendors, they’re excited because they see it as a way to actually make their software more sticky, but also drive more revenue.
MJ: In other words, more messaging.
WSP: Yes. Well, sorry. More relevant messaging. More relevant messaging. So when we talk about, you know, historically, people in the industry were like, “It’s all about less messaging.” It’s actually, if you think about it… There’s more and more ways to engage these days. You’ve got, from a digital perspective anyway… It’s not just about email and SMS. There’s things like live chats, chat bots, all sorts of stuff.
NM: Could be the message you’re seeing at the ATM when you’re getting your cash out, right.
WSP: A hundred percent. A hundred percent.
NM: I guess there’s an opportunity then for more analytics as well… behaviour at those points. If you are now able to track what people are doing at that point and put that information back in, then you can start seeing whether you’re being effective at those points as well, instead of always measuring like we do in online content for many years, and many still do. We just measure the consumption and whether it led to a sale. We’re not really checking further up where something might have been discovered, and people went looking and that sort of thing.
WSP: Yeah, absolutely, and I think this is the real opportunity, right? Is to deliver real true automated insights so that the end customer that’s using a platform like this, they actually start to learn more about their customers, and to your point earlier, Mark, the team actually spends more of their time coming up with customer experience innovation rather than focusing on customer experience automations, which is what we’re really trying to do.
MJ: This whole design thinking ethos that’s been prevalent for years now has been all about how well do we understand the customer. The conversation has actually been humans talking to humans. What you’re saying is, “Let’s get the humans out of the way, talking to humans, but the machines talking to the humans.” Is that right?
WSP: Yes, there is the context of the… I’ll explain it in context to automated insights. Just because the machine has generated automated insights, that doesn’t mean that there’s no place for human context. As part of the platform, we’re giving the customer the ability to add their own flavouring to the insight. We give the baseline insight, and what’s likely to happen next and the recommendation around what to do about it, but for the end user, they will be able to add their own context, enrich it, and add any different external attributes that potentially is not being read or is not a piece of data. It’s what we call a combination of machines and humans, right?
Because we still believe there is a huge amount of value and context that humans can provide, but what we’re trying to do is remove all of the manual stuff. If you asked any digital marketer, “How do you spend your eight hours per day, or 10 hours per day?” And 98% of it is manual work, which can be replaced, and allow them to have a better, more enjoyable time at work doing stuff they were probably initially hired for in the first place.
MJ: You’re taking away the manual stuff, the boring things, right.
NM: So how does this fit in with your prescriptive analytics thing?
WSP: So the prescriptive analytics part is just about what to do next. Traditionally, when a campaign was run, they would get the response data, extract it out, use some BI or business intelligence software to determine what happened, and then maybe they would use some predictive analytics tool to determine what’s likely to happen next. Is Mark any more likely than yourself to take up an offer for a home loan, for example? A zero or one. That’s the level of sophistication. Then they would take that list of names, stick them into the campaign tool, and then send them the next message.
And that period of time, it was quite a long period of time between when the campaign actually ran and when the next message would be sent, so what we’re doing around the prescriptive analytics piece is actually automating the decisioning in real time. So that is really around the next best action, the next best offer, and this sort of language has been talked about for years, right? When they talk about it, they talk about next best action into next best conversation. It all means the same thing, which is, “What is the next, you know, ultimately piece of engagement that we need to have with either this customer or this prospect, and what channel should we have it in and what time should we have it with them, et cetera?”
Historically, a lot of that has not been fully automated. It’s been what we call semi-automated. Certainly not in real time. It may be in things like batch updates. Everyone who bought a product from us online yesterday, let’s get a file of those customers and then do some analysis overnight and then determine what message we’re going to send to them… a follow up message tomorrow. What we’re doing here is just automating that whole process so the decisioning happens in real time.
MJ: Tell me about the tech. What’s the secret sauce? How do you make this whole AI thing work?
WSP: Yeah, so it’s a really good question. If we were to start this three years ago, we just raised our seed round, which was 250K to get us to build an MVP over the next 12 months… six to 12 months… if we were to introduce this…
MJ: Just for those playing at home, minimum viable product, right?
WSP: Yes, minimum viable product. We’re calling it the MLP, the minimum lovable product.
NM: This is not a prototype. This is a working…
WSP: This is going to be a working product. Yes. If we were to do this three years ago, it would’ve cost a lot more than 250K. What we’re doing is we’re using a number of different open source technologies to ingest extremely large volumes of data from point of sale data to transactional data to customer data, et cetera, and then we’re processing this data in real time. Now to process this data in real time, it has to go through all your models, and then it needs to crunch this stuff almost instantly.
WSP: Now’s the actual reality.
MJ: Well, look. You’ve been around engineers and software developers for the longest time, so I’m sure you know how to bring it all together. It’s really quite fascinating. We talk a lot about AI and machine learning and where it’s all going. It’s fascinating to hear it being applied in a marketing automation context. I think it’s probably one of the biggest areas… challenges, right, that marketers are facing?
MJ: I wondered, just, you know, we’re beginning to round this out. The CMO/CFO conversation. Are you actually appealing more to the bean counters than the marketers here?
WSP: Yeah, it’s a really good question. In fact, yesterday I was building out the user stories, and in our user stories, we started with retail just because that’s where we started our journey, and in the user stories is… I’ve got a user story for the CEO, the CMO, the head of Digital, the head of E-Commerce and the head of Customer Service and Support. Why? Because customer experience automation touches marketing sales, service, customer community, et cetera, and they all can get value from it, because they all have different problems and pains. They all have different primary goals. They all have different KPIs.
We are actually building the platform to adhere to all four or five of those different roles, so that when we walk into… whether it’s the head of digital versus the CEO, we know what buttons they’re looking to get pushed in terms of what are the key drivers for them with a platform like this. Having said that, what we’re also going to be is AI for every business. True AI. What we mean by that is, whether you’re a small business with a couple of employees and you’re just looking to get some basic customer journey automated without having to log into complex software and build creative and do complex testing, we want to give that same power that we’re providing enterprise to small and medium-sized businesses all around the world in multiple languages, et cetera.
MJ: That’s been one of the great divides. I’ve actually found that in start-up land where you can always tell the start-up that’s targeting the enterprise because it’s, “Call us for a demo and a trial, and then we’ll have this lengthy onboarding process,” as opposed to “Free trial. Stick in a credit card. Off you go.” Huge psychological difference between the two.
WSP: True, and it’s one of the… Actually in our early… part of our MVP is where we’re looking to validate somewhat about the assumptions that we have around both of those, and I think post-MVP, we’ll have a pretty clear position on that, but ultimately our whole company’s built on three core values. The first is simplicity. We want to make… Every automation vendor talks about, “How simple is this tool? Drag and drop.” That’s absolutely garbage, because once you lift the hood, it’s not just drag and drop.
I was sitting with a CMO the other day, and he said, “Oh, I’ve got Mary here. She does all of our marketing automation.” I said, “Really? Mary, you’re a superwoman, because you need to be a technical architect, a solution architect, a campaign manager, a data analyst, a campaign analyst, project manager, and the list goes on.” I said it’s about eight resources, and to build that in Australia, that’s about 800 to a million bucks in salaries alone. That’s why 80-odd percent of digital transformation projects… specific automations fail.
First and foremost, simplicity. Simplicity in the provisioning of the product for your business, but also in the use of the application. The second is trust and transparency. Every single vendor that’s out there don’t have a dashboard that they show you that shows the ROI of your tool. We’re going to deliver that. So it says, “Day one since you’ve been using Pascal51, this is the revenue impact and the ROI that we delivered to your business.” So trust and transparency there.
And the other part of the transparency is around what our models do. Every software company in the past that’s talked about a black box, most of them have failed. Why? Because customers can’t get in and look around, so why is Mark getting that message versus Will getting that message? What is the different attributes that is deciding that? We talk about, instead of the black box, the glass box. Right, so giving the customer the ability to see exactly why Mark was sent that message at that channel at that time.
And then the third core value really is around that value piece. Making sure that the customers that are using this thing, not only they’re getting value, but their end customer is getting value, so it’s part of one of the things that we’re doing is getting a metric back for their customers that we’re going to use as part of our KPIs with our customers.
MJ: Will Scully-Power, thank you very much for joining us. It’s been great to have you on the show. Really appreciate your insights, and all the best with making a new amazing, interesting idea reality.
WSP: Thanks, guys. Thanks for the time.