Episode Transcript
[00:00:00] Speaker A: Hey, everybody. You're listening to Top Quartile, where we bring you stories from the front lines of growth in community focused financial services.
Well, hello and welcome. I'm Dan Marks, your host, partner, and EVP at Infusion, and we're bringing back our very first guest, Infusion CEO Tim Keith, who helped us kick things off a few weeks ago. Tim, how's it going?
[00:00:23] Speaker B: Good, Dan, good. Thanks for having me back.
[00:00:26] Speaker A: Oh, yeah, always a pleasure. And you know, I'm excited. Since we kicked it off, the podcast is going really well. We're seeing great listing numbers keep going up. And today we're talking about a topic that's, that's pretty hot in the news because of the things that are happening with cookies. And then just there's always a topic around attribution. How do you. How do you know how your marketing's doing? How do you calculate an ROI on that? And so I'm looking forward to the conversation.
[00:00:52] Speaker B: Awesome. Yeah.
[00:00:53] Speaker A: And if anybody's interested in knowing the fascinating fact about Tim, you can go back to the welcome episode and discover that. So, Tim, as we get started, let's kind of think big picture. And we've both been in the industry a long time. We've seen a variety of approaches to what we're talking about. But let's talk about what are overall, what are some of the different methods of attribution that we've run across in all our years?
[00:01:18] Speaker B: Well, I think the subject is fascinating because you end up in this almost a conflict between, on the one hand, the desire to try to strain things out into very specific cause and effect type dynamics. And that is, I think, almost in direct conflict with the concept of the marketing mix, which is a fundamental premise of marketing, is that the sum of the different efforts that we make in marketing produces or the, I guess, the different elements of the marketing mix produces a sum that is greater than the parts.
So when we talk about attribution, if we start with grounding, regrounding ourselves, really, in the concept of the marketing mix, if you think about it from a consumer's perspective, you know, I'm driving down the road, I see your sign, your branch signage, three times or four times a week, depending on my driving patterns. I hear one of your ads on the radio, or I see a billboard, or I see one of your commercials on tv, and then I get a piece in the mail about a home equity line of credit. There's a good rate on it. And I happen to be thinking about doing a home improvement project. The reason we do all Those things is when I look at that home equity offer, I look at it in a broader context of my awareness of you, things I associate with you, with your organization.
And it's really ridiculous to say that the home equity piece is not affected by these other things you're doing or else you have to say why are you doing them? And so it's the combination of those things that should work together to produce again, a sum that is greater than the parts even a customer can't separate in many cases in their mind, you know what, which elements of the mix led them to choose you as a provider for a particular product or service they were in the market for at a point in time. And so if you think about that whole concept that's very fundamental to how you go about attribution, you're never going to get 100% cause and effect relationship between a particular marketing execution and the impact of that.
If you don't start with that premise, you're not going to get anywhere with the whole attribution discussion. If you do start with that premise, then I think you have a chance to make some reasonable assumptions about the value of the investment you've made in particular aspects of the marketing mix.
[00:04:02] Speaker A: Yeah, that's so very well said. I mean, in the past you would talk about kind of a marketing funnel. You start with the top of the funnel, you create awareness and then preference all the way down.
And then more recently, some of the theoretics have pointed out that the path to purchase looks like a squiggly line.
Clients like you talked about, customers sort of go through long periods of time where they don't even have a need.
[00:04:28] Speaker B: Right. And one of our mantras when it comes to loan marketing is proactive awareness building is really critical. So if I get a piece from you, you have a, let's say you have a 199 home equity line, 6 month promo rate or something.
And I'm not in the market for home equity, but that makes an impression that, oh, hey, bank ABC has good rates on loans.
Six months later I've been at home, working at home, looking, getting, you know, frustrated, looking around my house at things I haven't done that I want to, I want to go ahead and do that. The fact that you offer good rates, that, that impression is still somewhere in my mind. And when I start thinking about, well, who would I talk to about a home equity loan or line, you know, that almost activates that prior impression. If, and so again, how do you set up an attribution model to Take that sort of thing into account. And we're going to talk about some practical things as we get deeper in the conversation you can do. But I think it's starting with a realistic understanding of what the role of attribution is, what the limits of attribution, type of measurement, what those limits are.
[00:05:41] Speaker A: Yeah, absolutely. Well said.
And to your point about the mix, I've had multiple experiences in different industries with kind of a full scale econometric marketing mix where, where some very smart mathematicians try to build an econometric model with in some cases hundreds of variables to try to simulate that. And that can be really interesting, really impactful.
The challenge from a practical perspective is often it takes months to do the analysis. And so at that point you're three to six months after whatever you've done has happened and it becomes very challenging to have a practical benefit. So really for this discussion, we're going to focus on some of the more practical approaches with those. So let's kind of walk through Tim and talk about some of the different approaches as we, you know, we start, we're starting very big picture and then we're going to get more specific. But you know, what are, what are some of your experiences or comments around say, a pure activity based approach?
[00:06:44] Speaker B: Well, so I think digital marketing has obviously completely revolutionized marketing and you can just look at the advertising business and see how that's been transformed into a data driven business. But two aspects of that that are two sides of the same coin that relate to attribution are in a targeted environment. And so the first foundational component of measurement of any digital campaign is engagement. How many times was I able to show them the message?
And then how many times did that result in clicks where the customer took the step to engage, find out more information. Of course the engagement itself is beneficial in terms of, you know, creating essentially a communication stream with a customer. But the measurement itself is also a revolution in marketing. So in the old direct mail days, again talking about targeted marketing, you send a piece of mail, we have no idea, no view to how many people read it right away, how many people set it aside and read it later, how many people throw it away and never look at it. You have no view to that. You just kind of send the mail, kind of sit there, wait, and then when an appropriate amount of times passed, hopefully you're able to take an updated data and identify what, what are the people that I mailed, what do they bought? And then you're kind of guessing.
Whereas here we have direct evidence that, okay, we deployed this marketing, whether it's social media or paid digital. And this percentage of people that were exposed interacted with the marketing and then we're able to correlate that activity with actual purchase activity within the audience.
And so that's a huge leap forward, of course, in an attribution, in the whole subject of attribution. And so we had a campaign just this week. We were going through results with a client and it was, it was interesting. We had a couple of campaigns. We had a campaign where we're promoting mobile banking, digital banking services. And then we had a checking campaign. We're promoting checking accounts for the E Services campaign. We have 600 clicks on ads and we had roughly 400 of those 400 households adopted in the eService. So you have this ratio of, you know, 400 to 600 2/3 almost between the actual accounts open versus clicks, the checking, similar number of clicks, 600 clicks roughly, but 200 some open accounts. First thing at the most basic level is the engagement activity. The click activity gives you confidence that the campaign is generating a level of activity that is multiple of what you're counting as actual campaign driven activity associated with specific households in a defined window of time. So that's number one. I mean, am I getting two or three or four times the activity overall that I'm actually seeing in my new account file? That gives you some confidence that the overall response volume is a reasonable relative to the overall activity created by the campaign.
But then just more intuitively, on a digital banking campaign where I can click a link and sign up for the service, within a couple of minutes, you would expect to see a tighter ratio between clicks and actual response accounts because it's purely an online environment, whereas a checking account, it's a little bit more involved. I have to have certain information that may not be at my fingertips. Some clients have an efficient online account opening process for that product, others don't. And so you expect to see a bigger gap between your engagement measurement or activity level and your actual new account activity level. So first, first thing I would say is just as marketers and I've been doing this 25, almost 30 years, you've been doing this a long time. It's so exciting to be able to look at data that gives me direct evidence of interaction with the marketing within the audience that I targeted within the window in which they were the marketing was active and then be able to correlate that with the products purchased in that same audience because you have direct evidence that backs up, you know, Your attribution assumptions, I think that's really critical. And let me just drill into the second part of that. So the impression part and click part, pretty straightforward. And that applies to email as well. So I sent email to this number of people, this number of people, click and so forth. When I say correlate that with product purchase activity, what we're talking about is taking is when you create a campaign file of targeted households on the front end and then you, on that file, you copy onto that file various attributes of that relationship you have with that individual at the time the campaign is being deployed. Then you take an updated data file 90 days later.
And in the interim period, the digital ads are rotating continuously throughout that period. You take that updated new account or total accounts file at the end of 90 days.
So these are all accounts that are held at the organization. You match that to the campaign file and you look for accounts with open dates in the product category promoted open by people who were targeted. So three criteria, they were targeted.
They have an account open in the general category of products you promoted. And that account open date number three falls within the window in which marketing was active.
And so you can identify those activities and then correlate them with the impression and click activity. You have really the foundations of an attribution model that you can have some confidence in because you can link engagement created by the marketing with products purchased in the product in the categories that were being promoted by people promoted to during a window of time in which they were promoted.
[00:12:48] Speaker A: Yeah, and Tim, you said something really, really critical. You know, sometimes, and this happens in marketing, sometimes people who may not have the full knowledge of what's possible think about results reporting as purely stopping at that activity. Right. So how many times have we heard perhaps a brand agency person talk about a performance report and you look at the performance report and it's okay, I delivered X thousands or X billions impressions, and maybe sometimes it has some clicks on it. Well, that's not a performance report to your point. That's an engagement report. And so I think, and I was talking to somebody recently said sometimes you have to clarify your terms.
Attribution is not just stopping at what did I do? That's kind of what we meant by activity based. It's an ingredient. It's a really important ingredient. And we're getting better and better indicators all the time like you talked about with engagement.
And then self reported is another factor. And so sometimes people will say, well, I've got this many coupon code redemptions or even in the digital World, you talked about it. People will say, I got this many visits to my website or this many account openings.
Well, in a multi channel context, and we see this all the time. And we'll set aside the coupon description right now because that's a little bit old school. Although it still happens, but in the digital world, you know, the click stream gets interrupted. Right. How many times have we all. And we've all done this, you know. Yes. Sometimes you see an ad, you click right into it and you buy that thing within a few minutes. That happens.
But a lot of times what happens is it stimulates shopping behavior, Life gets busy, you go back to watching the ball game, you play with your kids, you know, work commitment pop up, life happens and some period of time goes by and then you come back and you buy it. And maybe you buy it on a different device at a different physical location. And this is all in the digital context, let alone maybe you decide, hey, you know what, I did some shopping, now I want to go in and talk to somebody. In a branch spanning channels can be very problematic if you just rely on a click stream.
[00:14:56] Speaker B: Yeah, and I think that's especially true for lending because there's this behavioral dynamic with borrowing that where the fear of being turned down is all, you know, it's always kind of in the back of your mind.
You know, you want to be absolutely sure the lender understands your situation, your needs, that sort of thing. And so that factors into. I want to mention one other thing, just kind of along these lines of just basic results tracking. And then I want to like to get into a discussion about control groups and the best way to think about control groups. But before we get to that, to take it a step further than just measuring, okay, these are accounts opened with open dates in the window when the campaign was active, associated with people that I was promoting to.
That's great.
But another, another element below the surface that can give you even further confidence in terms of the incremental nature of, of the results is to look at how buying behaviors correspond with material changes in the customer status.
So frequently as we track campaigns for clients, 90, 95% of our response population. So again, response being defined as customer who opened an account in the product category promoted during the period in which the marketing was active.
And when you look, okay, well the question I always ask, who are these people, you know, who are opening these accounts?
Frequently, you'll find 90, 95% of them have long term behaviors that are established. And then you see a material change in a long term Behavioral pattern directly corresponding with when the marketing was active. So let me give you an example.
I've got a checking household who has only had checking and they've been with me for six years.
They've never borrowed from me before, they don't have any other services with me.
And suddenly within a 90 day window of when they've been presented with a loan message, they take out a loan with me. So you have to ask yourself as a practical matter, this person, for six years they could have opened a loan anytime in that period. They've never opened any other service besides checking. And all of a sudden, within 20 days of the deployment or whatever it is, 30 days, 45 days of a deployment of a loan message, they suddenly borrow from me. Reasonable people would define that as a response to marketing because you're seeing a material change in a long term behavioral pattern associated with a point in time eventually.
I don't think it's unreasonable to say that. But you end up in these discussions where sometimes with the more statistician or statistically or how you would say it, statistician personalities.
[00:18:01] Speaker A: Yeah. If you take a more critical starting point and a skeptical mindset and an analytical bent of which there's a few of those people around banks, if they start with that and, and maybe they don't like marketing in the first place. Or then you get to some very esoteric conversations that I think you're, you're talking about.
[00:18:24] Speaker B: Yeah. So you get into the tree falls in a forest discussion. You know, it's going nowhere. All I know is this person never borrowed from it before. I sent them a piece of marketing and now all of a sudden they're a borrower. So you can draw your own. I think you can look at data like that. And so we kind of migrate into a control group discussion because that's really folks that have more of that mentality who are really kind of the statisticians in the room who bring good perspective to the discussion. But if you take a hard line type approach to that, I think you miss the bigger picture. And one of the fundamental issues, especially in cross sell marketing, is to have a statistically valid control population.
Especially during a discrete marketing event where you've got all kinds of other marketing going on and especially continuity based marketing. So this is my third quarter loan, cross sell, I did one and second first. I'm doing one and fourth. Having control populations where you don't have noise and they're measurable requires some rigor. And you start with size. So you have to have a large enough population as a holdout where.
And ideally that would be the number of what would be responses. So products purchased in the category promoted even though they weren't promoted is at least 100.
So that starts to. So if you've got 100 responses all of a sudden, I mean these are very large populations and if you're in a cross sell environment, you're starting with a limited population from the beginning.
So very frequently you find I will cannibalize half of my campaign to get a statistically valid control.
And in many cases the actual campaign itself becomes not valid statistically because what's left over is not large enough to be statistically valid either. You can't literally have a statistically valid individual campaign level control. And forget just the cannibalization, it's just the pure chinning the bar on the statistical validity piece. And when we say statistical validity that really becomes practical in cases like if I had five or six accounts that were opened three days later or three days earlier, so all of a sudden they're not in one of the populations. It totally changes the view of the campaign because the percentages shift enough and you project that forward and then that's really where you're on shaky ground.
And so the statistical validity piece, even if you kind of get to this place where I don't, I don't have faith that the engagement statistics are relative to my response statistics, I'm going to discount that. And then I'm going to, I'm not going to really look at the relationship part in that whole behavioral piece that I talked about earlier.
So I'm going to insist on having a control population.
You've got that issue to overcome which many cases it's not overcomeable statistically. And the worst thing in the world you can do is have a not valid control that gives you information that allows you to either say the campaign was more successful than it really was or less successful because cause a handful of accounts are pushing the percentages and directions that lead you to false premises.
So I think that's a practical issue that you have to deal with if you go down the road of kind of control groups relative to attribution.
[00:22:05] Speaker A: Yeah. And so one I think key point is as we're talking for anybody who may not know the technique that Tim's talking about is Matchback. What we believe is a superior approach in an omnichannel multichannel, you know, non straight line purchase path world is to not just rely on the ingredients so the impressions, the engagement, the click stream, all Those are important, they're necessary but not sufficient. And so we do is we match back actual account sales to the campaign audience. So we start with a specific audience of people addresses, run multi channel campaigns against them and then do a matchback against actual sales that happen in the window. So you know when Tim walked through that methodology that you'll hear that called matchback. And that's the kind of distinction between some of these other methodologies that you'll hear.
[00:22:59] Speaker B: Yeah. And the stuff we do for clients, which to me is the point of the spear, is more targeted, more frequent branding and search engine optimization and adwords and all that are great. You should be doing those things. The stuff we do is highly targeted and frequency based marketing. So if you are going to use controls, there's a couple things I would say is a really practical things you can think about in terms of the way you go about installing that kind of sanity check measurement. And that's a good way to think about it. So I'll talk about two attributes of that. One is a new customer onboarding, which is a very focused communication strategy targeting brand new customers to a bank or credit union over the first three to six months of their relationship with the institution to really make sure that they're fully adopting your digital banking services, that you're consolidating their financial wallet with the institution in that key early period.
So that's essentially a very high level definition of onboarding. What's unique about onboarding as contrasted with cross sell or acquisition marketing is every month by definition you have a new group of new households, you didn't have the prior month in the next month and the next month you're going to have the same thing. So what you can do because of the continuity of the program and the repetitive nature of it is you can randomly siphon off 10% of your new households every month. You can hold them out of contact and then eventually that holdout gets large enough to be statistically meaningful on a program to date basis. And then you compare that program to date universe to your contact to date, same periods built up over the same periods of time, both statistically valid. Then you can look at things like, you know, accountant household retention, if you, if I contacted you versus if I didn't fund up of your checking account balance, if I contacted you versus I didn't adoption of digital banking services, you know, these sorts of things. And then you can get really statistically valid and meaningful return on investment calculations based strictly on Lyft that give you a very high confidence that the money you're spending is productive and more productive than almost anything else you could do.
So that's a way to do a holdout and a program where the basic nature of it lends itself to, to that sort of measurement on a broader cross sell.
Because of the iterative nature of a good cross sell program where you've got messages going out continuously in different categories, the only way to do it without having noise in the numbers is to have a static control.
So you take your marketable universe, you randomly siphon off again maybe 10% of households, depending on the size of the institution. So you gotta be careful about that number. But an appropriate number of households and you hold them out of all contact, so they're not coming in and out of different campaigns where you're getting noise in your measurement, they're held out of all contact.
And you're not looking at individual campaign lift. What you're looking at is overall program lift.
So when I have a cohort that I'm actively marketing to in various ways over time, how is that cohort impacted compared to a cohort that is completely held out of that contact or that program?
Again, you're looking at things like retention accounts and households, change in deposits, change in loans at a per household level, change in accounts per household, these sorts of things. And one of the interesting things about doing that is you do see these broader lists, but the call out group is still being exposed to your signage, to your general advertising, to other things outside of the program.
So you do get a sense of how does proactive communication in an iterative cross sell environment derive broad relationship dynamics over time properly set up that can really give you some good data. And usually you find one or two or three things that are significant enough in terms of lift to say, yeah, this money I'm spending is a no brainer.
So there are ways to do holdouts that don't cannibalize 50% or even 20% of your campaign, but that are still statistically meaningful. But they're not.
They're more sanity check type metrics you can look at that are complementary to what we started the conversation with, which is correlating your engagement activity with your product purchase activity. And that's number one. If you're doing that and you're consistently seeing two or three or four times the click activity as you are products purchased in the same audience during the same period, that's really 95% of it.
If you still want to take that additional step of the randomized holdouts you've got to do it in ways that are statistically sound without cannibalizing your overall impacts. And so all that to say, intentionality expertise, having a coherent strategy about attribution, if you're going to do it, is really important.
I think clients struggle with either not measuring at all, or as Dan talked about, only measuring impressions and clicks, or on the other, the ditch on the other side of the road is holding things to a standard that's just simply not connected to reality in terms of the way that customers behave, the way that they take in marketing across channels, the whole concept of not taking into account the whole concept of the marketing mix, you know, these sorts of things. And so I think there's ditches on both sides of that.
[00:28:54] Speaker A: Yeah, well said. And so in this cookie world, you know, one thing you'll notice is there is a way to not worry about the cookiepocalypse.
[00:29:02] Speaker B: That's right.
[00:29:03] Speaker A: You know, I'm reminded of the princess, you know, bride line of, you know, life is pain. And anybody who tells you differently is selling something. Well, it sometimes feels like the, you know, and people talk about cookies. It's like, well, if people are screaming about the cookie apocalypse, they're probably selling something that needs a cookie.
[00:29:22] Speaker B: Right? Yeah, that's, it's, it's a, it's great subject. You know, for this discussion, I would say one of the, one of the advantages of the way Infusion does our work is we're not following you around online and then using cookies to come back and say, hey, that hotel in Miami is still available, you know, that you just looked up two days ago, you know, that's going away.
And there's a lot of panic in the industry about that. So I think that creates attribution challenges for people who are dependent on cookies.
We have data, behavioral data, that it's been predicted for 30 years on how consumers and businesses respond to different types of marketing.
So we start with our audience.
We don't start with where they've been online. We start with who we want to market to based on, you know, these proven behavioral things.
And so in that world, you're not counting on the cookie tracking to drive attribution.
You're able to associate account open activity from data files with individual households that you've targeted based on proven attributes. And so that's the scramble right now is people trying to go back, you know, pre cookie and say, okay, we got to rebuild models and be able to select lists not based on cookies, but based on predictive data.
Cookies, I think probably Made people lazy in terms of that. What used to be database marketing. We never stopped doing database marketing that way. And we continue to build our predictive models and our normative data sets that continue to predict these behaviors. And so, again, I think that puts us in a good place going forward. Because attribution is not based on being able to track someone's footprints online. It is associating account open activity with a household that was targeted with a message during that same period.
[00:31:28] Speaker A: And also, one of the opportunities it provides is for marketers to avoid waste when you start with an audience, as opposed to just the intent signal. And the intent signal can be useful, but if you start with the audience and what their likely needs are, it really feeds into that kind of continuous approach that we've been talking about and lets it be predictable and ultimately more relevant.
[00:31:50] Speaker B: Right.
[00:31:51] Speaker A: If you're putting a relevant message in front of a person, that can increase awareness, attention, preference, and increase irritation totally.
[00:32:01] Speaker B: And to go back to the hotel example, most of the times I've gotten those ads from Marriott, I already booked my hotel, you know, three days ago, and I'm sitting here thinking, don't you know that I already booked this hotel? Why are you placing an ad in front of me? You know, you almost outsmart yourself. I think. I think that's something we do as marketers sometimes with, especially with all this data. We outsmart ourselves because we forget the bigger. The bigger picture of it. So for us, it's, who do I want to talk to about what subjects? So specifically here are a groups of households where we're not meeting a financial need within their overall wallet of financial need. So therefore, I want to communicate to them about what we have to offer and how we could address their need.
So if that's my goal, then how can I communicate that efficiently and in ways that are, you know, that they can interact with? And that's, that's the definition of really effective digital marketing for us. You know, finding those audiences online, talking to them about irrelevant messages to their financial life, and then being able to associate how they react to that message to what they ultimately purchase and how that affects their overall relationship with the institution. That's probably in a nutshell, how I would summarize this whole subject.
[00:33:18] Speaker A: Yeah, very well summarized. And so I think the key takeaway for our audience is if you start with the audience and you think about attribution from day one, and you build an audience based on what their likely needs are. And Infusion has a very sophisticated approach here. It makes tracking results and attributing those results. Straightforward for me, when I've heard Tim talk about this and are thinking about our own experience, is that the key takeaway on ROI and attribution is to take a data driven approach, start with the audience and match back results in a reasonable time period to that audience. And that becomes a very robust way of understanding the impact that your marketing is having.
If you're not doing that, if you even want a second set of eyes. I mean, sometimes we do independent analysis for banks that have a very sophisticated program themselves, but they want a second set of eyes, a second set of analysis, a comparison, the infusion campaign norm.
That's something that we love doing and we have a couple of decades of experience doing across channels, across institution type and those kind of things. We'd love to have a conversation. If you want to bounce ideas off us, let us know. Thanks.
That's it for today on Top Quartile. If you haven't already, be sure to subscribe to Top Quartile wherever you find podcast on any podcast app. And while you're at it, we'd really appreciate a five star rating. And if you're interested in getting an opportunity assessment, head over to infusionmarketinggroup.com to learn more. Thanks for listening.
