Register for Napier’s webinar and learn why attribution might not truly reflect a campaign’s success. We will cover:
- Attribution models and the customer journey
- Why attribution is a childish measurement
- How B2B customers buy
- Business metrics
- Why you should measure incremental improvement
Register to view our webinar on demand by clicking here, and why not get in touch to let us know if our insights helped you.
Napier Webinar: ‘How to Integrate AI Into Your Marketing Campaigns’ Transcript
Speakers: Mike Maynard
Good afternoon everyone. Welcome to the latest Napier webinar. It’s good to have you all joining. I hope it’s all going well, and what I’m going to do is I’m going to talk about attribution.
So attribution isn’t always seen as the greatest, most interesting topic, but I think it’s really important. So a lot of what we do in digital is all about measurement, and it’s all about, you know, understanding what we do and making sure that we can actually measure what we achieve, and that’s obviously very important, but the way we measure it is really, really key to making sure we actually get the right numbers. And the problem is, is that, basically, quite often, we think up numbers that have no real basis. In fact, they’re total fairy tales, and we present them with three digits after the decimal point. And because it looks precise, we believe that it’s all precise. The reality is, is that those numbers can often be fake. And what we want to do is really try and dig in today about how we understand what is a really good metric for measuring the performance of our campaigns, and make sure that when we measure we’re actually measuring based upon business goals and not upon vanity metrics.
So let’s have a quick look at the agenda. Now, obviously, if anyone has any questions, I’d really encourage those. Please feel free to put the questions into either the chat or the Q&A. We’ll dig into both afterwards. Obviously Q&A is easier to find, and then I will answer the questions at the end, so type them in as you go along, and I’d love to answer some questions at the end. So we’ll talk a little bit about what attribution is. We’ll discuss some attribution models and how they relate to the customer journey. We’re then going to, you know, look at attribution and, you know, explain why it’s a childish fairy tale measurement. So we’re going to explain why attribution is poor. We’re going to look at how B to B customers buy, which is obviously very important. Understanding how they buy is key to making sure that we understand that our measurements match the actual behavior of the customers. And then we’re going to talk about measuring incrementality or incremental improvement. So all about actually looking at the best way to measure which, ironically, is a technique that very few systems will incorporate automatically. And we’ll talk about maybe why they don’t incorporate it. And then lastly, we’ll have a quick summary with a few tips on how to measure marketing campaigns.
So why do we care about attribution and measuring? Well, you know, what we’re trying to do, really, is answer the question that John Wanamaker posed many, many years ago when he said, half the money I spend on advertising is wasted. The trouble is, I don’t know which half Well, the idea is, is that attribution actually tags the people who come through, who see your advertising, and you can understand which half of the people are actually seeing your ad and doing something, and which half of the people are ignoring the ad. So it’s hugely important to be able to understand whether or not the ad is working, and hopefully, if we do use some way of measuring and attribution, is a technique we use, although, as I’ll show, it doesn’t necessarily give you the right answer. You can actually eliminate the half of the advertising that doesn’t work, so you can halve your advertising costs with no impact. So that’s the theory.
So what exactly is attribution? I think we all kind of know what it is. We all feel it’s, you know, linking your advertising to some results. But actually, you know, here’s the definition from Wikipedia, which is a bit clearer. So what attribution does is it identifies user actions, or they could be called events or touch points, and they try and understand which of those contribute to your desired outcome, and then assign a value to each of these events. So as an example, you know very, very simply, if you run an email campaign to somebody, they then pick up the phone and talk to a salesperson and then buy something for $10 you know, maybe you’d advertise, you know, allocate an amount to be related to the email and an amount to the person who answer the phone, because obviously you need. Both of those to work, so you’d allocate some money or some credit from the sale to each of those. So it’s really nice, because what it basically does is it gives you a monetary number associated with every marketing activity. So in theory, what it’s doing is telling you how much return you get from each marketing activity, and that is perfect, provided the allocation number is right. So attribution uses, you know, data in some mathematical way to allocate value to something you do, and there’s four key ways of doing it.
The first is called Single Source attribution. This is the very simplest. It’s typically first touch or last touch. And it’s very, very simple. So what it says is, is, if a customer makes a purchase, the very last marketing activity that touched them is responsible entirely for that purchase. Sales, people tend to love last touch attribution. They maybe don’t talk about it in that way. But I’m sure you’ve all seen the situation where you’ve run marketing campaigns into a prospect, you’ve worked really hard, you’ve got engagement, you’ve got PDF downloads, you’ve seen the progress. They phone in, they talk to a salesman. The salesman claims, yeah, they were never going to buy. They never saw any marketing. It’s just down to my call. That’s classic last touch attribution. It’s very simplistic and generally very wrong.
Fractional attribution tries to address the simplicity of single source attribution, and it allocates the value across a number of different activities, and that could be across all sorts of things. So the simplest way to do it is either to say everything you did all had the same impact, or the more recent it is, the more value you’ll give to it. Now clearly, we all know that some activities are more powerful than others. We all know that equal weights or time to decay are really poor ways to allocate value, probably better than first touch and last touch, but really limited. Now people try and get multi touch and curve models and things like that, so more complex models, but it’s still very arbitrary.
So the next thing that was developed was algorithmic, or probably probabilistic attribution. And basically what happens is, is a computer builds up a statistical model, and what it does is it tries to allocate the value based upon what it thinks has made the impact. So look at lots of different customer journeys. So very often we see this for digital ad platforms. For example, Google ads, they there. The default is using Google’s AI based algorithmic attribution. And so it’s creating a model. And this can actually be quite effective, although there are two key points. The first is, is typically the algorithm can only track activities within one platform. So with Google, it’s basically going to allocate all the value across anything you do on the Google platform, even if you’re running you know anything from PR through to attending trade shows, it’s going to allocate everything to the Google platform, because it doesn’t know what else is going on. So that’s a hugely important problem. And the second thing is, is you don’t actually have access, in general, to these algorithms, and say, not on Google, so you don’t know how accurate they are. And in particular, if you’re running a campaign that has relatively few conversions. And obviously, the closer you get to measuring your business metric, so that be a sale or a new customer win, the fewer conversions you’re going to get, the more difficult it’s going to be to build that model. It’s going to take time to build it. And if you don’t have enough conversions, the model won’t be built. So the attribution will, frankly, be making stuff up. It won’t have enough data to build the model.
So then lastly, there’s customer driven attribution models. And this is basically trying to understand what customers want and what they respond to. And quite often, this is done with things like, you know, literally, customer interviews and asking them and then allocating the weighting based upon what the customer says. And as we all know, as marketers, the customer would like to think they’re always right, but quite often they’re not. So again, all of these, these different approaches, have real issues. And fundamentally, the issue we’re seeing is that all these attribution techniques assume that any kind of marketing that interacts with the customer is going to have a positive impact. So let’s have a little bit of a look at this so you know if we look at what’s going to happen, here’s your classic sales funnel. And if we’ve got someone flowing through the sales funnel, they might have numerous touch points, you know, they might see some PR articles at the start, maybe some display ads or social posts. You know, towards the middle of the funnel, you know, perhaps you’re driving people with paid social to your website, or seeing them at a trade show. Maybe they’re downloading an e book, and towards the end, you know, maybe it’s a phone call or a salesperson meeting, or even a retargeting ad. So they’re seeing lots and lots of different items, and it’s assumed that all of these items are all going to positively contribute when we come to attribution, if the customer buys, and it’s assumed that none of them work if the customer fails to buy. So it’s a very simplistic model, and we can look at how these different models work very quickly.
So just graphically, this is first touch attribution as soon as they enter the funnel. We’ve still got a lot of work to bring them through the funnel, but we’re just going to assume that that work has no value. Last touch attribution right at the bottom. We’re going to assume all the work, bringing them through until they’re just about ready to buy, had no value, and then whatever touched it last is the most important. So not very good fractional attribution. I mean, basically we’re allocating, you know, different values to each stage based upon some arbitrary set of numbers. It could be either everybody gets the same so anything that touched the customer worked, or it could be some other time decay or something else.
And then we look at algorithmic, and we’ve talked about this again, and algorithmic, what we’re going to do is we’re going to allocate value throughout the funnel, which is great, but it’s some sort of, you know, magic black box algorithm that we don’t necessarily have access to, and frankly, may not have sufficient data to be able to make a Good decision if we’re talking about relatively low volumes.
And then lastly, we’ve got the algorithmic sorry, we’ve got the customer attribution, sorry, the title is wrong here, and then that’s based upon the feedback from talking to the customer. And so again, it requires the customer to give the correct answer. But ultimately, all of these are pretty childish. They’re all contributing to a bit of a fairy tale. You know, we’ve all had this discussion. You touched it last.
Not a great way to discuss anything overly simplistic for advertising, you know, or alternatively, we can implement the waiter that brings our food on the way he farmed and cooked it clearly he didn’t his last touch. But also, it’s important to note that it doesn’t matter how well the waiter does, if the farmer and the chef aren’t involved, there is no great result for the waiter. So attribution can ignore the interaction that’s required between different activities to generate the conversion. And generally speaking, attribution is pretty terrible at looking at where you need multiple touches, and it just assumes that each touch is somewhat independent. The next thing can be the issue around focusing on what’s measurable. So typically, if you’re running, say, a media relations campaign, you’re getting some great coverage in the press. People are getting excited. It’s really hard to link that through to sales. You don’t know who’s read your articles in the media, and so therefore you can’t really attribute anything to PR because there’s no way of tracking it.
However, if you’re running email marketing or Google ads, it’s really easy to track. The tracking is all there. So what happens is, is that attribution is a bit like the drunk person looking for their keys, and the drunk person looking for their keys will look under the lamppost because it’s light and they can see. However, we all know they drop their keys in the dark, and so do they, and it’s very similar when we look at attribution, attribution is looking at where things are illuminated through trackability, which is typical, typically digital, and quite often missing the things that are more difficult to see just because it’s going for the convenience of measuring, of focusing on what’s measurable.
And then lastly, and you know this, I think, is really quite important. I mean, attribution often doesn’t show you the money. It doesn’t always measure the impact on organization sales. And the reason for that is you’re allocating value for a sale to every marketing activity. Now it could be, particularly if this is a kind of, you know, a repeat purchase. So purchase through procurement for some. Think it’s in production that procurement person was going to buy anyway. And you know, if you keep hitting the procurement person around the time that they need to place the next order for their production requirements, you’re going to get some fantastic attribution. You’re going to make zero impact on the sales, but you’re going to get lots and lots of value attributed, and the same thing is also shown. You know, for example, on Google ads, when people advertise around, you know, different brand keywords.
A great example would be, you know, one of our clients, microchip, they have a platform called microchip Direct, which is their e commerce platform. So if you want to order a product from Microchip, you go to microchip direct. I can tell you that the best keyword in terms of attribution, if you’re running ads for microchip, will be microchip direct. Because you know, everybody searching for Microsoft direct has the intention of making a purchase. I can also tell you, because everyone already has the intention to purchase that the actual impact on the business is probably the lowest for microchip directs the keyword than anything else, and so attribution gives you completely the wrong picture. And actually, what this means is, ultimately, when we’re using attribution, a lot of people get frustrated because they’re trying to track attribution across a number of different channels, and what they find is one plus one equals 24 they add up all the value across their different channels that have been attributed to sales, and they find out that the total number is significantly more than the sales they’ve achieved. And so because of this, it’s really, really important to understand that, you know, the different attributions across different platforms. Don’t understand what’s going on on other platforms, so you won’t necessarily get the right value. So again, assigning a monetary value.
It sounds, you know, very convincing. But if you’re running, for example, a campaign on LinkedIn and then retargeting people on Google after they visit the landing page, both those campaigns will decide that they’re all responsible for the sales, and they’ll both want to claim all the value. And that’s not right. Actually, they both contribute, but they don’t share it because they’re different platforms.
Now, if there’s any engineers here, I can hear everybody shouting, we all know this correlation is not causation, and this is really important. So what attribution does is it measures correlation. It says people who saw this ad tended to buy, therefore the ad was fantastic. It doesn’t actually mean that because they match, there’s necessarily a direct link. There’s lots of famous examples of correlation not being causation. My favorite example probably is the direct link between violent crime and ice cream sales, as you can see here, the two curves follow each other very, very closely. Arguably, ice cream sales are lagging behind. It kind of suggests that, you know, maybe people go out and buy ice cream after committing violent crime, we should just arrest everybody buying ice cream. Quite clearly, that’s not true, you know. And equally, you know, it’s really clear that, basically, more violent crime happens when the weather is warm. People are more likely to be outside, meet more people, and potentially as well, there’s more alcohol consumed as well. But ice cream is not the cause of violent crime, and violent crime is not the cause of ice cream sales. And what we’re trying to do is we’re trying to do the same thing when we attribute. So we need something different. I mean, hopefully you’ve seen that attribution really is this fairy tale. It’s a bit of correlation, it’s a lot of approximation.
It’s platforms trying to grab as much credit as they can for their platform, and it’s not a very effective way of measuring the how well your marketing works. And the reason for this is that B to B customers don’t buy in a simple way. So firstly, different customers will have different approaches to buying so you can’t run one model that applies to everybody. The buying process is also complex with many people involved, and you’ve got to reach lots and lots of people, and those people want different information at different times, and frankly, you know, targeting one group of the decision making unit or the buying committee with content that’s really suitable for another I mean, that’s not going to actually help sell the product in most cases. And in fact, can even put the buyers off because you’re not supplying the information that they really care about personally, you supply it to the wrong person in B to B as.
Quite often the decision is made pretty early. And a great example of that is if you look at, for example, choosing semiconductors. So if you’re choosing to use a semiconductor, what will happen is you’ve got to buy a sample at semiconductor, put it into a board, build the board up, then write the software test the board, and then, only then can you start moving into production. So you’re actually choosing to the product well before you even buy that first sample or get that first sample to test. And so it’s very difficult to correlate the sale, which happens, you know, and in electronics and semiconductors, easily can happen a year after the decision has been made with the actual decision. So typically, what you’ll want to do is, you’ll want to tie sales through to marketing activities. But the reality is, is that’s quite hard to do. And also, in that particular case, once you move into volume, it’s almost certainly procurement that is purchasing the parts in volume, so they seem like the important buyer. But actually, if you’re talking about a 32 bit microcontroller, people in procurement are not swapping those microcontrollers out for different vendors. It’s the engineer at the start, who never spends any money, perhaps, who actually is responsible. So tracking is really, really hard.
Some of the buying is emotional, and that can lead to a very different level of importance for different touch points. And so that can be, for example, that a particular customer reference or maybe a white paper, or even a meeting at a trade show can have a huge impact, whereas perhaps display ads, whilst they slightly increase the likelihood of purchase, they’re not necessarily huge. Of course, display ads can also be one where actually, if you have sufficient display ads, then there’s very much a level of diminishing returns. Once you’ve got an awareness that means the potential customer is prepared to consider you as a vendor, then maybe those display ads running awareness perhaps aren’t going to add any more value if you keep putting more on and generally speaking, attribution has no concept of non linear effects. It always assumes some sort of linear effect. The more you do something, the more effective it’s going to be, B to B.
Customers are risk adverse. That means that obviously a lot of what you might need to do is give them reasons why they shouldn’t reject you, rather than reasons why they should buy. So that, again, can make it quite difficult. And you know, it’s very, very slow, as I mentioned. So what’s the solution? Well, the solution is easy for some markets and incredibly different for others. So the solution is to measure incrementality or incremental sales, and this is somewhat like AB testing, but what you’re AB testing is you’re AB testing the impact of running a campaign versus not running a campaign, as opposed to classic AB testing, which is, you know, for example, testing email a against email B. And in that situation, you know you can find that email B, for example, might perform better than email a, but you actually don’t know how much better email B performs than running no marketing. You know, it could almost be that email a is such a terrible email it puts people off, and so it results in actually a fall in sales, and email B makes very little difference to the likelihood of someone to buy. So AB testing can lead you down a route where you’re comparing two different activities, but you’re not necessarily really understanding whether you’re growing sales. So generally speaking, people talk about having control and test groups, and the test group gets exposed to a campaign, and the control group doesn’t. And then, ideally, you want to measure to sales, and not to just PDF download, although you can measure to that and see if you incrementally increase PDF downloads, again, incrementally increasing PDF downloads is better than one email winning an AB test against another. You know how much impact you’ve made. But of course, it’s very, very hard to do.
And so unfortunately, just like in all fairy tales, we have to have a wicked witch, and the Wicked Witch of incrementality comes in. It’s really hard in B to B to measure incrementality. It’s very difficult, for example, to get a proper control and test group. For example, if you’re running PR, it’s really hard to know who’s not seen your coverage online in publications versus who has. It’s very difficult to separate them in consumer markets, quite often, companies will choose different cities in the US or maybe regions in Europe. Uh, but that, of course, is fraught with difficulties, because people in different cities might behave differently. There might be different impacts. Of course, you’re measuring, you know, whether you increase the sales in each city.
But you know, in one city, for example, there could be, if you’re talking about selling luxury goods, there could be, you know, a large number of redundancies that causes luxury goods to fall, that has a bigger impact than your marketing. So it’s really hard to get these control groups. The buying journey is a long period of time. I mean, we’ve worked with clients who had multi year buying journeys. You really don’t want to know whether your marketing three years ago had a positive impact, and wait three years to find that out. So the time scale can be very difficult. The complexity also impacts things.
So the different members of the DMU, particularly, can interact very differently, and that can mask sometimes the benefits of some marketing. There’s often a huge number of touch points, you know, across a multi year buying cycle. Hopefully your marketing is reaching those prospects many, many times. So it’s almost impossible to isolate a single touch point, and within those long, complex buying journeys, each touch point has relatively little value. And also activities may interact. And it may be as simple as you just need to keep touching the customer at a certain frequency to make sure that they’re still engaged. Or it could be that, you know, if you don’t send an email, you know that’s around your certifications, then the following email about how high, how fast your processor is has no impact. So lots and lots of challenges from our wicked witch.
Fortunately, however, we do, as in all fairy tales, have a fairy godmother. And so really, our recommendation is you’ve got to do your best. I mean, measure campaigns where you can for incrementality, not for attribution. Obviously, sometime you’re not going to be able to measure incrementality. And very, very rarely are you going to actually be able to measure like full journey incrementality. You’re going to have to measure journey across intermediate metrics. So we would strongly recommend you know, firstly, always trying to identify, you know, whether you’ve increased sales or not. It sounds obvious, but it’s not something that a lot of people do. I think also consider, you know, testing incrementality across at a campaign level rather than a tactical level. So rather than testing a particular ad test a campaign, it’s much more effective to do that because of the length of the buying cycle.
Definitely, I’d recommend using intermediate metrics. So if you’ve mapped your customer journey, and hopefully you will have then look at moving people from one stage to the next of the customer journey, and use that as your measurement, because measuring across the full journey for most of the people on the call will be impractical, due to the time that people take to travel through it, and finally, don’t forget to measure non measurable marketing. And the example I’ve given multiple times, it’s the easy example, is getting media coverage. It’s something that’s virtually impossible to track digitally. It’s actually something you can measure really effectively using incrementality. So you know, it might be that actually what you say is we’re going to run some really powerful media campaigns around product A. We’re not going to, you know, run anything around product B, and we’re going to see whether product A’s sales increase and whether that therefore shows we’ve generated revenue. So incrementality is a great way to measure things that are non digital and traditionally hard to measure. So I hope you’ve enjoyed this. If you do all of these things, you will live happily ever after, and also your marketing will produce much better and much stronger results.
If you’ve got any questions, please feel free to put them into the Q and A or the chat. But before we go there, we’re going to talk about making chat GPT your best sales person for the next webinar, which will be on the 24th of June, 2025 so this is something I think a lot of people are really interested in. It’s about, how do you get your brand to show up in generative AI results? So if I type a query into chat GPT, and maybe that asks about, you know, the best DCDC converter, how do I get my brand to show up in those generative AI results? We’re going to talk a little bit about measuring how well you do, and the tools to that are only just emerging. And then we’ll also talk about what you can do to improve your visibility.
Perfect. So thank you very much for listening. If anyone does. Have a question. Here we go. We’ve got a couple of questions here.
So I’m being asked about software to assist with measuring an attribution. So the problem is is most software today typically focuses on attribution, and so most software today is looking to say there’s correlation. That’s not necessarily causation, and it’s a real issue for me, as you know someone that who’s quite passionate about marketing and measuring marketing, is that it really is. Did you touch that? That prospect on the way? Therefore it must have been positive. So very few systems have a great way of, you know, actually looking at what had a positive uplift on sales and how much that positive uplift was. So, I mean, we’ve asked about Marketo measure and Adobe mix modeler. I don’t want to talk about specific, you know, tools, but those tools are basically trying to match what you did against different prospects and see which ones touch the customers or touch the people that became customers. And as I said, you know, it is an indication, you know, attribution is not useless, useless, but it is kind of assuming, you know, in this fairy tale, that everything has a positive impact, and it’s also applying, you know, some sort of statistical analysis to say there’s a formula for it. So I would say that today it’s really hard to use software to measure incremental sales, because actually, most software is not designed to do that. And if I’m being cynical, a lot of software is designed to claim as much credit for the tactics you’re running with that particular platform. And Google and LinkedIn are great example of this as it possibly can so, it’s looking to allocate as much value to its own tactics. Some of the other tools that mentioned here in the question, they do look across more than one platform, but they are still limited, and they are still doing this really simple attribution, rather than measuring incrementality in most cases.
I hope you enjoyed this. I hope it was interesting. It’s always difficult presenting a webinar that’s about a topic like attribution, which isn’t on the face of it, the most exciting topic.
Hopefully you can tell I’m super passionate about measuring marketing activities well, so I’d really love to talk to you. If any of you have questions afterwards, please feel free to email me. My email address is there, mike@napierb.com and I’d be happy to have a chat. And I really hope I see you all in the next webinar where we can talk about appearing in generative AI search results. Thank you very much.
