Name: Why Google Smart Bidding Chases Bad Data (And How to Stop It)
Uploaded: 2026-09-16T15:16:25.960Z
Duration: 1 h 1 min 36 s
Description: Why Google Smart Bidding Chases Bad Data (And How to Stop It)

Transcript for "Why Google Smart Bidding Chases Bad Data (And How to Stop It)": Hey, guys. Welcome to today's smart bidding webinar, and a big thanks to those of you who are here bang on time. I'm sure we'll have a few more people joining us over the next couple of minutes. So while they continue to join, I'll quickly introduce our guests for today's session. We've got Heidi Sturrock. You've probably read her articles on Deep Sea Live and Search Engine Land. But Heidi is the lead Google strategist at OMG Commerce, and she also does a lot of valuable stuff over on her own website, heidistorrock.com, where you can read even more paid media insights from Heidi or get in touch with her directly. And then joining Heidi is Scott Carathers, friend of Lunio, previous podcast guest, webinar guest as well. Scott is the senior director of paid search at Journey Further and keynote speaker at, I would guess, about a thousand and one marketing events over the past year. Is that right, Scott? No? Okay. Close enough. Thank you both very much for joining us today. Scott, I know we're a bit outside of the season now, but have you got any more events on the calendar for 2026, or are we looking at 2027 now? think it's I think it's 2027 now. It's interesting you say that about the keynote speaker thing because I actually haven't done as many as you would have... You you would you would have thought. But I am on the lookout for more slots, so let me know if, if there's any good ones that you're planning on attending. I feel like every main event I saw that was going on in The UK had your face on the on the poster, but maybe that was I was just looking at the same posters over and over again. Heidi, I know you spoke at the SMS SMX advanced. panel in Boston back in June. Have you got anything similar lined up for next year, or is that kind of a surprise for now? Lots of stuff coming headed out into 2027. Probably, you'll find me in New York and Boston, maybe California. So stay tuned on LinkedIn. You'll see the announcement. Cool. Cool. Lots of fun stuff. Lots of lots of changes happening across the industry, so there's lots to talk about for sure. For sure. Yeah. More than ever. Alright. Cool. So we've had a fair few people join now, so let's get going. Again, welcome, everyone. Appreciate you making the time to join us today, whether you're here live or watching back on demand. But first of all, I wanted to take a second to reassure any previous Lunio webinar attendees that James has, in fact, not been haunted by the ghost of Ed Sheeran. To be honest, I think Ed Sheeran's got bigger issues he needs to attend to right now. But, no. My name is Ben. I'm the content manager. I work very closely with James here at Lunio, and I'll be running the session today. But if you are a long standing James Deanie fan, let me reassure you that his lovely Irish accent is not gone forever. He's here today helping me moderate the session, and then going forward, we'll be sharing webinar hosting duties in order to bring you the most valuable sessions we possibly can. But, yeah, with all that said, let's get the slides up here. And I'm just gonna quickly walk through what we'll be doing in today's session. So we'll start off by taking an informative look at how smart bidding actually learns, which will then kind of provide the foundation for the rest of the session. We'll then be taking a closer look at the bad data that's corrupt in Smart Bidding before explaining why corrupted campaigns never recover even after they're fixed. And after that, we'll take a closer look at how you can use data exclusions to make sure Smart Bidding isn't trained on the wrong data. And then finally, we'll be covering the big one, that August 17 Smart Bidding update. That will include the initial results and behaviors we've seen following the change along with what you should be doing right now given that we're almost exactly one month on. But before we get into all that, allow me just thirty more seconds to do a few more bits of webinar housekeeping. First up, we are recording the session today. It will be available to watch back on demand after the event. If you have any questions about anything we covered today, please do pop them into the q and a tab so they don't get lost in the chat. We'll try and get around to as many as we can, at the end of the session, and you can also upvote the questions you are the most interested in. That will help us prioritize which ones to answer first, so please do upvote any questions you like. We've also got some great resources today in the docs tab here in Goldcast, and I'll be directing your attention there at relevant points during the session, so do take a look at that. And then finally, we'll be running a couple of audience polls during the session. You can vote on those in the polls tab. You should see a notification pop up whenever a new poll is open, and you can respond by clicking on the polls tab and then submitting your answer. But on that note, I'm just gonna bring up our first poll on the screen here now to kick things off. So as I mentioned, we're about a month past the August 17 smart bidding update. And it's obviously still very early days, but it would be interesting to get a general consensus on how the changes are shaping up so far. So there's a few choices here. Please do vote for the option that resonates with you the most over in the post tab. But while we're waiting for answers to come through from the audience, I'd love to get some quick initial thoughts, on the change from you, first of all, Heidi. Have you noticed any changes at all so far following the update, or is it still too early to tell? Yeah. I mean, I think it really varies between advertisers. So, definitely, you know, working with a variety of accounts very large accounts and and then smaller accounts, we do see a bit of a difference there. So August 17 came and went, and what we see is accounts with historically large amounts of data associated with them. When we, you know, scale when we take a look back at, you know, a full conversion cycle, what that target was, we adjust the target, you know, it's business as usual. It it seems to be, you know, working fine, with with little disruption. I think where the struggle happens, it might be smaller accounts with fewer, instances of strong conversion history. And so, there's less validity in the target they're trying to set after August 17 for the algorithm to work with effectively. So there's a little bit disruption there. But, you know, I think as is everything, things eventually, the dust will settle. And I think, you know, we'll see a little more predictive results after this. So it's still early, right? We're still about a month out. So, my advice is just, really take a look at what your historical data was. You know, keep in mind this is for campaigns with limited by budget. Right? So so focus on those campaigns and, and set your target based on your past conversion history. Nice. Wise words. And then, Scott, are you kind of feeling the same as Heidi there? Or Yeah. I think, I think what Heidi mentioned there about, conversion cycles, and, actually, we are only a month out from it. And most advertiser... A lot of advertisers will have a thirty day conversion cycle that when there's no way to really say convincingly that it's gone either way. Generally, the feeling though at the moment, that I'm seeing and it might speak to the type of accounts that that we manage, but very little change so far from a performance perspective. There's something we can talk about later on in terms of, like, a control perspective, but certainly from a performance perspective, fairly minimal change so far. Yeah. And it looks like the audience kind of feel the same way. Most people saying that they're noticing things looking about the same, few people saying worse. Yeah. Interesting stuff. Alright. So as I said, we will get much deeper into that huge change later on in the session. But before that, I was reading Heidi's guide on smart bidding and data exclusions while I was prep prepping for the session today. And there is one quote from it that I wanted to share here purely because I think it's a really nice foundation for the rest of the content today. And it reads, smart bidding is exceptionally powerful, but it possesses absolutely no common sense. It cannot detect context or read the room. So with that in mind, I wanted to start off with you, Heidi. In your SEM insights video on this on the slide here, you start with the analogy that using smart bidding is kind of like training a new employee. And that if you train them on corrupted data, they'll conclude the spammy leads were brilliant, and they'll go looking for more. But just so everyone's kind of on the same page here, can you start by walking us through how the smart bidding algorithm actually learns maybe by touching on the signals it's evaluating in each auction or by telling us perhaps exactly what a single conversion actually teaches it. Yes. Absolutely. So first of all, Smart Bidding, your best friend. You know, it's a wonderful way to get the algorithm to learn a little faster and more intelligently to find your unique buyer. It's come a long way, right, over the years, and it's probably the strongest it's ever been. It uses so much. Right? It's going to take a look at who your ads are targeting, and it's going to really observe who's crossing the finish line for you. Who's who's your ideal customer? It's going to look at the devices that a person is most likely to convert on. You know, the location where they are demographically. For example, do people in cities convert better than rural areas? The time of day, do you have people browsing at night, you know, or do you have people during the workday on their PCs? So, there's lots of contextual signals that it's using right now to find that perfect avatar for you. And it works really well, and your campaigns can get really, really successful if you have a lot of that conversion data coming in. In. And so our conversion tracking, of course, is paramount to that. Right? We have to have the great things that we're tracking, the most meaningful things to feed the algorithm what it needs, so to define what success looks like. And then what's gonna happen is every time you get one of those important conversions that you set up through your conversion action, it's going to take all those signals together, and and it's gonna assign a value to it, and it's going to go after those. Right? And it's going to spend a little bit more to get that high value conversion, which is why oftentimes in an account, you'll actually start to see some of your secondary metrics increase like your CPMs or your CPCs. And and initially, people are think that's a negative thing, but it's kind of worth following the pattern because if you're seeing better conversion rates, better return on ad spend, better CPAs, it's usually an indication that the algorithm's, trying to go after more expensive auctions to get you that better return. So, that's the background. That's kind of how it works. Right? But here's where it can kind of go go south. So the the employee analogy, essentially. So think of your trusty algorithm on the back end attached to your bidding models as your employee. And let's say you accidentally train it on the wrong success metrics. So so for example, let's say that you're an ecommerce business and your success metric is sales. Right? You really, really need a strong CPA or a strong return on ad spend, but you've accidentally, you know, trained your c your your your trustee employee, your algorithm to go after newsletter sign ups. Right? So, yeah, it might still be kind of important, but really not what your, core metric of your business needs to go after. So, you know, really, when that instance happens, the algorithm it doesn't know. It doesn't know that newsletter sign ups aren't as important to you, you know, as as actual website sales, ecommerce sales. So it's it's going to start looking for people that are really great at signing up at newsletters, but maybe not buying anything for you. Right? And so that's an example of what we just talked about where you can, have a perfect setup. Right? So the perfect atmosphere of the perfect bidding model that you've chosen. You have, campaigns set up flawlessly, but it's attached to the wrong conversion conversion success metric. The algorithm's chasing something entirely different and feeding the signals into your campaign. Awesome. Yeah. Super detailed answer there. Thank you, Heidi. So smart bidding is very powerful, but before we go any further, I wanted to quickly touch on the dangers of smart bidding. One of those being how quickly it can go wrong. Scott, these days, you are an authority on all things Google Ads. I wanna profess with that, but, I do apologize for bringing back bad memories here. On a recent episode of the PPC live podcast, you told the story of launching smart bidding, setting a target CPA strategy live, and then it quickly spending a client's entire daily budget of over £10,000 in under an hour. Now I do know this happened quite a long time ago back when smart bidding was new, but, is there anything that experience taught you about how quickly automated bidding compounds its decisions? And is there anything else you kinda learned from that experience that still informs your bidding strategy or process today at all. I think the biggest thing I've learned is not to bring up ten year old mistakes on a podcast. Otherwise, you'll just go with some impassioned of them. You get people like me just, what, over and over again. Yeah. Yeah. We're talking about it a lot now. I think it was it was a very long time ago, so, probably the more technical strategies that have been employed... That that we employed since then or learned since then have then changed again, But the underlying principles that I learned, remain the same. So things like looking at that historic conversion data to understand how were the campaigns performing before you made that bid strategy change, what are the targets that you are now setting, What other guardrails do you potentially have in place that you could use to limit or control that campaign? Because, while this is a very extreme example, we do still see, when changes get put in place that that spend can run away or changes can happen really, really quickly, based on that historic data that you've got in the accounts. Nice. Yeah. And then following on from that, I want to get into the smart bidding failure modes. I've called this the bad day of hall of shame, and it's the recurring ways that corrupted data gets into your account in the first place, which in turn corrupts what smart bidding is learning. So you can see the five on the screen here, but messy conversions is where I want to start. And I wanna stick with you for a minute here, Scott. So before we get to things like broken tags, you've said that conversion action setup is arguably the most important thing in Google Ads right now, that you regularly audit accounts with 50 conversion actions where nobody's sure which are primary or where a single conversion action is stretched across totally different products, stuff like that. But how much of the bad data problem is really advertisers giving Google the wrong definition of success in the first place? Could you maybe also run us through what clean conversion action setup looks like? Yes. Yeah. Absolutely. I think to Heidi's earlier point about teaching the algorithm the wrong things, it is super important, like, to to... As you've just said as well, the conversion action setup is one of the most important, if not the most important thing you can do in the Google Ads account. If you look at things like the target in controls that we had previously, things like keywords, which were very slowly being moved away from with, things like broad match AI max, performance max, those targeting... Keyword targeting is not as strong as it used to be. The one signal that we do still have that we do have control over that we can tell Google how we want that targeting to work is that success metric. It is what is working for us as a business. What do we want more of? So how do we tell it that is through those conversion actions. In terms of a a clean conversion action setup, yeah, you're right. I have seen I have seen a lot of accounts where they have a lot of conversion actions. The the last account that I audit... Audited had, 57 conversion actions in it, 13 of which have been removed, but there were still 44 live conversion actions in there, which is a lot. I think a clean conversion action setup is one that, one, relates back to your business objectives, So you can tie it back to actually what are we trying to achieve as a business. What is it that we're trying to do? Is it to drive profit? Is it to drive leads? And it might be to hide it early, but it might be... Actually, we do all just want to drive loads of newsletter sign ups. That might be your your business objective at that time. You have to be able to type back to that business objective. And two, it has to be understandable and clean. So how anyone can follow 57 different conversion actions or understand what they're doing within an account is is anyone's guess. You can use primary and secondary conversion in there, but amongst that, you can have a lot of different conversion actions. You can have custom goals set up. You can use conversion value rules. There's the up weight and down weighting for new users nowadays. So there's a lot of complexity to that to that. So, making sure that you understand it, whether that's through the naming conversion... Naming conventions of those conversion actions, whether it's simply doing some spring cleaning and then stripping out those those old conversion actions that you no longer use. I think a clean one is one that, again, ties back to the business objectives, but is one that you can follow, understand, make sure that there's gonna be no issues anywhere in your account. Nice. So clean conversion actions tied to your business objectives. Very good. And then, Heidi, your data exclusions guide walks through four scenarios you've seen in real accounts, so namely numbers two to five on the slide here. But I was curious which of these you encounter most. Do you maybe have any, like, particularly egregious examples of horror stories you can share with us at all? Yeah. So, first of all, for everyone listening, don't feel bad if you found yourself in one of these scenarios. I've worked in the industry for over twenty five years. I can tell you even some of the most sophisticated accounts have experienced this. But, you know, right now, we're doing a lot with YouTube and top of funnel types of campaigns. And with that, sometimes to get a, in a a campaign kind of, warmed up if the conversions are either, going to be fewer and far between because maybe it's a long sales cycle or a really high ticket item, price ticket item, sometimes the advertiser will add a mic what we call a micro conversion, right, which is not your primary conversion as you define it by success like a sale if you're an ecommerce company, but an add to cart. So for example, a micro conversion could be an add to cart where it's not your primary conversion, but it's an indication of of possibly leading up to that. And and advertiser advertisers will put it in there to sort of warm up the algorithm and then eventually take it out when it can be strong enough to subs the campaign can be strong enough to, you know, exist on its own just using the the the regular sale metric. But but what ends up happening is, people will set up, you know, your sale metric, right, your your your ecommerce sale, and then they'll set up add to carts. Right? And, what where it kind of falls apart is, when the best practice really is to, create, you know, a custom conversion action that includes, like, the add to cart and the sale and assign it at a campaign level where that campaign is in fact going to need that unique setup, versus the entire account, that's where we kind of see things fall apart. So I'll I'll go in and audit an an account. And is, you know, similar to what Scott will say, I'm like, wow. You're getting a lot of sales, but why is the AOV so low, right, when the AOV is so low? Not because the sales are actually coming in with low average order values. It's because you're getting a conversion value reported from your ecommerce sale and then a ton of conversions. Right? Add to carts and conversions that are kind of, making the AOV, you know, artificially deflated there. So I would say, you know, be the most egregious attempts are when, you know, my the client comes to me and they're like, oh my gosh, you know, we need your help. We need to look and see how to clean this up because we're not sure what's reporting on what. So I'd I'd say like, even though people have multiple conversion actions within them, the biggest fault we see consistently is they're not being applied specifically to the campaigns where they're needed. They're apply they're being applied to the account total. So all the campaigns are grabbing them. Right? And they're they're just reporting these wild numbers. So I would say, you know, to anyone, listening right today, your conversion tracking setup is going to be, like, you know, Scott mentioned, fundamental. It's probably and arguably the most important part of setting up an an account. And so when you have it set up, great, but just make sure that those conversion actions are only being tied to those campaigns that they're relevant to if you have that type of setup where you're extending beyond just your typical one conversion action that you need for the account. Nice. Great stuff. So this slide exemplifies another key point where people go wrong with smart bidding. Typically, when people find a tracking bug, they fix it and assume that's the end of it. Job done, move on kind of thing. But what you can see here is that the bug going live is only really the start of the issue. The problem is that the model actually trains on that. And after you fix the bug, your dashboards may even start looking normal again only for you to find out kind of weeks later that your bids are still chasing after that bad data. So, Heidi, you mentioned that a client fixed a double firing bug in three days, but cost stayed inflated well beyond that. And it does kind of beg the question, why doesn't Smart Bid In just correct itself once the tracking is fixed? Yeah. So so let's go back to the fundamental way that these advanced algorithms work. They're using data modeling. Right? And so the data is being, looked at from historic points in the account to determine what will likely succeed next. And if you have, let's say, three days of really, really bad data coming into the account. In this case, it was when we had a tag double firing. So imagine double the conversions coming in. The algorithm's getting really excited. It's like, woah. These people are super high value. I'm gonna spend a ton of money going after them, and I'm going to train the next conversion cycle essentially to do that too because I it's a predictive modeling environment. So so what ends up happening is the algorithm, it doesn't know that even though you changed it on the back end that it should work with something differently in that moment. It's going to still use that historical pattern and slowly integrate the fix. Right? The more data you have in the fix fixed period and you're running your ads, the the better you'll see the results start to show. But before that happens, you're probably gonna look at a long period of time where the algorithm doesn't quite understand what it should be looking at it and what it shouldn't be looking at. Right? All it knows is your past historical context. And usually what we see, if you're not using a data exclusion, we'll see the algorithm usually try to look at the past thirty days in the account and say, what happened here? Let's let's go after all these all these these signals that I've received in the last thirty days. So even if you have three days of of of a data mishap of the conversion tag giving the algorithm the wrong data, if it's a lot of it, in this case, it was. I mean, the algorithm got got super excited. It was like, woah. Look at all these new customers I I'm getting for you. That's a really strong signal. Okay? And the algorithm is gonna latch on to that. And so that's why we use these data exclusions. So we can tell the algorithm, hey. I want you to totally have a blindfold on. Right? Imagine, like, putting a little blindfold on your employee. Like, don't even look at what we we trained you during those days. Nice. Yeah. And we will get into those data exclusions in just a second. But first, there is one other thing you need to be aware of when it comes to data integrity, and that is, of course, invalid traffic. And I wouldn't be doing my job if I didn't include a quick Lunio plug here because that's exactly what Lunio does. But I wanted to highlight our brand new retail report where we examined invalid traffic data across the entire retail industry. And what we found was actually a worrying upward trend where invalid traffic rose every every quarter across the last year. But of all the findings from the report, one really stood out as troubling, and that is a 72% higher invalid traffic rate in AI Mac versus standard search campaigns. And that's especially worrying given that Google is now pushing AI Mac speeches as default. Anyway, in the interest of time, I'll skip past this, but I've dropped a link to the full ungated report in the docs tab if you do wanna read the whole thing. Explains the methodology. There's a whole bunch of other findings in there. Really useful stuff as we head into the busy q four period. Maybe it will save you a lot of headaches. Anyway, moving on now to what Heidi touched on earlier, and that is data exclusions. And this is the more practical part of today's session. I'm assuming that if you're here today, you already know what data exclusions are. But in case you don't or maybe if you know they existed but have never actually used them, there is a quick recap on the screen here. To put it very simply, a data exclusion is telling your smart bidding to ignore this period when learning, like Heidi said earlier. So data exclusions might sound easy enough, but improper usage can cause far more harm than good. So knowing how to apply them strategically is crucial. And with that in mind, I'd like to pass back to you here, Heidi. Can you maybe walk us through how to apply an exclusion well? So maybe things like choosing the right conversion action instead of blinding the whole account, extending the exclusion window backwards, and stuff like that. Sure. So first of all, when you start, and everyone has had this happen, if you worked in the industry, you've had something happen to your conversion tracking, right? So the minute you notice it, don't panic. The first thing you should do is just try to understand the days that it happened. Right? Was it a day? Was it two days? Was it months? Right? Hopefully not months, but, you know, figure out when it happened. So you're gonna first understand the parameters, and then you're gonna try to understand after you figure out when your conversion tracking was incorrect, what campaigns did it affect? Did it affect every campaign in the account? Did it affect only one campaign in the account that was associated with that conversion action? And what you're going to do after you understand the scope of your problem, you're gonna go in and you're gonna use a data exclusion to remedy the situation, tell the account to ignore the period of time where this date this bad data was coming in, and you're going to, in some cases, only tell it to apply this reasoning to a specific campaign where the disruption happened. Now, this is a great, great option for you if you have a a situation where your data exclusion only needs to be a couple of weeks. I wanna just do a caveat here. When we talk about, being a good candidate for a data exclusion, The way that this feature in Google Ads works is that it's most appropriate for about a window of two weeks. Right? So it purposely was not created to cover months on end because the algorithm simply isn't designed to handle that. Right? It really does need a steady flow of data coming in. So if you have a mishap and most people are in those accounts every day, you'll notice it probably before two weeks even happened. It's it's about a two week window that you really should be using these for. So understand that, understand the setup part of it, and then think about something else. So remember we talked about predictive modeling earlier? So we all know and that if you have someone who sees your ad, right, or clicks on your ad, watched a video of you, chances are they didn't go into your site, or your app and convert right away. There's usually a a lag window that happens in between when they were first exposed to the advertising to the conversion. So consider that too because check your lag report. Most of you know this, the average window it takes someone to convert after they've had that initial reaction or exposure to your ad, and understand what that is. Factor that into when you want to apply the start date to your data exclusion. You're gonna wanna kind of if you can, push back the start date to accommodate that data, that lag window. Because remember that that predictive modeling, you're going to have the algorithm, try to, even apply its learning. It's it's faulty learnings, you know, to to to times before even when the click happens. So so just I I'd say consider that when you're setting it up. And then also, this is a famous thing too, and I'm gonna throw this out there because this is real this happens whenever we do our year end reviews or our quarterly reviews. They'll say, gosh. Why was performance so great compared to this period? And they're often looking at a time when, like, the tag missed, like, double counted or or in the inverse when they'll say, why was performance so bad back then and looking so good now? And it was because, like, maybe conversion tracking was out or undercounting. So document document when this happens so you could go back and you can say, oh, guys, remember, you know, last year or, you know, during this period, we had this mishap of conversion data. Because the unfortunate thing is even if you apply a data exclusion, as we mentioned earlier, the metrics in your account won't change. Right? They'll show still show the performance as it played out. Right? It won't it won't correct it in the platform. So just something to note there. Nice. Yeah. And just to follow-up on your answer there, Heidi, Scott, we will come back to you in just a second. I promise. But, you mentioned that the exclusion should rarely exceed two weeks. Mhmm. And you've also said in the past that if you're constantly excluding data, the real problem is your measurement. And you called it. a circuit breaker, not a Band Aid. So could you just maybe talk a little bit more about that and find out what should people expect during the recovery period after applying data exclusions? Sure. So let's let's think about the modern day search account. Right? It's not what it was even five years ago. And the fundamental way you're gonna succeed in Google Ads is through a strong conversion history. If you are constantly saying to the account, oh, just ignore that. You know? It's going to really be on the struggle bus to gain momentum and learn, which signals are going to get you the most successful return at the end of the day. So, again, the real problem is your measurement if you're constantly having to pull out and use this tool. You really want it to be an instance where you're using a data exclusion very rarely because, hopefully, that conversion tracking is super solid. So if you find yourself constantly having issues, seek out a way to get solid conversion tracking in your account. Maybe you need to go reach out to a third party provider. Maybe you need to hire someone who's professionally versed in your specific platform that you're using. So if you're on Shopify or something else, and fix the root of that. Here's the other thing I really wanna call out because I get this question a lot from my clients. They're like, oh, gosh. You know, what a flop our sale was. Could we just exclude that data from the account so the algorithm just, like, pretends it never happens and and and doesn't, get discouraged by it? And my advice is no. Don't. Because, again, conversion tracking, in its form of, bad performance does not warrant a data exclusion. It's actually valuable. Even though it's really disappointing and not really fun to have to report. Bad performance is still a key learning for the algorithm. It tells the other the algorithm what not to look for, right, because it's not meeting your target. So don't exclude bad performance. So keep it in there. Let it learn. Only exclude for, you know, the periods of time where your con where your performance was driven by a faulty untrue aspect of the conversion tracking, like misfiring, the wrong action, things like that. And after you apply that conversion action, just remember your reports are as usual. So so note it. The algorithm will eventually start quickly relearning, especially if you are an account that benefits from a lot of conversion action coming in at once. So that algorithm will relearn quickly. And usually within a few weeks, we see it bounce back pretty quickly. Right? So don't worry if things, you know, are still a little wobbly after the first week after you apply that data exclusion. Remember, the algorithm is starting to exclusively focus on that data coming in now, and it's gonna need a bit of time to kind of reestablish its base its baseline. So just try to stay patient. And my my, my advice to everyone is it's going to be super, super, you're gonna feel really compelled to make probably more changes than you would in the account to get it to stabilize further. That's usually not a great idea. Try to do business as usual with everything that you know that works and let the algorithm slowly come to terms with the new dataset and start reacting appropriately to it. Nice. Really solid advice. Thank you, Heidi. Alright. So as Heidi has just gone through, data exclusions are the fix once something has already gone wrong. Obviously, the better outcome, though, is catching something before it ever gets that far. And before I come to you on this, Scott, I just want to quickly point everyone towards two free scripts that are sitting in the docs tab right now. The first is a script created by our previous webinar guest, Nils Roymans. I think I said his last name right. Not sure. But that's his daily budget over delivery alert. And the second is Google's own account anomaly detector, which compares your performance today against the same day of the week over previous weeks. That will flag whenever clicks, costs, or conversions look unusual. Convergence being in there is the important bit because that's what makes it useful for spotting things like tracking breakages rather than just overspend. Both of those scripts are free. If you're not already using them, they take less than ten minutes to set up. There's no coding involved or anything like that. But, Scott, you've mentioned before that you run a whole philosophy of checks and balances that journey further, things kinda like extensive reporting alert, monitoring for any changes. But now, especially with a lot more, like, automation and stuff in the mix, can you maybe go in a bit deeper into what your current guardrails look like? And, maybe for someone who who's a solo in house marketer without the full agency infrastructure, rest? Yeah. So we've we've always had that philosophy of giving clients live data. So we have, dashboards that update daily so clients know exactly what's going on on a regular basis, none of this whole waiting for a monthly review type of thing. But with the advent of AI or the advancement of AI, our tech stack and our guardrail stack has absolutely bloomed. So there's a lot lot more in there now than there was historically. And and anyone can jump on that and do that. I think Nils' daily budget, over delivery alert is a great place to start. You can feed that into code and iterate from there and make changes based on on that original script if you so inclined. But our guardrail stack is... I think there's over 20 different alerts in there for things that we think are important. I think for someone working in house or a different agency, that is gonna be different for you because your best practice will be different to what our best practice is. But some of the basics in there are things like if your spend is going pretty crazy in one hour versus the previous hour or if your conversions have dropped to zero or, things like that. It is all built on, a dashboard that lets you change that by account. So, the threshold for certain accounts, like, some accounts might not get any conversions each hour. So having an alert that goes off every single hour when there's no conversions is no good, so you have to be conscious of things like that. But but it's it's, it's easier than ever to get these these alerts set up even if it's something super basic like in platform setting up some rules, where you can literally go into platform, hit rules, email me when certain event occurs. You can get an email through to to let you know that that's happening in the account. As I said, I think you need to think about, if you're in house, what is important to you, what is your best practice? What does that look like? So something that we've built very recently, which just runs once a week is it checks for any new campaigns across the agency that are set to presence, and interest rather than just presence. And we know that that's a a big thing that, a lot of people are impacted by is setting up that campaign, accidentally leaving it on presence or interest, and starting to get traffic from a location that you don't service. Just making sure that that's ticked off. And there there will be cases that actually that is the right thing to do and you do want to run the presence of interest. But thinking about things like that that that where you've seen errors in the past or things that you're always checking for yourself, take that off your mind, get something built to flag up and let you know when that's happening. Perfect. Great stuff. So we are running a little bit short on time now. So I did promise that we would spend the final part of the session talking about that pesky August 17 smart bidding update and kind of the resulting fallout from it. It's definitely the biggest smart bidding change I can remember in recent memory. But in case you have been living under a rock and have somehow missed it, I'll put a quick recap on the slide here. Essentially, before August 17, you tagged it as a ceiling, so a budget limited campaign with a $20 CPA could actually be running at fifteen dollars, and you just got to keep the upside. But now your target is a goal. If that same campaign is actually running at 15, that will drift back upwards towards 20 unless you go in and reset it to the number you actually want. So, ultimately, your target is a stronger signal than it has ever been, which means a target calibrated using polluted data is now a target Google will keep chasing. Meaning, the accurate targets that aren't trained on bad data are obviously now more important than ever. So let me kick things off with you here, Scott. On the PPC live podcast, you were fairly relaxed about the update. I think you said that if your targets are set correctly anyway, which they should be, you might not need to do anything, which, you know, fair enough. But now we're a month past the change. Has that held up for you? Have you noticed any unexpected surprises or anything like that? That was so blase of me to say that, wasn't it? Like, how cocky am I going into that big change saying something like that? But Respect. largely, as I mentioned earlier, performance performance generally across the board has remained relatively unchanged, off the back of that from what we've seen. Obviously, Heidi has mentioned working on some smaller accounts that, potentially have seen a little bit... A a little more, performance drift from what what we're expecting. But largely having, having that that... The importance on what we talked about earlier of aligning your conversion actions with your business objectives, means that those targets should be set in line with where you want them to be and what what what the outcome that you want from them is. So, yeah, largely performance, performance differences were were were fairly minimal. The one interesting thing that, I have as a takeaway from talking to the team more broadly about it, and it's something that, I potentially could have guessed or could have seen in in in the lead up, but Google promised more consistency from this, afterwards with the changes that they're putting into place, so less volatility from that performance. The effect of having that that great consistency or less volatility is that we're finding that accounts react more slowly to changes, when you're making them to those targets. And, again, that speaks to making sure that your, targets are aligned to those business objectives because your business objectives shouldn't really be changing on a daily basis, but everyone's got those clients that wants to change the targets, push the spend, pull the spend back, etcetera, etcetera. But I think that's something to take away and bear in mind that with greater consistency, with less volatility comes slower reacting accounts or slower reaction to those target changes, and that's what we're seeing more broadly at the moment. Nice. Interesting stuff. So on the screen here, we've kinda got the three stages of this update. We've got the advice beforehand after Google announced it. In the middle there, we've got the predicted impact from optimizer. And then on the right, we've got some much more recent results from Mike Ryan, who was showing that advertisers, are raising their ROAS targets kind of above the board after the update went live. So let me pass over to you here, Heidi. Before the change, you were kinda given the same advice as Aaron Levy from optimizer, up on the screen there. And that was to start nudging CPA targets down and ROAS targets up. So, obviously, as I've said and we we've said about a thousand times now, it's still very early days, but have you seen any signs of budget limited campaigns drifting back to their stated targets the way optimizers modeling predicted, or has the actual impact been kind of surprising to you in any way so far outside of that? No. Well, you know, our strategy with our clients has been might be different than what, others are doing. But for all of our budget limited campaigns, we weighted each current target and try to understand how different it was from the last thirty day convert or the last conversion cycle. And if it was a lot different. Right? Because as you had just mentioned in your example, you could set a target and you could potentially get a better a much better result. Right? And so we were trying to, audit what was currently set as the target and then what the actual performance was, and then we actually we changed the target to match closer what the the real the net performance was. And with those instances, we saw very few changes in terms of, performance. We were still able to meet our target. Again, I think it's a little different for smaller accounts that maybe have less volume of conversion data. So maybe they don't have a full cycle of conversion, enough conversions coming into what a full cycle conversion data would ultimately look like. So there's kinda guessing what what it is, or, you know, what what a realistic statistically significant target should be. So in in those cases, we see see a little bit of volatility where people are like, oh, gosh, we're all over the place. And in that case, it's kind of, you know, half guessing based on, your intuition as a specialist and then also looking at what data is available to you in the account. But I think, you know, in general, Mike Ryan has some had some very good insights here on the right for sure. I think a lot of people in the industry, paid attention to that for sure. Definitely. Yeah. And I do just wanna take a closer look at what he said there because, I think it caught a lot of people's attention for running very much against the grain of that initial advice. So as we saw on that last slide, Mike's data there showed that advertisers raising their target ROAS, they're raising them basically across the board, which is the instinctive move. But on the smarter ecommerce podcast, link in the docs tab, Mike spoke to Chris Starmuela. I have no idea how to pronounce his last name. He's called Chris. But he stated that the smartest thing you could probably do right now is lower your ROAS target in order to capture market share while everyone else is pulling back. But I just wanted to we're running quite low on time, so I just wanted to pitch this question to both of you. Whoever has an answer, feel free to shout out. But now the change is live. Do you actually agree with Chris and Mike here, and do either of you have any other kind of early doors tips outside of what's already been said? I think it's I think it's interesting. Right? I think a lot of people like to, consider a strategy like this based on their situation. Certainly, if you, lower your ROAS target beyond what you've typically experienced in real time as your result, you're going to probably get more volume at a little sacrifice some efficiency. If that's the wiggle room that you can afford in in in testing something like this, for your business, you should I I'd say I'd advise you to do that. I'd also advise you to do another thing. If you would like to fool around with your ROAS targets or CPA targets, do an experiment instead of just changing everything. So do an experiment if you have an eligible campaign that can do that, and then, give a portion of traffic to the experiment to work against your proposed target that you want, that you're hypothesizing could get you either better market share or better results. And that way, if you need to, pull back, right, from your from your choice, it's not really impacting the core core campaign. Awesome. Scott, Yeah. same sort of thing? I think that the keyword and everything Heidi just, Heidi just said there is to test. I think the advice from, Mike and Chris is a little too broad, and you wouldn't just take it at face value. I think I think you have to test and understand the impact of that, not just, blindly lower. So I'm not... I don't... Not suggesting that that's what anyone does, but, yeah, I think that that advice is a little too broad, and you need to look at a case by case basis. And as Heidi says, test test test. For sure. Yeah. Wise words. Cool. So in the last couple of minutes here, I wanted to wrap up the session with a super quick next steps guide for anyone who kind of, wants to start optimizing their smart bidding efforts based on what we've spoken about today. There's three things on the slide here. The first one is to start by auditing the last ninety or so days for any bad or corrupted data, then apply exclusions where you find data to be wrong. And after that, set up alerts using the scripts or either the scripts provided in the docs tab or anything Scott spoke about earlier to ensure you catch those anomalies before they get out of hand and start teaching your smart bidding the wrong things. And, of course, please do take a look through that docs tab for all of the resources and tools mentioned in today's session. But before we move into the q and a, let me quickly pass over to both of you here. Is there anything else you'd add to this list that you'd recommend Mark to go and look at? Or do you have any other final thoughts that maybe we haven't mentioned today, before we move into the q and a? I'll just quickly start with you here, Heidi, and then move to Scott. Yeah. No. I think this is solid advice, for sure. You know, I and I just think, now more than ever when our campaigns are running on bidding models that do use signals, quite heavily and historical data patterns to determine who they're going to be bidding for, It really is worth a resource in either your agency or in house team to have a team of people or an expert dedicated to making sure that there's auditing going on consist consistently, not just when you suspect there's a problem, but it to to ensure that these problems don't go on for too long. Because the worst thing that happens is when I get a phone call from someone who's like, oh my gosh. You know, we've been running on six months of of what we thought was valid statistical data. We were feeding the algorithm, and it's it was totally wrong. And and that strategy usually has, is different from what I'd advise someone who, you know, had a couple weeks of this happening. Nice. And then Scott? Yeah. I would just echo that. It's not just about, ordering for those spikes and errors as they're happening. Have a spring clean, get in there, and and and check out exactly what's going on in your account and get that tidy up. Perfect. Great stuff. So that neatly wraps up the content to to for today. So let's move into the q and a section. Got around five minutes for this, so let's try and get through as many questions as we can, kind of quick fire style. If you haven't submitted a question yet, this is your last chance to do so. Just pop it in the q and a tab. I will be starting with the most upvoted first. So this one is from Daniel Rostron. I'll share it on the screen here. If you had different services and service lines as a lead gen, would you still only have one core lead conversion action which tracks the lead submission across all of them, or would you set up separate actions for each service? Does anyone wanna take this one? I'll go. I have a thought. Oh, the... go, ahead. no. You you, go hide. it. Well, I'll go just wanna mention something really quick. So let's let's think about how Google Ads has evolved over the years. I would say my answer would have probably been different, seven, eight years ago than it is now. So, really, it depends, Danielle. And, Danielle, it depends because if you have different services and service lines that each have their own conversion action, but there's very few conversions coming into each. Right? That it might be worth putting them in one conversion action if it means you can give similar signals, right, to the algorithm to work at the campaign level so that it can be stronger a lot quicker. So think about it in terms of what how much volume each conversion action is putting in, if it could essentially be combined into one conversion action. And you can even have, the ability to assign multiple conversion actions as, you know, equal same weight to one campaign so you can still see what they look like broken out. But, you know, in general, let's try to consider sitter consider how your setup would also assist the algorithm in finding better customers for you. Nice. Do you have anything to add to that, Scott? Cool. Yeah. I would just say on that point about you can you can always apply multiple conversion actions to a single campaign. On that point, I would therefore split it out into separate, services because it means that, yes, you can have them together, but if they were all as one conversion action, you couldn't split them back out should you have the business need to do so. So say, for example, you wanted to run a campaign or have an individual budget, the only target is a specific service line and therefore your measure of success was that service lines conversion action, you can't split that out if you had more conversion action, whereas you can always roll them back up together if you have multiple conversion actions. So, yeah, I would, I would absolutely just be splitting those out into services if you have the technical ability to do so. Awesome. And then this next one is from Carl Rodell. Should we change or broaden our bid strategy when engaging data exclusions? So I have an opinion on this. I would keep it consistent with what you, I would keep it consistent. I would not I would not change it because, two things. You're gonna make two big changes at once. If you do that, you're applying the data exclusion, right, which is meant to help you. And the data ex as long as your bid strategy was working relatively well for you before at the target that it was set at, just let it continue to train on that with the new, fixed data coming in. And then if you have to make the adjustment, wait till after the, period in the account is stabilized. Right? So your your baseline is stabilized and then start optimizing as you normally would. But try not to change anything else after you do the initial data exclusion, at least at Nice. Great stuff. And then. the next one here is from Elliot Venus. If a conversion tracking issue lasts over a month, would you put in multiple data exclusion periods of fourteen days to cover the entire period or just put in one fourteen day exclusion period? And if just one, which period would you cover the most recent excluding attribution lag? I think we touched on this one earlier, but I think Heidi has just dropped off. I noticed she's having some Internet issues earlier. Scott, do you have any opinion on this at all? Yeah. I think this is, obviously a super niche example because there's not many clients out there that'll have a conversion tracking that lasts over a month. But, yeah, if I was to put that fourteen day exclusion in, I would put it in the the most recent day because that's gonna be most heavily weighted in the in the algorithm or what Google's looking at. Perfect. Alright. And that just about concludes the session for today. Huge thank you to you, Scott, and Heidi. God rest ourselves. And, yeah, big thank you to everyone who joined us today. Watched the session. Please do go ahead and follow both Heidi and Scott on LinkedIn if you're not already. You'll find links to follow them both in the docs tab. And, yeah, this session will be available on demand if you want to recap on anything we've covered here today. And finally, yeah, just a big thanks to all of you for tuning in either live or watching on demand. We really do appreciate every single one of you showing up, helping build a community around these webinars. And, yeah, I hope to see you again at our next one. That will be at the end of October, and it will be an AI Mac auto upgrade survival guide. We've got some excellent guests lined up for that one, so please do stay tuned and keep an eye out for sign ups. There will be more details to come very soon. Yeah. In the meantime, take care, and best of luck with your smart bidding.