PPCChat | Navigating Ever Advancing Automation in PPC

Posted March 6th, 2024 by Amita & filed under PPCChat.

This week on PPCChat, the focus was on ‘Navigating Ever Advancing Automation in PPC.’ Host Julie F Bacchini was curious to hear from PPC professionals about their thoughts on automation, especially with the recent launch of new automated solutions by the Google Ads Developer Blog.

Q1: How far have you waded into using automation in your PPC practices? Dipped your toes, in up to your knees or swimming in the ocean of automation?

Up to Knees, I’d say. Since we have clients across many different verticles, automation that scales across markets has been key so far. @Realicity

Swimming! I used GPT-4 in October/November to create quite a few scripts with email alerts to be sent daily. @alimehdimukadam

I’m also in the up-to-my-knees camp. (Kid has been talking about the beach nonstop lately, hence the wording of this question!)The types of accounts I typically work on struggle when fully automated. However, I also realize that automation is not moving backwards and will, at some point in the not-too-distant future, be the only option. So, I am trying to learn those new tricks as an old dog…@NeptuneMoon

Knees-ish. A lot is due to automation helping our team, but we have to manage it with client comfort, and of course, public sector rules. @JuliaVyse

Q2: How would you describe your general attitude about PPC automation?

Apprehensive

Bidding is better than manual – Usually, given enough time
Ad Copy Creation is cringe
Ad Copy Ideation is helpful
Labelling is helpful @Realicity

I have VERY mixed feelings about it. I like the idea of it, but the reality does not match the conceptual in practice. At least it does not in the accounts I work on, which are lead generation. Many of them are low conversion volume accounts and automation HAS NO ANSWERS AT THIS TIME FOR THOSE ACCOUNTS.I am concerned that so much is built with big ecommerce first (as the who is this for). And so many, many businesses are not of that mold. @NeptuneMoon

These machines are not as smart as they or their owners think they are. proceed with caution. @JuliaVyse

Rule #1: You might not need to automate that.
Rule #2: Every time I’ve set an automation “just for monitoring” I never looked at it again
Rule #3: You can’t automate an insight
Rule #4: Automate the routine @ferkungamaboobo

Lead-gen advertisers and PPCers have always had to figure out how to work in an ecomm first world, but that divide feels like the Grand Canyon these days. @NeptuneMoon

I’m worried it’ll be used a pretense by Google and/or Microsoft to keep hiding more and more insights and data from us advertisers over time. They’ve been chipping away at features and controls for a while and it’s a slippery slope. Automation could easily mean us never really fully understanding what is happening or what is/isn’t working in our accounts which is bleak. @JeremyKrantz

Great way to put this @ferkungamaboobo ! Rule #4: Automate the routine.@Realicity

@JeremyKrantz Just like Meta. @JuliaVyse

I mean, I think part of the challenge is that so much is “Automation” – custom scripts changing bid strategy based on weather is automation. So’s the core understanding of placement and bidding. @ferkungamaboobo

That Solutions announcement from Google Ads is unsettling. The SEJ piece says “Google believes this decision will provide the best experience for advertisers and avoid duplication of efforts, as Solutions will become the primary tool for campaign management and automation.” (emphasis mine)I have asked SEJ for clarification as to if this came from Google or is their interpretation. Cause if it came from Google – YIKES ON BIKES people.[I linked to both in the thread of the first post of this chat] @NeptuneMoon

Automation also makes people lazy and they don’t check on accounts. @josh

Suffering from shiny object syndrome from time to time helps me be aware of the latest. I am in favor of automation that helps workflows, streamlines certain optimization tasks, reporting, etc. Not at all in favor of auto-apply automation. Example – automated classification of search terms by intent > Manual review > add or exclude auto-apply keywords or upgrade to broad. @alimehdimukadam

See, I dunno. I love the auto-apply to pause redundant keywords. @ferkungamaboobo

The other flipside is that almost all automated recommendations in Google are complete and utter garbage. So I don’t think Google is as close as they think towards automating things because if I applied 90% of those, they’d end up doing more damage than good. @JeremyKrantz

Or like, I read the auto-apply keywords list every time it came out. @ferkungamaboobo

@NeptuneMoon In my opinion, they have simply renamed existing Google scripts functionality to solution as that is what I have seen so far in the documents. @alimehdimukadam

@JeremyKrantz I’m about 10 seconds away from all my public sector partners turning off Google entirely because we can’t run anything auto-created. it’s just not feasible. @JuliaVyse

And like, I won’t lie, I love enhanced bidding. @ferkungamaboobo

@ferkungamaboobo isn’t that a “Remove” redundant keyword feature instead of Pause? We almost never remove as we want to know which keywords have been manually/humanly approved. @Realicity

@alimehdimukadam Yes, it is scripts based, but if the statement in the SEJ piece was paraphrased from Google Ads, then I have questions…@NeptuneMoon

Yup, a primary tool for management & automation. That is a little concerning. @alimehdimukadam

It is when it gets into auto doing anything that I get off the automation train. Show recommendations? Ok. Make assets (that are not active without approval)? Ok, Do anything automatically, without expressed approval? NO THANK YOU @NeptuneMoon

I’m okay with it trying to come up with stuff, just not okay applying it without a review/consent. @JeremyKrantz

Can we also talk about how Google talks a big game about “brand safety and suitability” but the tools for this are still bludgeons? @NeptuneMoon

@Realicity It might be. Have used it both ways. @ferkungamaboobo

Attempt at a joke“Do you want to know how the world looks like happens there is complete automation? Check a Google Ads account with auto-apply recommendations onIt’s not a world we want to live in. @alimehdimukadam

I mean, I’d take one with auto-apply over one that’s run like it was 2009 any day. @ferkungamaboobo

I’d take 2009 over autoapply. @revaminkoff

Actually, what I used scripts most for was budget control. I forget the reason why even after monthly budgets were introduced, it just made more sense to have an hourly budget checker. @ferkungamaboobo

We also have to remember that we are not Google’s target audience for anything. Shareholders first.Direct to businesses second. The sales pitch of “run a fully automated ad strategy on our network that reaches potential customers at the best moments” and addresses data or privacy concerns? That is a pretty easy sell for lots and lots of orgs. Save on management fees and let our experts handle it – for free! @NeptuneMoon

I mean, but also, I think that easy sell is made easier by a lot of our sales pitches. I’m always shocked when someone trots out a 20-year-old line in a sales pitch. @ferkungamaboobo

Had to turn this into a poll, btw: https://www.linkedin.com/posts/revaminkoff_ppc-ppcchat-googleads-activity-7170833837495664642-hJLJ?utm_source=share&utm_medium=member_desktop @revaminkoff

Q3: Where do you use and/or plan to use automation the most in 2024? What has influenced your decisions about how and what to use?

If I had the option, I’d use it most in billing! but our preference seems to be to have machines make our art and music while people fill in forms. @JuliaVyse

Bidding, Ad Copy ideation, Image Ad Creation. @Realicity

If platforms like Google Ads make functions that were previously script based (you had to find the script and put your info into it and then upload and test it) I will be all over that! @NeptuneMoon

@NeptuneMoon I am so surprised they haven’t done that yet.@Realicity

I have played around with using AI to:

Give me copy ideas
Tell me what my client’s site is about
Tell me what their competitors’ sites are about @NeptuneMoon

@Realicity I think that is what the Solutions stuff in the links is? I need to dig around some more. Or at least some of what it is? I’m not totally clear after reading both pieces. @NeptuneMoon

Ad Copy Creation, Media Planning, Audience Research, Geo-Targeting Research w/ Population Count, Keyword Research, Competitor Research. @JeremyKrantz

In the PMax Answers piece from last week, Google Ads also claimed that their AI will “never create the same asset twice” which I found fascinating. Here is the link: https://support.google.com/google-ads/answer/14587068 @NeptuneMoon

@NeptuneMoon let me rephrase- I’m surprised they didn’t do that 3-5 years ago. @Realicity

I think it’s for the same reason that conversion rates are really hard for Google to use in their ranking algorithms: too much noise, too much specificity. @ferkungamaboobo

Still try not to use automation outside of some of the bidding (and will take automated billing any day!!) @revaminkoff

Oooh, I forgot G Ads help has an AI support assistant! I see it on the page I just linked. I will try that out and report back too…@NeptuneMoon

Well like, again. How many Headlines can one really make? Do you really need LLMs to say “Tallahassee Interstate Movers” in less than 25 characters? @ferkungamaboobo

@ferkungamaboobo Tallahassee Interstate Movers is 29 characters. @JeremyKrantz

I just want them to tell me which one is working!!! @revaminkoff

(Versus making me run all of them and then still making me run all of them and then saying why aren’t there another 15 that you’re using too?) @revaminkoff

@ferkungamaboobo Also, let’s be real…Google might want you to create a bazillion headlines, but it will find the 4 it likes and use them over and over and over. @NeptuneMoon

Automations – reporting, alerts, analysis & insights, copy ideas, anything that requires repetitive manual work and can be done with a standard checklist or SOP. @alimehdimukadam

I have been letting go of the idea that we can actually test much of anything over the past 18 months. @NeptuneMoon

( Acktshually from an old study I ran: @ferkungamaboobo

Finally, including “Now” or “Today” in the headline seems to turn off users from clicking, to the tune of a .97% drop at a 98% confidence. At lower confidences (90%), including the word “Now” in the body copy drops CTR by .33%.

@ferkungamaboobo Sample size? @JeremyKrantz

@revaminkoff and @NeptuneMoon I’m come to believe that of those 15 headlines, 14 of them are just supposed to be static Keyword Variations to replace DKI. @Realicity

16K ads, – only sampled the ones over 1000 impressions. @ferkungamaboobo

Copy all the recommended keywords to add to headlines/descrips when building an RSA in the UI and paste them into GPT and tell it to find a way to use all the words in 5-10 headlines & 3 descriptions. Then edit them as needed and paste them into the RSAs, higher good/excellent scores! @JuliaVyse I agree with the billing. Reporting/Analysis is a big one too.
I recently used it for data enhancement on a customer list that was missing zip codes but had city & state. I’m of the mindset that I’ll use it for ANYTHING…until it proves to me it can’t handle it. One thing it can’t handle a lot of the time is large data files. Greater than like 1000 rows sometimes. @timmhalloran

So not a publishable sample size. @ferkungamaboobo

In the future we will just enter some acceptable phrases to have in headlines and Google will do the rest. They are already playing fast and loose with what a headline even IS with their nonsense where they might only show one and jam another one at the beginning of the description area. @NeptuneMoon

It’s funny for someone who literally just posted from a granular ad copy study – but I don’t think search ad copy matters all that much tbh. @ferkungamaboobo

Beyond hitting the right beats @ferkungamaboobo

Like if you have bad copy, that’s one thing @ferkungamaboobo

Now I think it will get really interesting @ferkungamaboobo when more and more and copy is AI generated – either in platform or via a ChatGPT. And it all SOUNDS THE SAME. Still writing your own could pay huge rewards in that scenario! @NeptuneMoon

It’s a big issue in regulated industries though – the more Google forces AI, the more I expect some of them won’t be able to run? @revaminkoff

Legal has to sign off for a few of our clients (and for a reason) @revaminkoff

Q4: What are you stubborn about when it comes to automation in PPC? What is the reason for your stubbornness on that aspect?

I will not allow anything to auto-apply. EVER. One simply needs to peruse G Ad’s recommendations to see why allowing auto application could be a disaster. @NeptuneMoon

Trust.Seen so many “Automated Recommendations” that are horribly off. @Realicity

Anything about increasing budgets. Google, we signed an MPA. we can’t just increase like that. @JuliaVyse

I have a hard time releasing my ad groups from phrase/exact, even if it says to me, “Yo that’s already covered with your broad match” — My brain still goes “…..yeaaah but I’ll bid less if it matches for the exact match so Im’ma keep it okay? Okay.”@timmhalloran

Dang everyone stole all of my bullets. Bravo. @JeremyKrantz

Sense checking/brain power — the AI just doesn’t get it — still more often than not. @revaminkoff

Stubborn about not auto-applying
And not going by the book or completely rule-based, automated every time. Reason – It skips nuance and may even lack context. Similar to our fears of AGI and the recent Gemini fiasco. @alimehdimukadam

@timmhalloran Exact is what BMM was 5 years ago and Phrase is what Broad was then. Broad today is just Topics with an Inigo Montoya “I don’t think that means what you thing it does” GIF. @Realicity

Related to that query filtering. It’s really hard to not filter down based on match types. But when I’ve tested full broad vs. query filtered exact, phrase, broad – most of the time, the broad ended up working better (as long as I had a solid negative list). That’s been hard for me to stop doing. Also, overbuilding – especially with multi-location clients. It makes so much sense to break out locations by campaign, but nowadays you’re going to be stuck in a perpetual learning limited, especially with B2Bs that don’t have the conversion volume to get out of it. So I’ve been consolidating a lot more where I can, and using DKIs when I don’t trust the automation to match to the correct location if I’m forced to consolidate two locations where their distance worries me. @timmhalloran

Here is another question I have regarding in platform AI – will it take strong-performing assets from competitors and use it in your ads and vice versa? Will it just crib off of what works in one account and slap it in all the others?????? @NeptuneMoon

@Realicity Broad match in 2025 = “Any contextual signal connected to said keyword or topic allowing up to 6 degrees of separation from your stated intent.” @timmhalloran

There are aspects of this that are so therapeutic to hear from not just me.@revaminkoff

@timmhalloran Hopefully, automation will evolve to be able to allocate budget based on locations but within a single campaign. Because right now, a lot of things get split out for budget control on location A, B, C. @NeptuneMoon

@NeptuneMoon That’s my thinking – Google knows what words, phrases and sentiments are converting searches into leads and sales. @Pete_Bowen

@timmhalloran Good point on Overbuilding. We’ve implemented an Ad Group Consolidation process to help identify and address this. It was needed with some accounts that had 10+ years of build-out. @Realicity

Well, wait, we all just said an easy button is what we wanted for our clients — like would any of us be able to explain like, a linguistic eigenvector? @ferkungamaboobo

I think the Google reps are being judged on auto-apply adoption, and we were told at one point it was a criterion for Premier Partner last year – not sure about this year – but I’m with you – pretty stubborn on not adopting 95% of it. @revaminkoff

In the sense that I think “aboutness” is super hard to define in a programmatic way. @ferkungamaboobo

But LLMs! @revaminkoff

@ferkungamaboobo This is where I’d love to see some real talk about where AI and ML shines and where it does not (at least right now). Go big on the stuff it does really well and be honest on where it still struggles. Stop acting like it does everything it attempts equally well when we all know it does not. @NeptuneMoon

Well like, I’m trying not to touch the third rail of LLMs – this kind of aboutness scoring can probably be more easily done (and has been by people smarter than me) with other methods. @ferkungamaboobo

Along the lines of validation, do any of you have these conversations with yourself or your coworkers when editing campaigns:”Well we set the ROAS at 11:1 but I’d prefer not to drop it down to 9, even though I think it’ll work better because I don’t want to train the machine to assume we want to hit a 9. And then, every time I build a campaign in the future, even though I could have hit a 15 or 18:1 ROAS, the ML algo knows I’m okay with a 9, so it’ll penalize me for getting greedy…”@timmhalloran

1000% – that’s exactly the kind of weird human stuff that a good automation would handle? @ferkungamaboobo

But that’s also really hard to predict and model. @ferkungamaboobo

I think we buy into marketing materials from the ad companies too much. @ferkungamaboobo

Yeah. That’s one example but the idea that I’m working with a computer with perfect recall that knows my expectations for an ad account, and finds ways to spend the maximum amount while remaining within 10-20% of my expected return on a monthly basis so I don’t get too angry or too ecstatic – my head still wonders. But you’re right. Pretty difficult to model that type of behavior. @timmhalloran

PPCChat Participants