PPCChat | Google vs Human Automation: Why We Still Need To Be Involved | 3rd September 2023

Host Julie F Bacchini and guest host Navah Hopkins spoke about automation during this week’s PPCChat discussion. How can experts compare human automation & ad network automation? what are some areas where PPCers can trust ad network automation?, what is the role of human automation?and more

Q1: What does Human Automation vs Ad Network Automation mean to you? Do you distinguish between them?

Human automation means you’re setting the “if this/then that” rules of how you want your campaigns to function. Ad network automation are the native algorithmic changes. @navahf

This is a really interesting question… I think of it like this – human automation are things that I choose to use automation to do/assist with. And ad network automation, to me, represents aspects of the ad platform functionality that they automate. @NeptuneMoon

Case and point: I would see broad match as “ad network automation” where as exact match with audience and bidding targets would be human driven. @navahf

Love the explanations from @NeptuneMoon and @navahf above. Gotta get that human piece involved. @robert_brady

I feel like human automation is sort of like a policy. and you can have teams with hands on keyboards executing optimizations based on those policies, or you can use scripts and other tools. Typically these policies are in favour of you and your goals, while ad network optimization tends to be broad, based on scale, spend, fully machine automated, and focused on the networks goals. @JuliaVyse

It boils down to control for me. You can observe and influence ad network automation (like data inputs on Smart Bidding) but you can’t directly alter or force-stop it, like you can with alerts or rules. @TheCopyTrail

Could not agree with you more @JuliaVyse @NeptuneMoon @TheCopyTrail. Thank you @robert_brady

It really comes down to the nature of the automation. Ad Network automation is setting a real-time bid in an auction; human automation is configuring decision trees or “if-then” workflows. These two things should go hand-in-hand and be complementary; the ad network automation should handle the day-to-day management; the human automation should provide the guardrails to ensure the machine doesn’t do dumb things. @DigitalSamIAm

1000% agree – they are a partnership to ensure the business succeeds. Great point @DigitalSamIAm @navahf

I may be interpreting this question wrong but I’d define it as:Ad network automation: features and functionality within ad accounts that are automated offerings (smart bidding, optimized targeting, optimized ad rotation, automatically created assets, etc.)Human automation: automating changes to your account or camapigns based on data and inputs from your ad campaigns and other external sources such as business data (bidding scripts, dayparting scripts, etc) @justinhoffman

Curious if anyone has separate ideas of what constitutes automation vs. machine learning? I tend to use the terms interchangeably as do many others, but I’d love to know if someone has drawn a line in the sand with them.@TheCopyTrail

To me, automation means that something is done without manual input. So, some bidding strategies, how Google uses signals, etc. Machine learning is what Google does with millions of data points to supposedly drive better results in which ads show for which queries, etc. Over time, it can track performance across a bajillion variables and “learn” which combinations lead to conversions. @NeptuneMoon

Q2: What are some areas where Ad Network Automation wins and it’s reasonable to trust? Where do you only trust human automation (and why)?

I don’t have a great answer for this because (drum roll) it depends! Plus in omni, you have to just trust your partners a lot of the time. Ever bought on Spotify? there are LIMITS to what you can even do. @JuliaVyse

I really am a fan of value based bidding and feeding conversion values into the system. Our study on match types and bidding strategy found that Max conversion Values did better on CPC and CPA (which surprised me) @navahf

Discover the surprising revelations about Broad Match vs. Exact Match and bidding strategies in our data-driven analysis. Is it time to rethink your approach? https://www.optmyzr.com/blog/optmyzr-study-broad-match-bidding/ @navahf

first thing that came to mind is bidding strategy for most keywords. Sure there are some low volume/brand keywords you need utmost control over – but I reckon bidding strategy is needed for the majority of keywords to ensure you are competitively entering the auction. @TheMarketingAnu

I no longer am as willing to lean into broad match as aggressively, but I do like PMax. @navahf

With sufficient data, the automated bidding does a good job most of the time. It will still chase a wild click at an outrageous price, even well past an initial learning period, but generally, it finds a decent groove if it has enough data to work with (which is a big IF). @NeptuneMoon

Ad Networks will always be able to out-trade humans, provided the machine is given correct information + proper targets. The simple reality is that Ad Network automation has far more access to data + more data in the view than an advertiser has. @DigitalSamIAm

Bid management for sure. In any recent H2H tests of 3P vs smart bidding or manual vs smart bidding, smart bidding has won for me. Google and other ad platforms utilize data points that 3Ps can’t (such as user search history, for example), in real-time (though I am questioning this value prop given the recent revelations about Google moving the auction price lever to meet rev targets) I tend to trust human automation when there isn’t enough data to make statistically significant decisions and some intuition and human discrepancy is needed. Also if there are other external data points that ad network automation is blind to. @justinhoffman

One area where I would only trust human-controlled automation is while applying business data to campaigns. Whether through a third-party tool or something homegrown, you should have a layer in between your account/data and ad networks so they can see the adjustments to goals without seeing the underlying data. @TheCopyTrail

You must trust your conversion tracking to lean into automation – if you don’t, humans will have to take the wheel. @navahf

I also think that human intervention to put guardrails in place is still critical. I do not trust Google to take all the learnings in an account history and use them properly when I give it permission to “do its thing.”I know G wants us to just let them do their thing, but I can’t in good conscience just let it rip. So, I will input negatives we have figured out over time, etc.  Despite what they say, there is no real good reason to have the machines start at zero (meaning not putting any guardrails in place). @NeptuneMoon

I’m so glad you made that point – exclusions (audiences, negative keywords, etc.) are how we teach the system. @navahf

And each system learns differently. I can’t tell you how many start-fresh campaigns Google has to be re-taught our never-in-any-context-ever negative keywords/placements. @JuliaVyse

@navahf While it is certainly an argument for why PPC pros are very much still needed, it is not just self-serving. The machines are very good as some things and incredibly dumb at everything else.I think our value will be in taking our clients from being one of x advertisers all in the big Google Ads stew and putting them into the game on second base! @NeptuneMoon

Yes – every network has different rules of engagement! @navahf

Letting a Google Ads campaign work without exclusions and guardrails is like telling a creative team to come up with a campaign with no brief. You will get results, but they won’t be the ones you want and you’ll blame the wrong people for it. @TheCopyTrail

so like…PMax? @JuliaVyse

Q3: Are you currently feeding conversion values into ad networks? Why or why not? And what method(s) are you using to send conversion data to ad networks?

Yes!!!!!!!! And we do this through crm integration. @navahf

only where possible. In cases where conversions aren’t possible, we have a lot of black box conversations. @JuliaVyse

Shameless plug for Optmyzr and how we enable our customers to feed that conversion data into their ad accounts. @navahf

Not the easiest to do for lead gen… and as a lead gen PPCer, let me take this moment to say that the current platform automation is very geared for ecommerce.Lots of lead gen businesses just don’t have the internal infrastructure to send lead data back. And, many are uncomfortable doing it too. @NeptuneMoon

Yes. If you’re not, you’re behind. @DigitalSamIAm

@NeptuneMoon Some areas are way behind here. Higher ed in particular because admissions is usually in a separate system from lead gen/recruiting efforts. @robert_brady

@robert_brady Yes. And the assumption that every business uses a CRM that can just easily export sales data is kinda hilarious. @NeptuneMoon

Not yet, but I’d like to. The biggest obstacle for us right now is that we don’t have an internal framework built to properly value leads based on the volume of leads we have. The volume just isn’t there to score them and it’s very noisy. If we get to a stage where we can definitively say that leads meeting xyz criteria are worth x value, we’ll definitely get this in place as I think it’s the gold standard of bid management. It’s tough for lead gen. @justinhoffman

I have heard that objection quite a bit @Justin Hoffman – but I think there’s a lot of misconception here, especially around the system. I have described this to clients like Roulette. Yes, you’d love to know exactly where the ball will land. That’s the ideal. But the simple reality is that getting 20% more accurate (i.e. weeding out just the spam leads that are easily identifable) provides a strong signal that can improve the overall quality of your leads from Google. @DigitalSamIAm

That makes sense @DigitalSamIAm. Our approach now is to tweak the signal that we’re sending to Google based on quality factors but it’s much more boolean since we’re not using a value-based system. We don’t optimize towards all leads, just those that meet xyz criteria. We could bucket leads meeting certain criteria and assign values based on lead to acquisition rates (with the understanding that there isn’t any stat sig) and move to max conv value, but I wonder if that would produce the same outcome as just adjusting the signal and using max conv? I’m curious if you’ve deployed this for a lead gen business and if you still applied some value to raw leads not meeting any baseline quality criteria (raw form submissions, for example) or if the minimum conversion value threshold still has some baseline level of filtering (such as a business email address)? @justinhoffman

So, my overarching philosophy has been: (1) spam/garbage leads have no value. You can even re-post the value of those conversions to zero (which we do) using value adjustments. What I’ve found in doing this is that, within ~100 leads (usually about 2 months for our clients), the percentage of spam/agency leads falls off substantially. In one account, it went from ~30% to ~5%. The total number of leads generated didn’t move much, but the quality was much better. I think bucketing makes perfect sense. @DigitalSamIAm

Q4: Which type of bidding are you using? If Smart, which variant and why? If manual, are you layering in third-party scripts/rules?

Max Conversion Values where I can. Max Click with a bid cap where I don’t have conversion volumes. @navahf

It is not reasonable to opt into smart bidding without conversion thresholds. @navahf

For shorter campaigns Max Clicks is where it’s at. For longer campaigns with a conversion, max conversions. but it’s a real challenge in public sector. @JuliaVyse

One thing I found really interesting about Optmyzr customers is that 66% use smart bidding to a certain degree – while 12% use manual. I am still shocked at the manual percentage but that’s why manual will never die. @navahf

I generally start with Max Clicks to get data flowing.I do try Max Conversions too, but with lower conversion volumes, it struggles. And to be clear on that threshold, in my experience, fewer than 50 conversions per 30 days and the machines struggle. I know G “removed” that threshold but in reality it is still in play. @NeptuneMoon

I would love to see everyone who has struggled to use max conversion values “converted” to it in 2024. @navahf

@navahf So are you using Max Conv Value without a target? If yes, how do you prevent the algo from chasing unprofitable conversions in the name of maximizing value? @robert_brady

No, I still put in goals – realistic ones. @navahf

Can we also talk about how most clients have no idea what their target CPA (cost per acquisition) should even be? I love how it’s taken as a given that all businesses just know this and can plug that number right in. I’m not talking about tiny businesses either. @NeptuneMoon

I often have to walk people through the math of their funnel to get a target CPA. @robert_brady

Q5: Have you tested broad match in Google Ads in the last 12 months? What were the results of those tests? Were you surprised by the results – why or why not? If you have not tested Broad Match in Google Ads lately, what is keeping you from testing it?

Yes. And it’s been at least that long since I had a broad match campaign win the experiment over my usual phrase/exact setup. Not surprising with the expansiveness of Broad and Google’s need to drive their own revenue. It’s an easy dial for them to change. @robert_brady

I can’t justify it when conversions are not at a volume where the machines can find their groove. Add to that the increasing lack of transparency of the queries that are triggering the ads and the antitrust revelations about Google’s auction shennanigans and it will keep being a tough sell for many lead-gen businesses. @NeptuneMoon

Yes, and it’s worked for me. But I’m in very specific circumstances. I primarily run campaigns based on conversations happening on fairly broad topics like flu shots and burgers. Broad + audiences work really well in cases like that, and not so great in other circumstances like lead gen. @JuliaVyse

This is where I got burned. I see Broad match do ok occasionally, but the forest view saw Broad match lose when compared to exact even factoring in volume. @navahf

Yes; you have to test it at this point. Right now, our priority goes something like EM < BM < PM. There’s not a lot of compelling reasons to use the new Phrase Match at this point. @DigitalSamIAm

I have not seen great results with broad match for my clients, who have all been B2B SaaS for the last couple of years. Something I’ve done to mitigate the non-relevant searches that match to my keywords while using broad match is restricting the audience targeting to Targeting, and selecting relevant in-market and industry audience segments (and remarketing lists). @kytaylor88

I predominantly use a mix of phrase match (since it’s effectively BMM at this point) and exact match. @kytaylor88

BM + Smart exclusions + Automated Bidding works very well in certain spaces; for more precise or term-specific areas, EM dominates. @DigitalSamIAm

PM is just a terrible middle ground. @DigitalSamIAm

Seeing broad match lose on CPC, conversions, CPA etc was a wake up call that Exact can drive value and you don’t need to pay a premium for value.  @navahf

Phrase match is broken broad match.  @navahf

You know how Facebook has the % match on LAL audiences? And that you can choose like 1% or 10%?I feel like Google has that on broad match. They can adjust it to be like 75% close enough instead of 80% semantically related. And BOOM! millions more dollars of ad revenue appear. (while hiding the crappy matches in the “Other” section of the search terms report) @robert_brady

Long-tail phrase match has done mostly good for me, not perfect but definitely better than true broad. @kytaylor88

I’m talking keywords that are 3 or 4 words or longer that include intent terms. @kytaylor88

Google’s pitch on broad match is full of hubris the level of Icharus, honestly. To tell businesses with a straight face that all of their knowledge gained through in many cases YEARS of advertising and finding the best converting terms should just be set aside is laughable. And when the machines still don’t distinguish between things like “milk chocolate” and “chocolate milk”? Come on.I had a client with an airport city abbreviation in their company name. Know how many airport queries we showed up for (even with phrase and exact match) A TON. @NeptuneMoon

At this point, PM has none of the upside of BM and none of the control of EM. It’s just bad @DigitalSamIAm

Yes, using it now. Couple of things I’ve noticed.

• Search term relevance and budget are a sliding scale in that as you increase budget, relevance deteriorates for BM
• Spend on search partners as a percentage of overall budget tends to increase as you increase your campaign budget for BM
• Lead quality as a whole is lower, despite optimizing towards the same conversion signal as other match types and driving similar converting search terms
• Pre 2023, broad match was useful for picking up long tail searches that if you were directly targeting as keywords would result in low search volume status and not be eligible to show (not sure if this threshold exists anymore or not, but any term that receives <15 searches over a 30 day period is automatically given that status, and so it’s a privacy issue to target that level of granularity)
• I think exact/phrase close variants can probably achieve the same thing, and I’m thinking of ways to test this. @justinhoffman

I guess I’m just a phrase-match apologist. I was always a heavy modified broad user in the past, the primary switch to phrase back in 2020 and then dealing with the sunsetting of mod broad never really phased me. @kytaylor88

PPCChat Participants

• Navah Hopkins @navahf
• Julie F Bacchini @NeptuneMoon
• Robert Brady @robert_brady
• Julia Vyse @JuliaVyse
• Ashwin @TheCopyTrail
• Sam @DigitalSamIAm
• Anu Adegbola aka PPC Live UK founder @TheMarketingAnu
• Kyle Taylor @kytaylor88
• Justin Hoffman @justinhoffman

Related Links

• Types Of Negative Keywords List One Can Create For Amazon Ads
• Duplicate Keywords Report For Amazon Ads
• Top Bing Ads Search Parameters That You Should Track