YT YouTube2Text

YouTube2Text → Transcripts → Lessons From 20 Years in PPC: How Agencies Should Prepare For The Future | PPC TH 117

Lessons From 20 Years in PPC: How Agencies Should Prepare For The Future | PPC TH 117 — Transcript

by PPC Town Hall · 9,178 words · 1,298 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hello and welcome to another episode of
0:02PPC Town Hall. My name is Fred Vales.
0:04I'm your host. I'm also CEO and
0:06co-founder at Optimizer, a PPC
0:09management tool. So for today's episode,
0:11we have someone from Optimizer, but he's
0:15not been with Optimizer for that long.
0:17He has been in the industry for a very
0:19long time though and he's a brilliant
0:21mind who's been at a lot of agencies
0:24who's managed a lot of PPC accounts and
0:26really knows how to make things work in
0:29this constantly evolving landscape where
0:31the tools change the team structures
0:33change and recently of course artificial
0:36intelligence is also stepping in and
0:38trying to do a lot of the work that
0:39we've been used to doing. So Erin Levy
0:42is my guest today. And with that, let's
0:45get rolling with this episode of PPC
0:47Town Hall.
0:52>> Erin, welcome to the show. Good to uh
0:54see you. We uh we see each other all the
0:56time and good to see you on the podcast.
0:58>> I was going to say thank you for having
0:59me in apparently my own studio.
1:03It's been exciting to uh to get to know
1:05the team and get to know the product and
1:06get to work with you a lot more.
1:08>> Yeah. Um likewise. very happy that you
1:12uh ended your little break from PPC and
1:15decided to come back and we spoke and
1:18here we are and one day yes you'll have
1:20to be in studio in person right now
1:22you're remote still on the east coast
1:25>> yes
1:25>> but um
1:26>> yeah I mean so you've been in the
1:27industry quite a long time you've been
1:29on the list of the most influential
1:31PPCers you've given keynotes at all of
1:33the events that you know the audience
1:35has probably been to like SMX Hero
1:38um for those people who haven't had the
1:40pleasure of meeting you, who may be
1:42newer in the industry, tell them a
1:43little bit about where you came from and
1:45how you got here.
1:46>> Yeah, I think I think like most people,
1:48I fell into the industry kind of by
1:50accident. Um, so I've been I've been in
1:53paid search for oh, this hurts to say 20
1:56years now, 19 years, something like
1:58that. Um, I fell into it. I was looking
2:01for uh I was being an overachiever and
2:03looking for a a summer job in November
2:06for some reason when I was in when I was
2:07in school and wound up landing a
2:09full-time co-op. And so I got to help uh
2:12DuPont, the chemical company, get their
2:14search program off the ground when I was
2:1619, 20, one of those. Um and really
2:20really liked the power that it had and
2:21how effective it was and frankly that it
2:23was new because me a little intern was
2:25getting attention from the seauite which
2:27was awesome to me. Uh and then from
2:30there I went to a couple of agencies. I
2:32worked at the agency that is now
2:34Fanatics. Um it was an ecom platform. So
2:36we got to manage both sides of the
2:37business. Uh then I worked at Seir for a
2:41long time. I think
2:43I was somewhere around employee number
2:4412 or 13 at Seir. And I'm probably best
2:47known for spending almost 10 years at
2:49Tenuity.
2:51Um formerly known as Elite SEM when I
2:53joined it. And so while I was there, I
2:55ran the paid search team and my team was
2:58at its peak probably about 115 people.
3:01Uh and then as you mentioned, I took a
3:03break. Uh so I took about a year and a
3:07half off to play a lot of golf as you
3:08can tell from my hilarious tan lines. Um
3:12do a little bit of learning, do a little
3:14bit of gardening, do a little bit of
3:15furniture rebuilding, and not enough
3:16home maintenance. But uh then, you know,
3:19you and I had talked almost as soon as I
3:21took a break. and graceful as you are,
3:23you gave me my space, but then when it
3:24was time for me to get back into the
3:26working world, I called and here we are.
3:29>> Yeah, that's great to have you on the
3:30team. Um, and we'll be traveling
3:32together to Hero Conference San Diego.
3:35So, anyone who's going to be at that
3:36show and wants to reconnect with Aaron
3:38or see myself will be there. Um, I'll
3:40also be in London in a few weeks for SMX
3:44London where I also be doing a master
3:46class and Erin's graciously helped me
3:49prepare some of those slides as well.
3:51Yes,
3:51>> but Erin, you said something interesting
3:53and uh let's jump into this. So, you
3:55were 19 years old, maybe 20, but young
3:59uh green behind the ears, and you
4:02stepped into a big company, DuPont, and
4:04you helped him figure out SEO. So that
4:06makes me think of uh Chuck Robbins who's
4:08the CEO of Cisco and just this last week
4:12he was asked whether he was going to
4:15reduce staffing due to AI because a lot
4:17of people
4:18>> uh um like Amade from Anthropic and
4:22people like Sam Alman are basically
4:23saying like listen a lot of people are
4:25going to be put out of their jobs. Mark
4:26Zuckerberg says the same thing about
4:28marketers. Um, but Chuck takes a
4:30complete opposite point here and he says
4:32like he actually loves having meetings
4:34with interns and and younger people
4:36because they tend to be the ones
4:37learning these new technologies in
4:38school. They're on the cutting edge of
4:40how things are actually done. So they
4:41they usually come in without bias and
4:44with a fresher perspective and they help
4:46the company move in the right direction.
4:48Um, so I think that's super relevant to
4:51the industry that we're in and you
4:53having been at so many agencies like
4:55talk to us about how you educate someone
4:58who comes in fresh and how you think
5:00that's going to evolve in this new
5:02landscape.
5:03>> Yeah, I think that's a really good
5:05question. Uh, certainly when I started
5:07there was not uh quite so much
5:08documentation. There was not quite such
5:10a robust blogosphere out there. Uh, and
5:13so I had to figure out a lot of it
5:14myself, which was of course fine at the
5:17time, and I think a lot of people have
5:18learned that way. But, you know,
5:20thinking back to a lot of the things
5:22that I was taught as I got more serious
5:23about my career. So, when I first
5:25started off, I did paid search, but I
5:26also did every other channel. So, I had
5:28to learn how to code emails and learn
5:30how to do affiliate and things like
5:31that. And what I found there, what I
5:34found in frankly the own my own training
5:37that I put together with my team at
5:38Tenuity,
5:40a lot of our work with junior folks was
5:43teaching them how to do. Um, so we would
5:46teach them where buttons were. We would
5:48teach them what search query reports
5:49were and what we wanted them to look
5:51for. Like, oh, look for all the terms
5:53that spent $100 but didn't convert.
5:56Let's exclude them as negatives and all
5:57that jazz. And I think the challenge
6:01that I see in our industry and have seen
6:03for a long time, this is not a new
6:05challenge, but that very very binary
6:07sort of transactional view of what
6:09Google ads is like, oh, it's easy, it's
6:12just a math problem. Um, find out what
6:14converts, find out what words made it
6:15convert, let it go. But never thought
6:18us, never really taught us to look at
6:20outside inferences.
6:21Um, so I've made this joke quite a few
6:24times, but nobody wakes up and just
6:25decides to have a super googly day. Uh
6:28well some people do and they work at
6:30Google but something always happens
6:33before someone searches something always
6:35happens while they're searching.
6:36Something always happens after
6:39Google and many others of course like oh
6:41we can tell based on their browsing
6:43behavior or whatever. I think we all
6:45know that a lot of search engines can be
6:47a little obuse on that front, but really
6:49bringing it all together is teaching
6:50people how to be more
6:55the take the TV apart to see how it
6:56works type of people. Um, not just
6:58looking at, oh, we have to try new broad
7:02match or in current state of affairs, we
7:04have to try AIAX.
7:06Do you? What does it do? What are you
7:09hoping to accomplish? How is it
7:10different from what we're already doing?
7:12or you just trying a new toy because
7:13someone said you had to try it. And so I
7:15think we're a lot of the industry is
7:18falling into that ladder trap that you
7:20have to use AI, you have to try broad
7:22match, you have to try AI max, but
7:25without understanding exactly like how
7:27the tool works, what it's going to do to
7:28our accounts, and frankly what it's
7:30going to do to our brand or our client
7:32as marketers. Does it accomplish
7:34something that's relevant? Do we have a
7:36goal in mind that we hope to accomplish?
7:38Frankly, most people don't a lot of the
7:39times. and will this tool help us
7:42accomplish it? Being able to assess that
7:44sort of I'll call it a problem solving
7:45structure, but looking at the whole big
7:48picture instead of just how do I do my
7:50job as easily as possible so I can check
7:51it off a box,
7:54>> right? And you made the point that when
7:56you did this, everything was new and so
7:59you had to be behind the scenes. You had
8:00to experiment a lot. I would argue that
8:02we're at a stage where everything is
8:05kind of new again. there s sort of the
8:07old way of doing things or the the know
8:09the way that we all know it around
8:11search terms and bids. Uh but that AI
8:14shift is very real. People are searching
8:16in different ways. They are formulating
8:18what they look for in different ways.
8:20They have different touch points along
8:21the way to getting to that day of
8:23feeling googly and doing that final
8:26search. So I think there is a need for
8:29experimentation. But I like what you're
8:31saying, which is don't just use the tool
8:33because it's shiny and it's new, but try
8:35to understand what that tool can do. Um,
8:37and then I would argue, and I'm curious
8:39if you've had this, but oftentimes these
8:41tools are designed by Google under the
8:44idea of, okay, it's going to do this
8:45thing, but actually no code is perfect,
8:48and it always kind of like does maybe
8:50something else or unexpected things. And
8:52that could be an advantage, right? So,
8:53how do how do you strike that balance in
8:55an agency setting of maintaining your
8:58client's happiness with doing the things
9:00that we all know we should be doing, but
9:02balancing that with enough of the new
9:04stuff where we could figure out, okay,
9:06here's an opportunity that nobody else
9:08has figured out because we're just
9:09better than everyone else at knowing
9:11exactly how AI Max works and where to
9:14deploy it. Well, and there's a third, I
9:18guess, pillar in that little structure
9:20as well, which is your partnership with
9:21vendors, which put more simply is your
9:24relationship with Google. You don't want
9:26to take a product like AMAX and be like,
9:27"Ah, it's stupid. We're not going to
9:28test it for anyone." But trying to
9:31figure out how to prudently test it and
9:32test it where things are are more
9:34important is, I think, the tough part.
9:36So to me a big degree of it is
9:44gross oversimplification here but I
9:46think the way that most people use chat
9:47GBT for search is essentially delegating
9:50tasks to it and so you can look at it at
9:52the same the same way that you might
9:54look at an outsourcing exercise or
9:56offshoring exercise which I've been
9:57through a few attempts to do it marginal
10:00success sometimes not so good other
10:01times
10:03if you can't accurately describe what
10:05you're trying to accomplish how is a
10:06tool going to accomplish it for you? Um,
10:09you were kind enough to share a lot of
10:11the presentations you've done before and
10:13a lot of it is about really robust
10:14prompt engineering. How to write a good
10:16prompt, how to write an effective
10:17prompt. I think prompt engineering is
10:20super important, but again, if you don't
10:21really know what you want the tool to
10:23do, how are you going to engineer a
10:26prompt for it? So yes, teaching it the
10:28semantics like that it
10:32GPT is pretty bad at math sometimes and
10:34it forgets that CPA is a calculation and
10:36not a raw number. Okay, we can tell it
10:38that. That's fine. How do we get it to
10:40look for things that are actually
10:41effective? How do we get look for it get
10:44it to look for things that are actually
10:45going to help our accounts and as an
10:47agency? How are we going to get it to
10:49help power us strategically? You said
10:52being the best at AIM Max in the world
10:55isn't really a value proposition for a
10:57service vendor because everyone can kind
10:59of hit that button, but being able to
11:01explain what it will do to the broader
11:04ecosystem, explain what it'll do to the
11:05c the client and explain how it can help
11:07save the business money, I think is the
11:09most important part. So that goes in
11:10again to unpacking exactly how these
11:12tool tools work relative to a foundation
11:15that we'll have to establish in some
11:17way. And to your point, I think that's
11:19going to be the hardest part for
11:20agencies, and I'm frankly glad I don't
11:22have to do it, is what do we train
11:24people on now? Because we know consumer
11:28behavior is changing.
11:31We also know, and this is something that
11:32I realized during my little hiatus and
11:35coming back in is it's also really
11:37important to take a look at the internet
11:38from an outside view and remember that
11:40frankly most people aren't very good at
11:42it.
11:43>> So, a lot of people are going to get
11:45frustrated with AI overviews. A lot of
11:47people are going to keep on searching
11:48their same way. Others may evolve, but I
11:52took probably a little bit too much joy
11:55in watching my mom try to Google things.
11:57Um, and seeing how our parents use the
11:59internet and seeing how our older
12:00friends or less technologically
12:04uh whatever the opposite of inept is.
12:06Um, seeing folks who don't use the
12:09internet for a living, how do they use
12:11it? Because that's going to be most of
12:12the customers and how do they react to
12:14it? behavior is changing, but it's not
12:16all the way changed yet.
12:18>> So, what do we teach people on?
12:20>> Exactly. And that's maybe what makes
12:22this moment in time so difficult is that
12:24we have to be dealing with the past as
12:26well as get ready for the future. Um
12:30be because like like having been at
12:32Google,
12:34it's people didn't know how to search in
12:36the beginning and they would put in the
12:39in and like be very verbose about how
12:41they specify things and then they
12:42started to realize well these words
12:43don't actually help return different
12:45results. So let me use that limited
12:47space that I have for something that's
12:49maybe a bit more meaningful that helps
12:51filter down the results to what I need.
12:53But people have gotten there over time,
12:55right? And and so what what's kind of
12:56funny now is that we've been conditioned
12:58to speak Google ease in terms of how we
13:01search, but now we can actually start
13:03speaking human again and AI mode should
13:06presumably give us those answers. But
13:08but it is a reconditioning and like I
13:10said, there's um thinking about
13:12prompting. Now, now when I think about
13:14prompting thing, you can either be the
13:16expert and be very specific and explicit
13:18and basically the micromanager like
13:20here's my task, here's the five steps
13:22that I want you to follow and just go
13:23and do it exactly that way. Uh, but what
13:26I find fascinating and and and maybe
13:27like with more junior PPC people is tell
13:32them what the goal is and ask them how
13:34might you approach this and don't say
13:36like here's the goal, go and do it,
13:39right? But but say here's the goal, how
13:41might you do this? And then you see the
13:44thought process. And again, this is the
13:45the benefit of that junior person. They
13:46come up with novel ideas and maybe
13:48approaches that you had thought about.
13:50If they come with an approach that you
13:51know is flatout bad, well, they've told
13:53you, you can tell them that's not how
13:54we're going to do it and here's why. Um,
13:58but I I've certainly found myself being
13:59informed by these models and and finding
14:01novel ways to approach problems that
14:03otherwise I would have followed the the
14:05old methodology a bit more. Um, yeah,
14:08that's how I like using it. You know,
14:10it's interesting. You this you might
14:13have seen my eyes raised and a little
14:14light bulb go off over my head. So, I
14:16taught for a number of years. Um,
14:19one is an adjunct and one is just sort
14:21of like a survey level course. Uh, and
14:24and the survey level course at Drexel, I
14:26mean, we just had students come in one
14:27day a week. It was sort of an overview
14:28on all media types. I basically taught
14:32all media, which was tough to do in an
14:34hour. But the project that the students
14:36got
14:38is it was kind of a mad libs. They got a
14:41brand, a problem, and a delivery method.
14:43And so you would wind up with the
14:45stupidest stuff in the world. Like we
14:46had one group of students once that had
14:48to
14:50do a contest to announce a recall of
14:54milk and they had to focus on social
14:56media. But I mean that one was
14:59hilarious. But a lot of them, a lot of
15:00other ones were like, "All right, you
15:02had a PR snafu and you need to
15:04proactively get ahead of people, but
15:06also make sure that people are searching
15:07for it, that you're in front of them.
15:08How would you tackle this?"
15:10And that sort of thought process again,
15:12this little light bulb just clicked. I'm
15:13like, "Oh, there's something there for
15:15present times, too. There's something
15:17that we can put in there from from a
15:19problem solving perspective. Just give
15:20students a really weird problem
15:23that chat GBT certainly couldn't figure
15:25out how to do a contest for milk
15:27recalls." Um, but it makes it a little
15:30tougher. It's it's almost like in
15:31academia instead of trying to fight uh
15:35people using GPT for essay writing, make
15:38them think more. Make the essays have to
15:39be actually valuable instead of just
15:41giving a book report. So getting people
15:43more creative ways to think that the
15:44answer can't be accomplished just by
15:47putting a really simple idiot proof
15:49prompt in.
15:51>> Yeah. And and so that makes me think
15:53about education, right? And and one
15:55thing that I've really enjoyed that so
15:57in the GPT5 live stream uh the
15:59announcement of GPT5 coming out one of
16:02the things that was demoed was uh
16:05explained visually the concept of how a
16:07wing on an airplane produces lift
16:10and for for folks who haven't seen it
16:12amazing like it did take about five
16:14minutes give or take. GPT wrote a web
16:18page which had a visual diagram of a
16:20wing and you could change the angle and
16:22you could change the speed of the
16:23airflow and the little thing would show
16:26you is this airplane now taking off or
16:28is it going down? Um, and that's so
16:31cool, right? Because if you think about
16:33the Bernui principle and all of these
16:35other things that make planes fly, like
16:36it's very dry. It's like formulas and
16:39it's hard to visualize. You stick your
16:41arm out of the the car window like ah,
16:43now I sort of get it. if you can
16:45visualize this on a web page, like how
16:47amazing is that? And so I took that
16:49principle. I said, what if I could
16:51visualize how a Google ads auction
16:53works? Because I mean, for 20 years,
16:55you've been explaining this too, but
16:57like, okay, well, it's ad rank and it's
16:59basically CTR and CPC, these two
17:01factors, right? So, but then how do you
17:03calculate the actual CPC that you pay?
17:06And how does the organic section come
17:07into play? And and Google has these
17:09thresholds that they talk about for
17:11quality and minimum CPM. how does that
17:14fit in? And so I just made a tool. It's
17:17visual. So I'll put it in the show
17:18notes, but people can play with it. Um,
17:21and my idea is this a great way to
17:23explain it to your clients. They may not
17:25know the nuance of an ad auction, but if
17:26they were like, "Hey, Aaron, go and
17:29manage this campaign for me, but like
17:30here's my budget. It's $10,000. You
17:32cannot go above it." And you're like,
17:34"Well, okay, let me show you the bids in
17:36this industry, and with that $10,000, I
17:39can only get you this many clicks." and
17:40like look at this auction simulator like
17:43what do you expect? I can't get you
17:44beyond position six. Um and you
17:47visualize it and you let them play with
17:48a slider. It's like oh okay if I spend
17:50some more time improving my quality
17:52score ah my ad starts jumping and my
17:55cost doesn't actually go up. Um so that
17:58that's one really cool way that I've
18:00found that you can also use it for
18:02educational purposes. I really like
18:04that. And and going back to agency
18:07world, you just mentioned the thing um
18:10that's frankly the biggest time I'll
18:12call it a time suck, but that's a little
18:13disrespectful to clients. The biggest
18:16the biggest time suck at agencies. The
18:17reason why we can't just automate
18:19everything and go to AI is because of
18:22clients. the the biggest use of time
18:26that we had was in client services, uh,
18:28presentations, reporting,
18:32and those are areas where, believe me, I
18:34I tried when I was at TUNI to try and
18:36figure out a way that we could have our
18:38QBRS automatically made. And certainly
18:41GPT and Copilot can make some pretty
18:43presentations. And there's things like
18:45beautiful, but
18:48they're just fitting out book reports at
18:50this point. Um, and that's the real hard
18:53part. And again, hearkening back to
18:55academia where I think yes, GPT and and
19:00other sorts of generative AI can do a
19:01ton of work for us. It can also just as
19:04quickly call out the imposttors. It's
19:06pretty easy to tell someone who just
19:09copy and pasted a report or a bunch of
19:11insights from something they slapped
19:12into GPT because it will be
19:16it's almost as if you can watch it
19:18think. And so a lot of the reports, you
19:21know, give me give me 10 insights about
19:23my campaign like, "Oh, my CPA went up.
19:26Why did that happen?" Then a very
19:29rudimentary format would go like, "Oh,
19:32cuz CPCs went up and your CTR went down
19:36because you had less clicks and more
19:38impressions." I'm like, "Okay,
19:42cool. That's awesome. Great to know. Now
19:45what?" It's like, "Oh, you should get
19:47your CTR higher." like, "Thank you.
19:49Thank you very much." Um, it's not good
19:52at providing insights. It's not good at
19:53providing strategy. And that's going to
19:55that's what's going to be the biggest
19:57piece of work.
19:58>> Well,
19:59>> for real time savings,
20:01>> like I I totally hear you on that,
20:03right? But one thing that's um so I've
20:06been doing a little bit of vibe coding
20:08and
20:08>> waiting for this. I'm so excited for you
20:10to counterpoint this.
20:11>> Yeah. Well, so so in VIP coding, you
20:14basically just prompt the machine um
20:17with the app that you want to build and
20:18then the app gets built and then you can
20:20say, okay, I don't like this section or
20:21add more data like this here. But it's
20:24all prompt driven. Um now I've created
20:26something that's somewhat complex. It
20:28has a lot of data structures. And what I
20:31find really fascinating is that in the
20:33past when I was developing and I got
20:34stuck on a bug, I would have to open a
20:37database, look at a table, maybe figure
20:39out how that table was related to
20:40another table, go look at that table,
20:42and then I had to open the file that was
20:44supposed to render that data, but that
20:46file would have dependencies, and I'd
20:47have to open those files. And so just
20:50the literal process of looking at these
20:5210 different pieces of information that
20:54I needed to understand what might
20:56potentially be going wrong, you can give
20:59it that to an agent. a large language
21:01model and it very quickly opens these
21:03things in sequence and in a matter of
21:05seconds comes back and says I've read it
21:07all here's what's going on now it's not
21:10always right it doesn't always look
21:12deeply enough but just the fact that it
21:15uh you know say it takes five prompts so
21:17I spend three minutes doing this that is
21:20still faster than me processing each of
21:22these files in turn um and kind of
21:24losing my place and like did I really
21:26look at that did I not so that's where I
21:28think these LLMs
21:30maybe aren't at the stage yet like
21:32you're describing. So, they get stuck in
21:34a very like fundamental place, but if
21:37you feed them more information, that's
21:38where they can go deeper. And that's why
21:40I'm excited about Optimizer, too,
21:41because Optimizer has all of this deep
21:43insight and deep access into your
21:45reports. You can connect it with your
21:48first party data. And so, we can start
21:50pulling these things together. Uh we can
21:52even pull context across multiple
21:55accounts that you manage for the same
21:57portfolio, the same client. And so all
21:59of that research can now happen much
22:01more quickly. So, so I think what you're
22:02describing is is a fundamental problem
22:04right now, but I see it as an
22:06opportunity to build something that's
22:07better because the tech to do it is
22:09absolutely there.
22:11>> I would agree with that. I mean, I do
22:14love toying around with a lot of stuff
22:15that you built and a lot of stuff you
22:16described. Not a pitch, I promise. But
22:20um
22:21I mean I'm thinking even to
22:24some specific examples from from my past
22:27like I worked on a large pest control
22:28company for a long time and they were
22:30having a slow start to the year and we
22:32spent
22:34weeks well not literally weeks because
22:36we got the old phone call on a Friday I
22:38need you in our office on Monday to tell
22:40me what's going on thing. So hours and
22:43hours and hours 48 hour weekend to try
22:45and figure everything out.
22:4710 people, so probably
22:50300 person hours spent working to figure
22:53out that it was rainy in the spring, so
22:55there weren't as many bugs. And so it's
22:59literally that's literally all it was
23:01was that the, you know, swarms were
23:03slower or we had won uh this has
23:06probably happened to a lot of people, a
23:08fashion brand of some kind. We're like,
23:11wait a minute, why did we just have our
23:13best day ever? Of course, we figured
23:15this out with Google alerts or whatever.
23:17Oh, Ma Kunis was wearing our bag. That's
23:20why. Oh, because everyone wanted what
23:24she was she was wearing. And those are
23:26the sorts of inferences that I know that
23:29generative AI can solve with a little
23:31bit of coaching. I don't know that the
23:34everyday paid search person is going to
23:36get it there. So, until the product is
23:37there, I don't know that it's going to
23:39be the instant time savings that folks
23:41are hoping for in agencies,
23:44>> right? because it sounds like you still
23:45have to come in and ask the question and
23:48um so in the case of Makunis wearing
23:51your bag it's like that is one image on
23:53social thank god for Google alerts where
23:56they identified the bag probably but if
23:58they didn't how would you know right
24:00until somebody told you
24:03>> the other example about the weather like
24:04that that's really interesting because
24:05that seems like could apply to a lot
24:07more scenarios and like listen ever
24:09since the first day that Google ad
24:11scripts were announced it was all like
24:12hey bidding by
24:14like okay there is a perfect example why
24:17that might actually matter and that's
24:19why we support those use cases in
24:21optimizer but that was interesting too
24:22is like we support you setting up
24:24weather-based conditions but we don't
24:26necessarily tell you like what is the
24:28weather you should be looking for what
24:30are the patterns that we're seeing and
24:31that's where a machine learning model it
24:33doesn't even have to be geni but machine
24:35learning can sort of point you in the
24:36right direction
24:38>> it certainly could but then of course
24:40what data sources do you pull from and
24:42and how what level of do you have to
24:46give it? So in that particular example,
24:48like weather had to be unseasonably cold
24:51for an unseasonably long period of time,
24:53which would then delay the breeding
24:55period of termites and mosquitoes. So
24:57then our budget would spike back up in a
25:00month.
25:03We didn't figure this out.
25:05could machine learning, could geni
25:07probably, but who's going to coach it
25:09and who's going to build it,
25:11>> right? And so, um, one of the ideas
25:14behind large language models is that you
25:18make them as smart as they are by
25:19basically feeding it all the data
25:22>> and not getting in the way as the human
25:24and and this a little bit in Google ads
25:26too, right? Like the more that we look
25:27at Pmax campaigns, if you get in the way
25:30of it and say, "Hey, here's keyword
25:32themes. Here's negative keywords. Here
25:34here's audiences, we don't actually
25:36always see that correlate to better
25:38performance because you've made some
25:40human assumptions about what will and
25:42will not work. And you're getting in the
25:44way of the machine doing what it does
25:45best, which is find patterns and set the
25:48correct bid for something." Um, and like
25:51I've always said, there's no such thing
25:53as a bad keyword. There's only such a
25:54thing as a bad bid. Uh, right.
25:56Everything can be relevant at some
25:58level. Maybe you can't bid below a
26:00penny. So, in that case, maybe it's a
26:01bad keyword, but it's still a value
26:04question. It's not necessarily a
26:05targeting question. Um, so, so, but but
26:08you having been at so many agencies,
26:10like what if you were to start an agency
26:12today, thinking about this first party
26:14data, you have sort of the world's more
26:17broad data set like Reddit, economic
26:19indicators, is that something you would
26:21build into an agency to be successful?
26:23Like, how would you think about that?
26:29I don't think that I would build it as a
26:32proprietary agency tool. And the reason
26:34being is that once it's built out, once
26:36everyone will then have the idea. So
26:39they will see your pitch. They will see
26:40your web collateral. Yes. Could
26:44are they as smart as I am? Maybe, maybe
26:46not. Could they build it better than I
26:47could? Maybe, maybe not. But it feels
26:49repeatable to me. It feels much more
26:50like a a software player or a vendor
26:52play or something that you would build
26:54into a tool. Um, you make an interesting
26:58point about first-party data as well. It
26:59would be amazing if every client had
27:01their first party data in order. They do
27:03not. Um, the number of people that are
27:06still operating off of post-it notes and
27:09spreadsheets and things like that, or
27:11their first party data hasn't been
27:13cleaned up in 20 years, so nothing
27:15matches and it's all backwards. and then
27:17some CMO tried to migrate to a different
27:18place and it got broken.
27:21I think if there was a place that I
27:23would invest as an agency owner, it
27:25would be in data infrastructure
27:28because building building the tool,
27:29building the software, building the LLM,
27:31I think if you're building it for
27:32yourself, things will catch up in a
27:34couple years. like we like we were just
27:37talking about at the beginning. We're
27:38kind of at a a crux or a nexus in in
27:41terms of
27:43I don't know the world if you will that
27:45these things are just getting started
27:47and they're obviously moving very
27:48quickly. So investing all of that time
27:50and money into something that would be
27:52proprietary for me for a little while
27:54doesn't seem like a good investment.
27:56spending a ton of money in making sure
27:58that we can ingest and understand and
28:00clean up data. So then that could be fed
28:03in into something else. That to me
28:06sounds like an amazing investment. And
28:08funny enough, it's something a where
28:09agencies underinvest and b where clients
28:11don't want to pay because nobody has a
28:13data budget. And that to me seems like
28:16the most worthwhile investment for
28:17people.
28:18>> Yeah.
28:20Well, that's back to the fundamental
28:21point of uh you know automation can only
28:24do as good of a job as the data that you
28:26put in.
28:27So fully agree you should get your data
28:30structure in place and uh and make an
28:32investment in that. Um I mean h how did
28:35you see at the agency like people
28:36handling conversion data and offline
28:38conversion imports? Were they at least
28:41decent at that or is that a struggle and
28:42still an opportunity for agencies?
28:45Um,
28:48most clients were competent at getting
28:51it done. They could handle it. Where we
28:53would run into brick walls is making
28:55sure that the sales people were doing
28:57it. We could connect the pipes, but we
28:59couldn't make make people put them in
29:01the right place.
29:02>> And so solving for the human problem was
29:04the biggest the biggest challenge. I
29:06mean, I went through this experience
29:07with a call center once where,
29:10oh, we're doing amazing. Like, we have
29:12more MQLs than we've ever had this week.
29:14Our conversion rate is really high. What
29:16is happening? This is amazing, but is
29:18it?
29:20Call center was doing bonuses for the
29:22most NQLs generated. So every SQL was an
29:24MQL that month because people wanted to
29:26get their bonus. And so it's just like,
29:29okay, okay. That's that's that's the
29:32biggest challenge that I think we faced
29:34as agencies is number one, uh, people
29:36would only tell us marketing stuff and
29:38like, no, no, no, we need to know
29:39everything. We need to know everything
29:40that's going on in the business. We need
29:41to know where all your data is. We need
29:43to know your PR schedule. We need to
29:44know your TV. Um, so then we can use it.
29:47Then we can feed it into a proprietary
29:49tool or frankly anywhere and make better
29:52decisions as opposed to just here's
29:55here's your marketing budget, go spend
29:56it.
29:58>> Yeah.
29:59All this stuff. Um, and uh, you know,
30:02you mentioned too that working with
30:04clients is one of the most timeconuming
30:07things. Doing the reporting for them.
30:08Sometimes the client gets in their own
30:10way of success.
30:12Um, here's something I want to spitball.
30:14I've been doing this in other PPC town
30:15hall episodes, but one idea is why not
30:18see the client meeting as the master
30:21prompt for the work that needs to be
30:23done? Because most of the time, like
30:25agencies don't like client meetings
30:26because it's just time that you sit
30:28there and what happens at the end of
30:29that call? Well, you've got action
30:31items, which means more time. What if
30:34you could think about that 30 minute
30:35meeting as hey let's really figure out
30:37what it is you want to do how we're
30:39going to go about it and at the end
30:40we're going to take that transcript give
30:42it to the AI and the agent and at least
30:45the simple stuff like adjust budgets for
30:48this thing that we just told you is
30:49going to be a promotional period like
30:51that should be doable by the AI um
30:55what do you think about that is that
30:59>> so first of all I don't want it to come
31:00off like I hated client meetings because
31:02I love them Um, I frankly thought they
31:04were the best time cuz that was when you
31:05would actually get the business
31:06information. It was the preparation that
31:08was awful and very often the data format
31:11that people needed because they would
31:12need an analysis for their own budget.
31:14They would put their own budget tool,
31:15whatever. Um,
31:19it's funny. I think you and I are
31:20speaking the same language but in a
31:21slightly different view. You most people
31:24probably don't know this. I'm a little
31:25bit of a lite personally. Like I tend to
31:27not be lean into crazy new technology
31:32that often in my personal life,
31:34obviously in my work life, but so we're
31:36saying the same thing in exactly
31:37different ways. What you just described
31:40is a good structure for a presentation.
31:43You just described having a presentation
31:45that's purposeful, that you have things
31:46that you want to show them, that you
31:49have a goal, that you have something
31:50that you want a client to say yes to at
31:52the end.
31:54So fundamentally in in both cases it's
31:57again it's good prompt engineering or
31:59good goal setting and so I think as long
32:01as we are training people on junior
32:04folks and senior folks alike training
32:06people on how to go into these
32:08conversations with an extremely
32:09purposeful route certainly in its
32:11current state then we can have
32:13something automate most of the
32:15presentation development certainly all
32:16the charts and numbers and then we put
32:18the words on it but as long as you go in
32:20there with something that you really
32:21want out of it then you'll be then
32:23you'll be very successful and then
32:25you'll have your go forward instructions
32:26because you
32:28>> had your engineered a prompt or
32:29structured a meeting really well
32:31>> well now that you put it that way it's
32:33making complete sense to me it's like
32:35your perfect slide deck has an
32:36introduction about we are your expert
32:39PPC management team so you're going to
32:42act in that role here's what we've tried
32:44in the past and here's what we've seen
32:46call text here's what we're trying to
32:48achieve goal here's now let's talk about
32:50that goal So we put in place some limits
32:53and parameters. Um and then like I said
32:55at the end of the thing it's like is
32:57that what you want us to work on? Yes.
32:59Okay. Go and do it. That is exactly how
33:01a prompt would be. So uh yeah I love
33:04that. I think it's a great takeaway for
33:05anyone listening who maybe has more
33:07loosely structured meetings. Like think
33:09about it in terms of how you would
33:10prompt an AI. And um if you get it to
33:14that point then eventually I think
33:15you'll be at a level where you give it
33:18to the AI and it'll do most of the work
33:20for you.
33:21Well, well, and so many of these
33:23questions are I mean, speaking to client
33:25meetings specifically, very often they
33:27would be like, "Oh, what do you want to
33:28see in this quarter's presentation?"
33:30Just do the same stuff we did last time.
33:32Don't let them do that. In the same way
33:35that you wouldn't go to generative AI
33:37and like, "Ah, just give me some ads
33:38that they look fine. Here's our website.
33:41Just go for it." You just take them. Um,
33:43you kind of have to press people. You
33:45have to press AI to get creative. I was
33:46speaking with a a
33:48former student. and he graduated now who
33:50I mentor who's uh from from Villanova
33:53where I went to school and
33:56I unfortunately funny enough I do have
33:58post-it notes in front of me and I did
34:00throw it away which I should have had AI
34:02remind me of what it was but uh I
34:04believe he called it the death of
34:05originality uh as the thing that we're
34:07the most worried about which is very
34:09easy to fall into a trap with if you
34:11have someone who's uh undereducated or
34:13doesn't want to think very much and
34:15they're working with a powerful tool
34:17like GPT or
34:20to a lesser extent, uh, even Pmax. Um,
34:24every PMAX ad looks the same. Uh, I know
34:26when I get targeted by a retargeting ad
34:28cuz it's the same stupid thing with the
34:30same headlines that we all thought were
34:32best practices. And so I think if
34:34anything the most important thing that
34:36we can teach in current state of affairs
34:38for presentations for prompting for
34:40marketing in general is
34:42keep your creativity and remember it and
34:44don't expect whatever tool to come up
34:47with that for you.
34:50>> Yeah. And and remember too how the tool
34:52got to suggest those things. It's
34:55because historically there's been a lot
34:57of use of that and that may have
34:59correlated to good success.
35:01But for those things to become
35:03successful in the first place, at some
35:04point they had to be innovated. They had
35:06to be someone tried it and they saw more
35:08success than all of the other stuff and
35:10then everyone else followed suit. So if
35:13you want to think about unleveling the
35:14playing field and having an edge, uh I
35:17completely agree. You need to be
35:18creative and be willing to experiment
35:20with something new. But but I think
35:21that's also a client conversation,
35:23right? Like listen, we can do the easy
35:25thing for you, not that hard. we're
35:27going to get decent enough results, but
35:29if you really want to be best in class,
35:30here's what we need to spend for a
35:32little bit of testing budget and here's
35:34some of the things we need to try,
35:35>> right? And it might not work every
35:37single time, but we're going to learn
35:38from it. Then we can learn, let that
35:39learn somewhere else, and we can let all
35:41of our idiot competition
35:43do the quote unquote best practices that
35:45they read on a blog somewhere, which by
35:47the way was a phrase that I banned from
35:49my team at Tenuity. They were not
35:50allowed to say best practices. Best
35:52practices are lazy. They're average.
35:54They're the starting point. They're what
35:55everyone else suggests. like don't don't
35:57say like it's a best practice like no no
35:59no no let's say what we're trying to
36:00accomplish and why.
36:03>> Very interesting
36:04>> hard to do.
36:05>> Um so Erin uh any AI tools that have
36:09some connection to marketing that you've
36:11been like really engaged with that you
36:13found fun some some interesting stuff
36:15that you've seen?
36:17>> Frankly I'm still relatively um junior
36:20in my AI journey. Like I say I had a
36:23year plus of lititness which has been
36:25kind of fun. It's been kind of fun to
36:26come back in as an infant. U first of
36:29all, an obvious plug for Optimizer
36:31Sidekick because love that little guy.
36:33Um,
36:35for those who don't know, it's a natural
36:37language uh what you call it a plug-in
36:39tool toy that sits
36:41>> a chat feature inside your optimizer
36:43account that basically uses LLMs and a
36:45deep integration into your ads data to
36:48uh help you with whatever you need.
36:50>> Clippy for the 21st century.
36:53You know, I saw clicky. I don't know if
36:55people have seen the new Naked Gun
36:57movie.
37:00>> First of all, what? There's a new Naked
37:01Gun movie.
37:02>> Yeah,
37:05>> check it out. Good movie. And Clippy
37:08makes his return. And uh I'll let you
37:11all find out if he's super annoying or
37:14super helpful.
37:15>> Well, there's my evening plans. Um but
37:18no, beyond that, I've just been toying
37:19around with the kind of standard ones.
37:22Um,
37:24why did this just leave my brain? Uh,
37:26GPT Claude, uh, toying around with
37:29Gemini, toyed around with Copilot. Um,
37:33we were toying around with some, we were
37:37toying around with some early
37:38integrations of
37:40whatever Gemini was called before it was
37:42Gemini. Um, and integrating that into
37:44some Google Sheets plugins. Um, doing a
37:47bunch of cool stuff that way.
37:49Um, I did mess around with beautiful a
37:51little bit, which is more or less
37:53presentation development, but I haven't
37:55gotten as sophisticated as others, which
37:57I think if anything makes me a good
37:58candidate for a conversation like this
38:01because I think I represent most kind of
38:04mid-level people at agencies at this
38:05point that they're going to learn enough
38:06to do their job. And so, how do we make
38:09them use it effectively?
38:12>> Right? Right. And it's sort of you have
38:14these amazing tools, but not everyone's
38:16gotten good at prompting or figuring out
38:18how to connect the pieces together. And
38:20and that's where you then see specialty
38:21tools coming out which very simply are
38:24built on top of GPT5, but they figured
38:26out how to give better context, how to
38:28do better fine-tuning and really get it
38:30to achieve a certain task for you. Um,
38:34and uh, yeah, I I'm playing with a lot
38:37of it right now. Um, vibe coding these
38:40auction simulators. I'm uh coding a
38:44ghost blogger functionality so that I
38:47can write blog posts as if I'm talking
38:49to a reporter and it asks me about my
38:51thoughts on certain topics and then it
38:53constructs something that is much more
38:55relevant to who I am. Um and then of
38:59course all the things we're doing in
39:00optimizer. So again not too much of a
39:02plug for optimizer but I'm super excited
39:04that sidekick is actually starting to
39:06work across the whole account. So, if
39:09you as an account manager need to go in
39:10and say, "Listen, I just need um to make
39:14a report for my client, and I don't want
39:16to use the custom reports that are in
39:18optimizer, but I'm just going to click
39:19through the interface, and as I look at
39:21stuff, I'm just going to talk out loud
39:23about what I see and what I think that
39:25means." And so, at the end of the
39:27session, the optimizer sidekick has
39:30heard you talk about like a variety of
39:31tools, and then it compiles that and
39:32says, "Okay, well, let me group it.
39:34Here's some keyword stuff. Here's some
39:35budget stuff. here's some like strategic
39:37long-term vision and it actually spits
39:39out a narrative that's not just
39:42generically generated by the LLM, but it
39:44is actually you as the account manager
39:46and your expert opinion combined with
39:48what the data is showing. Um, so this is
39:51sort of that future vision that I'm very
39:53excited about and I think is going to be
39:55hopefully very helpful to people.
39:57>> Well, and it's tangible. It's it's it's
39:59easy to work with and frankly that's a
40:00lot of the way that I've been my own
40:03again I'll call it um AI journey has
40:06been all right I'm going to try
40:08something a little bit more complicated
40:09this time I'm going to I have had an
40:12idea in the back of my head for an app
40:14forever that I'm going to vibe code now
40:15because now I'm getting more comfortable
40:17with it. Um but just toying around with
40:19simple things and figuring out what it
40:20gets wrong. In my keynote for uh SMX
40:23which was gosh probably over a year ago
40:26at this point. It was around this time
40:27last year.
40:29I sort of peppered in bad AI throughout
40:32which was just single prompted which was
40:34at the time my perception of AI and I
40:36think a lot of people's. But now as
40:37we're getting much better and uh much
40:40stronger at it and the models are
40:41getting better and learning faster.
40:44All right. Now that now this is a toy I
40:46can play with. Now I can build an app
40:47for people to find other people to go
40:49play golf with. Don't steal it. It's
40:50mine. But
40:53>> it's funny too because you did that
40:54keynote uh dissing AI and I probably did
40:57a keynote the one before at the previous
41:00SMX like all hyping AI. So we're the yin
41:03and the yang here. Uh but I think that's
41:05good, right? And that's why I'm happy
41:06having you in the company to sort of
41:07like bring in that agency practitioner
41:10perspective and maybe uh boots on the
41:13ground and like how does all of this
41:14actually play out because because I
41:16living in Silicon Valley I mean I'm so
41:18surrounded by it and that vision where
41:20we think we could go but like what's
41:22happening in most of the world um let's
41:26let's bring it back to reality at some
41:27level. Yeah, I'm in South Philadelphia
41:29where nothing has changed for 150 years
41:32except my corner store is a house now,
41:34but besides that, um, no, it is I mean,
41:37I've always tried to take that approach.
41:39Uh, especially when I was managing a
41:41really large scale team. Yes, I had my
41:44super high performers, but I didn't have
41:45to build stuff for them. They would they
41:46would go build their own stuff. I had to
41:48build stuff for everybody, and I had to
41:51build things with guard rails for
41:55I had only A players, but some were A
41:57minus players. Come on, everyone.
41:58Everyone has a bell curve. But so you
42:00have to make these things tangible for
42:02everybody and make them usable for
42:03everybody, including people that are not
42:05comfortable with technology, which I
42:07think makes us kind of good foils here.
42:10>> So, and that's kind of an enviable
42:11position, right? Like having a players
42:13and maybe a few A minus players. How do
42:16you do that as an agency? How do you get
42:18those players?
42:20>> Well, number one, we gave our teams a
42:25ton of freedom. And so in turn, if there
42:28were people who required straight up
42:30direction, direction, direction,
42:32direction, they would get found out
42:34quick. Um, it would be people asking for
42:37step-by-step directions for doing
42:38everything. Instead, we would sort of,
42:41you know, we would have a framework and
42:43here's a rough thought process, but you
42:45take it, you build it in your style, you
42:47build it in your own way. So, I always
42:48like to say that at Tenuity, you could
42:50look at an account and tell it was a
42:52tinuity account, but you could also see
42:54that it was like Alana or Jeff's account
42:57um because they would have their own
42:58style that would take off our own start
43:00point. And again, speaking to large
43:03scale agencies, almost always the person
43:06clients almost always want the person
43:08who's going to be working on their
43:10business to pitch the business. So in
43:12turn, the person would get to pitch
43:14themselves just as much as we would get
43:15to pitch the company. And so that would
43:18make clients feel a lot more comfortable
43:19because then a person wasn't kind of
43:21forcing
43:23or or maybe neutering their own style um
43:26because it had to fit into a company's
43:28best practices,
43:30>> right? So give them autonomy.
43:33Um tell them what you're trying to
43:36achieve and not how to go about it.
43:38>> Autonomy, a lot of trust. Um,
43:42and then you'll find that the the lower
43:45level players will shake out quickly.
43:47Um, you know, we had atuity, we had
43:50unlimited vacation. I don't think I've
43:51ever had a company where I had to count
43:54my hours. I mean, even you and I working
43:57together when we started, I was like,
43:58"So, what hours should I work?" And you
43:59said, "I don't know. Whenever you need
44:01to do your work."
44:03That sort of an attitude has always been
44:04viewed as sort of less afair, like, "Oh,
44:06how do you make sure that they're not uh
44:08slacking off?
44:10If anything, I think you find that out
44:12faster that way with with freedom and
44:14trust.
44:15>> Yeah,
44:17good advice. Okay. Well, um Erin, you
44:19did a keynote at SMX uh like a year ago.
44:22People can go and find that, but you're
44:24also going to be speaking at the next
44:26SMX next, I believe. Uh what can people
44:29look forward to?
44:31>> Uh more me um more more contrarian
44:34stuff. I mean, a lot of what I'm excited
44:36about in working with optimizers is use
44:38a super strong team of analysts and
44:39obviously a ton of data to play with.
44:42And you know, in the past, I certainly
44:44had that available to me, but not to
44:46this level and not to this volume. So,
44:48what I'm most excited about is to start
44:49dissecting some of these things that are
44:51coming up. I'm really excited to poke
44:53into AIAX and find out if it really is
44:56something special or if it's just DSA
44:58with different buttons. Um, and trying
45:01to figure out how many of these things
45:02are incremental, how many of these
45:03things are driving value. That's really
45:04what I'm looking for is not
45:07new tool. People tried tool good, but
45:10try to figure out what what did the tool
45:12do? Is it doing something new or is it
45:13just shoving food around your plate?
45:15That's the sort of stuff that I'm going
45:16to be diving into the most because those
45:17are the questions that we get the most.
45:19>> Great. Okay. Well, everyone sign up for
45:22that. And then uh if you've enjoyed PPC
45:24Town Hall, do subscribe, give it a like
45:26so that other people find it. and Aaron.
45:29With that, uh, people can find you on
45:31LinkedIn. Uh, you're part of Optimizer,
45:32so they can reach you there as well. Go
45:34and see Aaron and myself at the upcoming
45:36conferences. But with that, we'll wrap
45:38it up here. Thanks for watching, and
45:40we'll see you for the next one.
45:42>> Thanks, boss.

About this transcript

This page contains the full transcript of Lessons From 20 Years in PPC: How Agencies Should Prepare For The Future | PPC TH 117 by PPC Town Hall, generated from the public captions YouTube serves with the video. The transcript has 9,178 words across 1,298 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.