April 12, 2025 • 40 mins
I am always looking for ways to improve. Send me a text and let me know your thoughts! - Kevin
Jacob Bank is transforming how marketers work with AI agents – and the results are extraordinary. As founder and CEO of Relay, Jacob brings a unique perspective shaped by his background as an AI researcher, founder of Timeful (acquired by Google), and former product lead for Gmail and Google Calendar.
In our conversation, Jacob cuts through the confusion surrounding AI terminology to explain what truly matters: AI agents are tools that work behind the scenes with human-level intelligence on tasks that previously required people. The technological leap enabling this revolution isn't just incremental improvement but a fundamental shift in how AI reasons and plans multi-step processes.
What makes Jacob's insights particularly valuable is that he's applying these technologies directly in his own marketing. At Relay, he operates as essentially a one-person marketing team, leveraging AI agents for competitor research, content creation, customer insight synthesis, community management, and lifecycle communications. From automatically analyzing customer calls to generating personalized outreach and tracking performance metrics, these agents handle the "busy work" while Jacob focuses on strategy and relationships.
Perhaps most compelling is Jacob's candid story about his content marketing journey. After nine months of consistent posting on LinkedIn and YouTube with minimal traction, his persistence finally paid off – exploding from 1,000 to 5,000 YouTube subscribers in a month and generating 4.7 million LinkedIn impressions in 90 days. This transformation didn't come from changing posting frequency but from refining his approach and persevering until breakthrough.
Looking ahead, Jacob believes marketers who build authentic connections and communities will thrive even as AI-generated content proliferates. The successful marketer of tomorrow will be a curious, adaptable generalist who embraces new tools while focusing on the human elements that technology can't replicate.
Ready to transform your marketing with AI agents?
Follow Jacob on LinkedIn or YouTube for detailed tutorials
Visit RelayApp.com to start building your own AI-powered workflows.
🎧 Tech Marketing Rewired is hosted by Kevin Kerner, founder of Mighty & True.
New episodes drop regularly with unfiltered conversations from the frontlines of B2B and tech marketing.
Catch the video version of the podcast on our YouTube channel.
👉 Subscribe, share, and connect with us at www.mightyandtrue.com
📩 Got a guest idea or question? Email us at kevin@mightyandtrue.com
🔗 Follow Kevin on LinkedIn for more insights and behind-the-scenes takes.
📰 Subscribe to Kevin's Substack at: Subscribe to Kevin's Substack at: https://kevinkerner.substack.com/
📰 Want more tech marketing insights? Subscribe to the Drag & Drop newsletter for fresh thinking, real talk, and tools that make your job easier.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Kevin Kerner (00:00):
Hey everyone, I'm Kevin Kerner and this is Tech
Marketing Rewired.
I just had the chance to sitdown with Jacob Bank and I've
got to say this one was a realpleasure.
I've been watching Jacob'sYouTube videos for a while now
and I've always appreciated hisenergy, his curiosity and his
willingness to jump into the AIchaos with the rest of us.
In this episode, he breaks downthe real difference between AI
agents and traditionalautomations.
(00:21):
He also gives us abehind-the-scenes look at how
his team at Relay is usingagents in their own marketing,
and let me tell you, this isreally smart stuff.
Plus, jacob shares a story ofhow he finally cracked YouTube
and LinkedIn his own personalmarketing after a year of
testing, tweaking and totalpersistence.
And don't miss the AI rouletteat the end of this one.
It's a good one.
Let's dive in.
(00:41):
This is Tech Marketing Rewired.
Jacob Bank (00:50):
Well welcome, jacob.
I'm glad to have you on theshow.
I'm super excited it's going tobe a fun conversation.
Kevin Kerner (00:54):
So the first thing I wanted to do is just have you
introduce yourself, talk alittle bit about your journey
and then I definitely want youto say what you're doing now at
Relay.
Jacob Bank (01:04):
Yeah, I'll start with what I'm doing now and then
I'll go back to quickly how Igot here.
I'm currently the founder andCEO of Relayapp, which is a
platform that enables everyone,including non-technical users,
to build AI agents, automationsand workflows.
My road here was kind ofwinding.
Once upon a time, I was anartificial intelligence
researcher.
I thought I was going to pursuea PhD and become a professor.
(01:24):
One day, I ended up droppingout of my PhD to start my first
company, which was an AI-powereddigital calendar called Timeful
.
We got acquired by Google and Ispent six years at Google where
I led product for Gmail, googleCalendar and several other
collaboration and productivitytools, and so I basically spent
my whole career at thisintersection of AI and
collaboration and productivitytools, and that's what's taken
(01:45):
me to where we are now, wherethe stuff actually works.
Kevin Kerner (01:47):
finally, Well, I think we've now discerned that
you're a lot smarter than I am.
Far far from it, but I pretendto be on YouTube.
Yeah, that's really awesome.
Okay, I want to talk firstabout.
I want to.
You know, we have a lot ofpeople that are hearing all
about agents and agentic AI andall these things.
How do you describe or maybedefine what you see as AI agents
(02:09):
and maybe even the differencebetween that and agentic AI?
Jacob Bank (02:13):
It's easy to get lost in the terminology and I
don't think it matters that much.
So let me explain it how Ithink about it as a user of
these tools.
There's three basic ways that Iuse AI every day.
Number one is I go to a chatbot.
That's like ChatGPT or ClaudeProbably most people have tried
that by now.
Grok, pick your favorite.
You go talk to it and say likehey, I'm about to talk to Kevin
(02:36):
on Mighty and True, do someresearch about him for me.
So I show up prepared.
That's a chatbot.
You're in a window writing textmessages to an AI and it's
responding.
The second model is a co-pilot.
A co-pilot is you're in thetool that you normally use every
day and the AI is somehowhelping you do things.
So, for example, one of thefeatures I worked on on Gmail
was that as you're typing outyour email, the AI will suggest
(02:58):
what you might want to say.
Next, if you write code, youmay have used a tool like Cursor
that suggests what code youmight want to write.
Or if you've used Google Docsor Microsoft Word, they now have
co-pilots where they can helpedit your writing or suggest new
ideas.
So that's a co-pilot.
A co-pilot is you're in a tool.
You're not in a chatbot.
You're in a dedicated tool toproduce something and there's an
(03:19):
AI experience within that toolthat's helping you do it.
And then the third class of AItools I use every day are those
that are happening behind thescenes.
I'm not actively chatting withsomething in a chatbot.
I'm not actively using it in aco-pilot.
The AI is just doing work on mybehalf.
And that's where people havethis kind of technical debate
(03:40):
about what's an AI automationversus an AI workflow or what's
an AI agent.
But I think the fundamentalthing to keep in mind is that
there's this class of AI thatcan work on your behalf behind
the scenes on things thatpreviously would have required
human level intelligence.
For people who are reallyinterested in the technical
distinction, an AI automationwould be like a single step,
like generate a summary of thispodcast.
(04:02):
An AI workflow would bechaining multiple of those steps
together, like generate asummary of this podcast.
An AI workflow would bechaining multiple of those steps
together, like generate asummary of this podcast and then
create a post on LinkedInpromoting it, where we have
multiple different AI actionsthat are chained together into a
workflow.
And an AI agent is technicallya little different.
With an AI agent, you just giveit a goal and a set of tools
and say agent, go do work for me.
(04:24):
And so I would give the agent agoal.
That's like hey, I want topromote this podcast episode.
The tools you have access toare you can read transcripts,
you can write LinkedIn posts andyou can write Twitter posts.
Go, promote it as you see fit.
And so automation, workflow,agent.
The line between those threecan be very blurry and I find
the you know the debates aboutthe distinction between them on
(04:45):
Twitter and on LinkedIn kind ofbeside the point and overly
pedantic.
So the way I would think aboutan AI agent I wouldn't worry
about the difference between anagent and a workflow and an
automation that's just notpractically relevant for any of
us.
The thing I would think aboutis an AI agent is something that
can work on my behalf behindthe scenes, without me directly
chatting with it or using it asa co-pilot, and it can do things
(05:06):
with human level ofintelligence that I previously
would have needed a person aperson to do.
And that's a perfect examplelike promoting writing a really
good promotional post onLinkedIn about a podcast episode
.
Kevin Kerner (05:17):
Is there a reason why now it's getting so much?
Because these workflows havebeen around for a while and I
guess there's been ways to diginto AI in these workflows.
Is there a reason why now, ifwe're just in a hype cycle, or
is there is there a technicalreason why now they're becoming
more accessible or more peopleare just more interested in them
?
Jacob Bank (05:33):
Models are way, way better.
The models are way, way better,like, even basic things like
extract invoice information froma PDF barely worked two years
ago and now works flawlessly.
Things like write a LinkedInpost in my voice two years ago
didn't really work.
You got this like AI generatedslop with all the em dashes and
(05:54):
all the you know words thatpeople poke fun at that chat
should be to use, whereas nownow it's, the models can do it
very well.
So the models have improved, Ithink, along two dimensions that
I think are important foreveryone to know about.
One is their core quality at asingle task has improved.
They are way better atextracting information from PDFs
.
They're way better atsummarizing long form
(06:15):
transcripts.
They're way better atgenerating prose.
But you've probably also heardof this other kind of models
called reasoning models, andthose reasoning models are able
to do sort of multi-step chainof thought actions where they
have to make a plan about how toachieve something and then
deliver on it, and they've alsogotten way better at that.
And it's that second, that leapinto reasoning, that has
(06:36):
enabled people to shift fromvery well-defined AI,
automations and workflows intothings that feel a bit more
agentic.
And when I say more agentic, Ijust mean you're giving the AI
more latitude to figure out theplan of how it wants to
accomplish its goal.
Kevin Kerner (06:50):
I watch a ton of your videos and I always find it
funny when you're building someof the build with me and you
put something in the AI and itbrings back a result and you're
surprised how good it is.
You're like, whoa, wow, thatgot that thing.
Jacob Bank (07:03):
Honestly, this happened to me literally today,
where I was like I was doing asession today where we were
going to do some promptengineering and the idea was
we'd go into the session andthen it would get it wrong the
first time, we'd try again.
It would get it wrong.
The second time We'd try again,it would get it wrong the third
time We'd try again.
And I wanted to show peoplekind of here's my progression of
how I test and iterate to makesure I get a prompt.
(07:24):
That's really good and thesession kind of kind of blew up
because I just did it reallywell the first time for all my
test cases and you know I foundsome different.
You know I found some ways tokind of push it into areas that
it wasn't quite able to do.
I gave it harder and hardertasks.
But I'm like I'm in this everyday and every day I wake up
surprised by how good somethingis.
Kevin Kerner (07:45):
Wow, that's wild.
So you can see the improvement,maybe not day to day, but week
to week.
Jacob Bank (07:53):
Definitely week to week, and so I would encourage
everyone.
I'm not one of these people whothinks all human work is going
to be irrelevant by the end of2025.
And so I always try to presenta pragmatic, realistic view of
what AI is capable of today.
But I would really discount theopinions of people who say it's
all hype, it can't do anythingLike.
(08:13):
Those are people who are nottrying hard enough, because if
you find the right use cases andyou find the right tools and
you put in a bit of time togetting it good, it's really
amazing.
And one of the things that I'mtrying to push back on is a lot
of times people will say is likeI tried to get Chachi BD to
write me a blog post and itwasn't good, and I'm like, well,
what did you ask it to do?
And you say, like I gave it.
(08:34):
You know one sentence of theblog post I wanted to write.
Did you give it any context onwhat key points you wanted to
make?
Did you give it any of yourpast blog posts so it knows your
style?
Did you tell it anything aboutyour ICP and the kind of
language those people use?
Did you give it any contentfrom customer calls about issues
that might be top of mind.
Like imagine if you asked ajunior in college intern from a
one sentence prompt to justwrite a blog post.
(08:55):
They're not going to writesomething good based on that
information and so you reallyhave to think of like the mental
model right now for me that Ifind most helpful is imagine a
really, really bright sophomorein college intern.
That's like a good startingpoint and but that changes every
week.
If we talk again in threemonths, it might be.
Imagine a really good new gradand if we talk again in three
(09:18):
months, it might be.
Imagine a really good contentmarketing professional who's
been in the in the game for fiveyears.
Because it's the pace of the.
The evolution is incredible.
Kevin Kerner (09:25):
Yeah, the people that don't get really good at
that soon are going to be kindof left behind or they're going
to have to learn really fast,you know.
The other thing I was thinkingabout in terms of the maturity
of your tool and other tools isI don't know if you agree with
this is that it seems like theability to dig into other
applications.
Other applications are sort ofretrofitting to, and it used to
be API calls and you know itstill is to some degree, but man
(09:46):
, it's getting so much easier todig into other applications and
the more you build thatecosystem out of all these
different things that can beconnected with and have this
agent rules engine inside of it,it just it seems like it could
be really powerful.
I would put that thing ineverything you know as much as
you can get it into it's hardwork.
Jacob Bank (10:03):
But here's my current take on what it takes.
For an AI agent to be valuable,it needs three things.
Number one it needs to connectto your actual tools, because if
all it does is spit out sometext that you then have to
reformat and copy paste like,it's just not going to be able
to do work on your behalf if itcan't deeply connect to your
tools, both to read data andunderstand context and also to
take action.
(10:24):
Number two this is something Ifirmly believe the AI has to
show you a plan of what it'sgoing to do before it does it,
Especially in these early days,as we're all just building trust
and trying to figure out whatcan it do, what can it not do,
how it shows you the plan in thesame way that a great new
employee says hey, Kevin, thisis the task you gave me.
Here are the five things I'mgoing to do, and I'm going to
(10:44):
get this on your desk byThursday.
Does that look right?
And then you say, yes, no,we're like oh, also check this
source or add this step here.
And then number three is as theAI is executing, things will
come up and it'll need help it.
But, like, tell me if you wantsomething different.
And so if you can connect toall the tools, if you can be
(11:08):
very transparent and trustworthyabout what your plan is and if
you can loop in the human in theright way, as you're doing, I
think that's how you build auseful AI agent.
And on that first point that youwere making about tool
connections, there's two basicapproaches.
One is what's called an APIthat you mentioned, and this is
basically a special set ofcapabilities that software tools
offer for other software tointeract with them.
(11:29):
So you know, Salesforce willexpose an API that lets other
tools look up records and createrecords and search over records
.
But many, many tools don't havean API or don't have a complete
API, and so for those tools,you basically need to just go in
the browser and mimic a humanclicking around.
Now, in my experience right now,I have yet to succeed with a
(11:52):
browser clicky tool.
So there's Operator from OpenAI.
Manus got a lot of hyperecently.
There's Cloud Computer Use.
I would love to see use caseswhere you're using it for real,
but I have not successfully usedOpenAI Operator for anything
yet.
But that doesn't mean you canwrite it off forever.
I mean right now I'm buildingall of my workflows based on
(12:13):
APIs, with some light webscraping, but I totally can
believe that six months from now, that I would be able to fold
in Operator to some of theseagentic use cases that I use,
that's cool.
Kevin Kerner (12:23):
Operator to some of these, these agentic use
cases that I use, that's cool,but you and you haven't seen any
end applications cutting offaccess or limiting access, as
they want their own AI tools.
Jacob Bank (12:31):
The one case that I have seen is that LinkedIn has a
pretty locked down terms ofservice around what automated
actions you can take, and so Idon't know the details of it,
but they got into a little bitof a dust up with Apollo and
Seamless and a couple others acouple weeks ago, and so I think
, in general, most platformsbenefit from being open and
(12:54):
having good APIs, and so Iexpect that most companies will
invest more in their APIs andnot less, as AI agents become
maybe the primary thing.
That's interacting with thesetools like making an AI agent
click around in your browser isvery inefficient.
Like we have way more efficientways to give other computers
access to data.
Now you will see some providerslike my hypothesis is that
(13:17):
LinkedIn sees tools like Apolloas competitive with Sales
Navigator, and they're violatingtheir terms in a way that makes
them have a competitiveadvantage, and so there will be
some players that try to likehold on to their advantage by
through litigation or throughterms of service, but like I
just don't think that's going tobe the way the world is moving,
in the same way that in theearly shift to the web, like
(13:38):
there were some publishers who,like really tried to hold on
tight and say Google, you can'taccess my information.
And openness openness, kind of,kind of won, and so I expect
the same to happen in the future.
Kevin Kerner (13:47):
Doesn't that always happen?
It happened with email,happened with Napster and
Spotify and those things.
It happens in every theconsumer always wins.
Jacob Bank (13:55):
That's above my pay grade, but that's my guess.
Kevin Kerner (13:57):
Well, I will quote you on.
I won't quote you on that, Okay, so I want to dig into some of
the use cases because I've seensome of the stuff that you've
done and, of course, a lot ofthe people who will be listening
to this have are marketers ormight be maybe this marketing
ecosystem, like marketing salesand product.
Can you give some examples ofuse cases using agents that
might be helpful for people Likehow would you use an agent in
(14:18):
those functions?
Jacob Bank (14:20):
Yeah, so this is our tool Related App.
But I don't want to make thisan ad for Related App.
I want to show you kind of thegeneral things that we're using
it for and I'm just going towalk you through, kind of step
by step, many of the use cases Iuse AI for.
So here is one on marketresearch.
So these APIs are changingreally quickly Anthropic Croc
(14:42):
Perplexity and so we set up anAI agent that basically tracks
the changes to their websitesand lets us know when we have a
new model that we need toresearch or if they're
deprecating an old model.
So AI is really really good atchecking a website every day and
tracking changes.
We use AI for our communitymanagement.
So we have a Slack communitywith several hundred members and
(15:03):
we have a bunch of things thatwe want to do.
We want to send them apersonalized welcome message
when they join.
We do kind of communitychallenges from time to time so
we want to tally up their pointsand send them leaderboard
scoring.
So we have a bunch of assistivekind of community management,
community management tools.
We do a lot of AI poweredcompetitive research.
We follow the blogs of all ofour competitors.
(15:24):
We follow the YouTube channelsof all of our competitors.
We follow the pricing updatesof all of our competitors, and
so these typically like eitherthey run on demand like every
time Zapier posts a new YouTubevideo, we get a little summary
of it sent to us or you can runthem once a week or once, I
think.
Our competitor pricing analyzerruns once a month.
It's just like, hey, look atthe pricing pages of our major
(15:44):
competitors and just let me knowif anything's changed in the
last month, because pricingchanges tend not to happen on a
daily basis.
We have a partner program, andso we have a workflow that
screens applicants.
So our partner program isdesigned for agencies and
consultants that help peopleeither with automation or AI or
even function specific likemarketing people either with
(16:05):
automation or AI or even, youknow, function specific like
marketing, and so what we do iswe have a Google Doc with our
program guidelines, and then weautomatically look up the
person's URL and research themonline with perplexity and say,
based on what you found outabout this person and based on
our program guidelines, do youthink they're a good fit for our
partner program?
And then we have a human in theloop step where we just confirm
to double check.
The AI says here's, you know, Ithink Kevin's a great fit,
(16:30):
here's why he runs thismarketing agency, blah, blah,
blah.
And then a human in the loopconfirms it.
Here's another thing that isreally important.
Ai is great at synthesisSynthesis across.
For example, in this use case,I always record all of our
customer calls.
I probably do between 10 and 20customer calls in a given week
(16:50):
and I want to get aggregatedinsights from all of them.
So I have this AI agent thatevery Saturday it looks at all
the customer calls that I did inthe past week and it creates a
single synthesized report thatbreaks down.
I can actually show you theprompt for this one, because
it's kind of cool.
It breaks down.
You know, for each customercall, carefully look at their
industry, their role, their typeof business, what are their key
(17:12):
use cases for AI and automation.
What do they like about RelatedApp?
Where do they struggle withRelated App?
Where do they struggle withRelated App?
What other tools have theytried?
Kevin Kerner (17:18):
And then it synthesizes it into a single
report Digging into Gmail,finding the notes from any
meeting probably prompted bysome tag in the Gmail itself.
Jacob Bank (17:30):
Exactly so.
It just looks for the yearmeeting recap.
Kevin Kerner (17:34):
And then it's just sending it to Slack as a
consolidated note.
Jacob Bank (17:38):
And you can send it to Slack.
You can send it to Slack, youcan send it to email, you can
create it in a Google Doc,whatever works for you.
We have a bunch of kind oflifecycle marketing stuff.
So we send personalized emailsto new users.
We send personalized emails tonew paying customers.
We do outreach to churned usersand we kind of look into, like,
what they were doing in theproduct automatically and try to
(17:59):
get a sense of like, ah, didthey churn because they ran into
an issue or because their usagetailed off over time?
And then I review those emailsbefore they go out to make sure
they're accurate.
We have a bunch of SEOworkflows.
You'll see this folder's got alot of stuff in it.
We create programmatic pages.
We refresh all our blog postsautomatically from time to time
(18:20):
to make sure they're still up todate and accurate.
We automatically publish blogposts for new YouTube videos.
We automatically create postsfor new features that we launch.
Kevin Kerner (18:33):
Is the agent writing to your CMS, your
website?
Jacob Bank (18:36):
Yeah, we use Sanity.
We use Sanity as our CMS and it.
It outputs posts directly there, usually again in draft mode.
So then I go in and edit alittle bit and and hit publish.
Um, we also do a bunch of AIpowered social media marketing.
So whenever I make a YouTubevideo, we automatically
transcribe the recording.
Use that to generate theYouTube description.
Use that to generate theLinkedIn post.
(18:58):
Use that to generate theTwitter posts and the that to
generate the LinkedIn post.
Use that to generate theTwitter post and the Blue Sky
post.
I have a bunch of writingassistants that if I write in a
little after this call, I mightsay, hey, just had a great call
with Kevin.
We talked about this marketinguse case, this marketing use
case, this marketing use case,and then it'll expand that into
a full LinkedIn post for me withmore detail.
I can also feed it thetranscript of our call if I want
(19:19):
to.
I track all my performance onLinkedIn automatically.
So I think this works.
Every month yeah, every monthit looks at all my LinkedIn
posts for the last month andthen it automatically fetches
how many comments and reactionsthey got and it puts out a
tracker and it lets me know.
I can actually show you mytracker because this is pretty
fun.
It shows me like, um, oh shoot,sorry, I clicked on the wrong.
(19:45):
The wrong one I was.
This was the AI writingassistant.
I wanted to show you theperformance tracker, so I'm
showing.
Here is a spreadsheet.
It basically has the url of thepost, the post title, the date
I posted it, the day of week,the time of day, the post
content and then the performance, and then you can look at like
wow, it turns out all of my topsix posts were on Friday,
saturday or Sunday, sosomething's going on about my
content that happens to beresonating on on weekend.
(20:08):
So that was.
That was a very kind of liketactical and detailed, detailed
view, zooming out our major usecase in every marketing function
that I'm aware of.
We have a ton of use case inlifecycle marketing, email
marketing, community marketing,partner marketing, social media
marketing, traditional productmarketing, and we use it for
content creation, we use it foranalysis, we use it for
synthesis, we use it forresearch, we use it for updates,
(20:31):
and basically what this hasmeant is I am the marketing team
for our company, me plus.
Ai is the marketing team, and Istill have a very active role
to play in setting up theseagents, training these agents,
guiding them, giving themfeedback, because, as you've
seen, like a big part of ourmarketing strategy is based on
my relationship with ourcustomers, and I do all the
(20:52):
YouTube videos myself, I answera bunch of the support messages
myself, and that's only possiblebecause I have AI doing all the
busy work sort of behind thescenes.
For me, it's incredible.
Kevin Kerner (21:00):
It's just absolutely incredible.
That was the one thing I wasthinking is, when you're showing
me that stuff number one itreplaces a lot of other tools.
I was thinking Tapleo, you knowwhich is, or any sort of social
media creator tool, schedulertool, whatever.
It replaces a lot of thattechnology because you're just
building an agent space to do it.
It's kind of like citizen-leddevelopment.
Jacob Bank (21:18):
You're doing it yourself versus going to it,
yeah, and this approach won't beright for everyone.
So relayapp is a veryhorizontal tool.
You can think of it as like aspreadsheet.
Like every single company inevery single role uses
spreadsheets right, but you usethem for vastly different use
cases at vastly different levelsof sophistication.
Some people prefer to use aspreadsheet for modeling their
finances, some people prefer touse a dedicated financial
(21:42):
management tool, and so and foreveryone.
For me, I love the flexibilityof working with horizontal tools
because it lets me customizethings just the way I want to
and it gives me much morecontrol.
But it does take more effortbecause you have to tinker with
things a little bit.
You have to experiment.
So me.
So what I would encourageeveryone listening to do is
explore both use case specifictools and horizontal tools, and
(22:05):
my guess is we're going to endup with some blend of both.
You're not going to use Relayas your CRM.
You couldn't even use Relay asyour CRM.
It's not a theorem.
It doesn't have the right datastorage.
Like you got to use a CRM asyour CRM you.
You can definitely use Relay toaugment some of the
capabilities that might bemissing in your CRM, like
logging certain kinds ofcalendar events or doing a
certain kind of enrichment.
Kevin Kerner (22:22):
Well, you also had some emails in there.
They're going out through Gmail, but it's just amazing to see
that that's mind blown on someof this stuff.
Jacob Bank (22:31):
Can I show you one other thing that's super cool.
This is something that weliterally just built today.
So we've had a lot of successwith this combination of
LinkedIn posts and then liveevents like webinars.
This has been a.
It's a motion that really workswell for us, because the
LinkedIn post gets a lot of topof funnel traction and the
webinar really teaches peoplehow people to the webinar Like I
(22:58):
looked at Zoom webinar, Ilooked at Luma, I looked at all
these dedicated tools and theyall just felt kind of heavy and
kind of wrong for me.
Like cause.
I wanted this feeling ofpersonal interaction.
I didn't feel like you weregiving your email to some
faceless, you know newsletterwriter, and so I eventually I
tried to do like a public Googlecalendar.
That didn't really work, and sowe decided what if we built our
own custom landing page forevents and we used a tool called
(23:22):
Lovable, which folks haveprobably heard of?
Lovable, bolt and VZero theseare probably the three most
popular natural language-basedweb app creators and I want to
show you this tool that we wereable to build in about two hours
in Lovable.
It's a totally custom websitethat says here's when the
Relayapp live sessions arecoming up.
It shows you the list ofupcoming sessions.
(23:44):
There's one tomorrow, there'sone March 20th, there's another
one on April 9th.
You can click to expand to seeother future sessions.
You can select a session andenter your email.
Hit, sign me up.
And now.
This is lovable on the frontend and on the back end, it's
using a relayapp workflow toautomatically add that email
(24:06):
address to the calendar event,automatically adding that email
address to our loops emailmarketing tool, automatically
sending a personalized email.
And then this is something thatour head of product, who is
working this out just today Ithink this is a great touch Like
you can watch previous sessionsand that links right to our
YouTube channel where all of ourprevious sessions are posted.
And this is crazy because noengineer was involved in this.
(24:29):
It's totally custom and exactlyfits the need and the use case
that we have.
It took two hours to set up.
All like no code tools.
Lovable on the front end andthen relayapp on the back end.
It actually works.
It actually works.
We are actually going to usethis in production and send that
link to thousands of people.
Kevin Kerner (24:46):
Amazing.
Yeah, I've been on the no-codething for a few years now.
We built our I think I messagedyou about us using Glide Like
we use Glide to the max man, it,man.
It is such an easy tool to useand it's all spreadsheet based,
so I'm really interested to seehow Relay can work with it,
because we can get data in andout of it.
Jacob Bank (25:02):
One funny thing Go ahead, oh go ahead, Go ahead.
Kevin Kerner (25:05):
Well, I was going to, I was going to, I was going
to, but Glide is a tool that's alittle bit more cautious about
access, I think.
But I was going to, and whatwe're trying to do on the Mighty
and True side is find peoplethat can build in these no-code
tools for our customers.
Like, the makeup of a marketeris different now.
It seems like it's this sort ofcreative growth, entrepreneur
(25:27):
role, but you have to be able touse or learn these tools, which
are pretty accessible.
You can easily learn them.
I'm curious what you think themakeup of a like.
If you were a marketer, likewhat would you?
Or if you were going to hire amarketer or replace you in this
stuff?
If you were a marketer, or ifyou were going to hire a
marketer or replace you in thisstuff, what would you want them
to do?
What would their makeup be?
What type of person is that?
Jacob Bank (25:45):
So first, I don't know, because it's been hard for
me to know what marketer tofind, which is why I've been
doing it myself.
But let me tell you some of thethings I would think about.
First, I would think about whatmarketing strategies are likely
to work better in the futureand which marketing strategies
are likely to work worse in thefuture.
So, if you're used to writingSEO optimized blog posts for
(26:08):
really competitive keywords oof,I'm not sure how future-proof
that's going to be Like AI isjust going to get really, really
.
It's going to be hard todifferentiate there.
Ai is going to get really goodat writing the content and the
companies that already have thedomain reputation and the topic
authority are going to winanyway.
If you're really good atwriting cold emails, like ah,
are cold emails going to work ina world where AI generated cold
(26:28):
emails are a dime, a dozen?
I don't know.
And so what marketingstrategies I think are likely to
work in the future?
I think marketing strategiesthat enable the customer base to
build a real relationship withthe team and the builders.
I think those are going to bedurable into the future, because
people want to buy a productthat's from people they trust
and that they know.
(26:48):
And that's why I actuallyexperimented with fully
automating my YouTube videos andI got them pretty good.
It could click around thescreen, it could do the
voiceover, it could generate thescript.
But I got some feedback thatwas like hey, we're missing you
in the video, because they seeme on YouTube, they see me on
LinkedIn, they see me doingcustomer support, and so things
(27:10):
that are based on building trustand a relationship with the
team.
So certain kinds of socialmedia, certain kinds of content
creation and then also kind ofcommunity led channels, because
in a world where there's AI youknow AI generated content
everywhere it's kind of likewhen I need to find a plumber
for my home, I can't really goto Yelp anymore, because I just
don't really trust Yelp anymore.
It's so flooded with reviews Idon't know if they're meaningful
(27:30):
.
So what do I do?
I ask my four neighbors I saydo any of you have a good
plumber?
That you've worked?
And I think that's the world.
Software is going to go the way.
Software is going to go too.
And so I'd want a marketer whoreally has great instincts in
building community, buildingword of mouth, building
advocates, cultivating advocates.
So that'd be one thing I wouldlook for.
Are they aligning themselves tothe kinds of marketing channels
(27:50):
that I think are likely to workdurably in the future?
That's number one.
Number two are they totallyunafraid of playing with new
tools and learning new things,like and in fact, their first
instinct should always be isthere a tool that could help me
with this?
And you know, there are peoplethat we've worked with that are
like that, and there are peoplethat we work that are not like
that.
They're like I know my playbook, I know my way of doing things,
(28:11):
I'm doing it.
So I really need people whosefirst instinct is let me try to
find the best tool that'savailable right now, and are
always updating their mentalroster of like here are the
tools that are great, here'swhat AI can do, here's what AI
can't do.
And then number three is I needsomeone who's a little bit more
(28:33):
of a generalist, because I thinkcompanies are going to be
smaller in the future.
Cursor 150 million ARR with 22people.
Bolt 40 million ARR with 22people.
Bolt 40 million ARR with 17people.
Lovable 20 million ARR with 15people.
I might be messing up the exactnumbers, but, like, spirit is
definitely right.
(28:53):
There are just going to besmall, small companies with
player coach generalists thatare harnessing many AI agents to
do a lot, and so those are kindof the three things I'm looking
for.
I don't really care if you'vepreviously written a blog post
that ranked on Google.
I don't really care if you'vepreviously run an amazing PPC
campaign.
I want to see that you canlearn the tools that are going
to enable you to do the rightforward thinking thing.
Kevin Kerner (29:12):
It's a certain amount of curiosity and
fearlessness and desire toautomate and sort of systems led
thinking that I think is goingto win out in the end.
This is awesome.
We could go on.
I want to ask you reallyquickly, so we have a few
minutes.
Here is your marketing story.
You mentioned some of it like alot of this has been
founder-led and those sort ofthings, but you have this
(29:33):
amazing video story.
What do you think's real?
How have you really done whatyou've done with the brand by
yourself?
Where have you seen success?
I know some of it's been in theYouTube and on the video side
of things.
I'd love to hear just quicklytalk about that success.
Jacob Bank (29:47):
Let me tell you a quick story, and it has been
failure after failure afterfailure.
It's been really, really hard.
So the book that really shapedmy thinking in the early days of
the company was the mom test.
I highly recommend it toeveryone.
If you're in the early userresearch phase or you're looking
for your first five or 10customers, that book basically
outlines a methodology by whichyou can talk to people and get
(30:08):
real feedback, Because nicepeople like Kevin if I say, hey,
Kevin, I've been pouring myheart and soul in this thing for
the last six months what do youthink?
What are you going to say to me?
You're going to say I love it socool.
Let me know when it launches.
So there's a methodology tokind of asking questions in a
way that will give you the truthabout whether you're onto
something useful.
And then I was able to get sortof our first five-ish, 10-ish
(30:29):
customers just kind of throughmy network and using that
interview-based methodology.
Then the book that reallyshaped my thinking for the next
phase was a book called Traction.
It was written by I think hisname is Gabriel Weinberg, the
founder of DuckDuckGo.
And that basically outlines amethodology that says there's 20
or 25 different channels, fromlive events to paid ads to
(30:50):
social media and what you shoulddo is, at every phase of growth
, you should experiment with thefew channels that you think are
likely to work at that phase ofgrowth, and then you're gonna
have to rerun this exercise manytimes as you reach different
phases of growth.
So at the first phase of growth, we were going from, like you
know, the zero to 10 customers.
The channels we picked wereintroductions through my network
, finding people who had thisproblem on Reddit and
(31:11):
interacting with them, and doinga little bit of like outbound
on LinkedIn, based on people wholiked adjacent products, and
those got us from one to 10users.
But I quickly realized, like,oh shoot, those are not actually
gonna get us for one to 10customers.
Going to individual Redditthreads is not gonna get us from
10 to 100.
Like that's not gonna work.
And so, from 10 to 100, wethought what could work for us?
(31:31):
Partner marketing, because weintegrate with a bunch of other
tools Airtable Notion,smartsuite, et cetera.
So if they list us in ourmarketplace, if we do a webinar
with them, if we do a jointtemplate with them we get some
traction.
We built some basic bottom offunnel SEO blog posts.
We tried a bunch of stuff thatdidn't work that I won't even
mention here.
We tried outbound, we tried afew other things that totally
(31:52):
didn't work.
And then those tactics got usfrom 10 to 100.
But we were like, oh shoot,those are not going to get us
from 100 to 1000.
We sort of tapped out what wesee as the opportunity and
bottom of funnel blog posts.
We've kind of seen what theceiling is with partner
marketing not going to get usfrom 10 to 100.
And so we rethought and thought, okay, what are the things that
are going to compound with ouruser base and get us from 100 to
(32:13):
1000?
And we decided it was going tobe a combination of social media
.
It was going to be acombination of social media,
video content and then kind ofcommunity and partnership
facilitation of word of mouth.
And I've been posting two tothree times a week on LinkedIn
and YouTube for the last yearand had no results.
(32:34):
No results for the first ninemonths.
I can show you the chart of ourYouTube channel.
So here's our YouTube channel.
I'll kind of describe it andI'm not saying we have some
amazing channel.
Now we're closing it on 5,000subscribers, which is decent at
our pace.
But if you go to the chart ofthe historical subscribers, I'll
(32:56):
show you the last year Ourchannel kind of just turned one
year old and I'll show you thelast year Our channel kind of
just turned, just turned oneyear old and I'll show you the
subscriber count.
Sorry for navigating through thewhole YouTube UI here.
You can see here.
If you look at the last year,it was just flat, completely
flat.
From March to basicallyDecember Instinctively got up
(33:19):
like six or 700 subscriberspainful, painful, painful.
And then all of a sudden wekind of found the thing that
clicked.
In February my videos got a lota bit better.
I was cross promoting themeffectively from LinkedIn.
We got a bit of word of mouth,the algorithm started picking us
up and now in the last monthwe've gone from 1000 to 5,000
subscribers, and LinkedIn is asimilar story.
(33:39):
Let me show you this.
Did your?
Kevin Kerner (33:41):
volume of videos.
Jacob Bank (33:44):
Nope, nope.
Volume didn't go up, it waspretty much two a week.
Kevin Kerner (33:48):
Great for you for sticking that out.
That is unbelievable.
Jacob Bank (33:51):
Some people would just go oh well, video's not
working.
Got Gotta let it go, that'sgreat to see and this is where
you know you really have totrust your marketing or
entrepreneur intuition.
Sometimes the right answerskeep going and sometimes the
right answer is switch, like Iknew.
Outbound was not working and wehad to switch.
So look at this on LinkedIn Iwas posting three times a week
(34:11):
this whole time Nothing, nothing, nothing, nothing, nothing,
nothing, nothing, nothing,nothing, nothing, nothing,
nothing.
I had one post that hitDecember 31st, 23rd, the day
before Christmas, and then Ikind of figured it out and now
I've got 5 million impressionslike 4.7 million of them were in
the past 90 days, and I haveyou know 800,000, 750, 600.
(34:33):
And, and it goes, it goes down.
Kevin Kerner (34:35):
Amazing.
I mean, that's.
That's one of the, that's oneof the most clearly product led
growth examples I've ever seen.
It's really fantastic.
Jacob Bank (34:42):
And so I don't know what the lesson from that is
exactly.
I think it's a combination offind a channel that you believe
in, that you think is going towork for you, and then just be
ridiculously persistent and then, once it pays off, double down
on it.
So now LinkedIn is working fortop of funnel, youtube is
working for mid funnel, and sonow I can just focus on those
two channels.
Kevin Kerner (35:02):
That's awesome.
What a great, a great story.
Okay, there's so many thingsI'd want to ask you beyond that.
We're running up against time,so I wanted to do one more thing
.
I haven't tried this before ona podcast.
This is the first time I'vedone it on this, on this podcast
series.
So I put a question, intoperplexity you would probably.
I'm not going to tell you whatthe prompt was, because you're
such a prompt expert, but it wassomething around.
Hey, I'm meeting with Jacob,his founder.
(35:23):
You know we're talking about abunch of different things here.
Give me one question I shouldask him that is provocative, and
so I'm going to hit send onthis and see what it gives back
to me.
Hold to me, hold on a second.
I should probably just grab myscreen, but I'll just give it to
you.
Oh gosh, if you could go backto your time at Google and undo
one decision you made as aproduct leader, what would it be
(35:44):
and how do you think it wouldhave shaped your entrepreneurial
journey differently?
That's completely unedited.
Jacob Bank (35:47):
It's actually a great question and I have an
exact answer for it.
It wasn't a product decision,but it was a way I operated that
really handicapped me for thefuture.
I was the product lead of Gmailfor six years and I ended that
time with 800 Twitter followersInteresting.
That's what was I doing.
What was I thinking?
(36:07):
A?
I was way too insulated fromour real users when I was in
that position and I was lookingat all the data and I was behind
the scenes and I was, you know,interviewing small groups of
users and making decent productdecisions.
But if I look at the bestproduct leaders, they are the
public face of their product.
They are on Twitter, they're onLinkedIn, they're making demo
videos, they're doing interviewsand they're directly
(36:30):
interacting with users and, forwhatever reason, google culture
doesn't really value that orincentivize it or encourage it,
because and I can see why likeGoogle doesn't want a bunch of
loose cannon PMs talking aboutstuff on Twitter because you can
end up with regulatory issuesand PR disasters and stuff.
But when I look back at it, Iwas like what was I thinking?
(36:51):
I was in, I had this platformwhere we had 2 billion users and
I could have interacted withthem more, both by way of
getting their feedback andmaking the Gmail product better,
and then I would have had thatskill set and that personal
brand that I could have broughtthrough to my startup, because I
now am up to whatever 20,29,000 followers on LinkedIn.
I've still got nothing onTwitter, by the way, but I had
(37:14):
to learn all of this reallypainstakingly from scratch, and
the only thing that actuallyspurred me to do it is my
company's future depended on it.
So I was like, okay, I betterlearn Now.
It's not optional, it'srequired.
And so what I would recommendto everyone in any role, whether
you're at a big company or asmall company or agency or
consultant.
In the way the world is moving,where people want to build
(37:35):
authentic relationships with thebuilders or creators of the
products that they're using,like, you got to get out there
and you got to just put yourselfout there.
Make tweet, make LinkedIn posts, make YouTube videos, host
webinars.
Don't be afraid of lookingsilly.
That was the biggest mistakefor my time at Google.
Kevin Kerner (37:50):
Amazing.
That's an amazing insight.
Great question by perplexity.
I take no credit, but man,that's amazing.
Yeah, and there's a certainamount of courage.
You need to put yourself outthere.
I mean, I've been doing thisfor a long time, so it's like I
think the same thing.
It's like, golly, I was doingall this stuff.
What was I doing the whole time?
Because it was behind thescenes working.
It's crazy.
Yeah to you, and I'm such a fanof.
(38:12):
I think part of what makes yourbrand work is you're so
enthusiastic and so willing tolike share with the community.
And it's real.
You know you really want to dothis stuff.
So it's just, I feel, veryfortunate.
Jacob Bank (38:22):
I've always been a productivity nerd.
I absolutely love this stuff,and I think the best service I
can provide is to show the real.
Show the reality, becausethere's so many demos out there
that are very curated, and so Ithink you've you've seen a
couple of my my live sessions.
Like I make mistakes all thetime, I mess up my prompt, I
(38:43):
don't, I don't attach the rightinformation, but that's the
reality and so, and so I justwant to show people yeah, I make
mistakes too, and here's how Iwork through it when I do,
because it's a when, not an if.
Kevin Kerner (38:53):
Yeah, it's fantastic.
Well, there's a bunch of waysto get ahold of you.
What's the best way to followwhat you're doing in this space?
Jacob Bank (39:02):
LinkedIn is the best beginner way.
I post pretty broad contentabout how to use AI to help you
get stuff done.
None of it is too Relayappspecific, really, it's just
general content on ways you canuse AI.
That's the easiest.
And then, if you're ready to goone step deeper, I post very
detailed YouTube videos 45minutes or an hour about how to
(39:22):
set up an AI agent for yourself.
So those are the two best waysto follow along.
But please feel free to DM me.
You know ping me directly, dmme, email me at jacob at
RelatedApp Like.
I would love to hear your usecases, your feedback and any
ideas of how we can make AI workeffectively for you.
Kevin Kerner (39:37):
That's fantastic, and I've been through the
YouTube videos, the builds, withyou.
They're very accessible.
So if you're a marketer whowants to get into the stuff and
you're thinking I can't do this,it is really easy.
It's not hard to do.
You just need to give it a tryand you do a great job at
walking people through it errorsand all all the way, so it's
fantastic.
Yeah, Jacob, thank you so much.
I am really happy to havetalked to you.
(39:57):
We'll have to do a recap a yearfrom now to see how this all
turns out.
Jacob Bank (40:00):
I think the fun thing is yeah, a year from now,
I'm sure two thirds of what I'vesaid will look painfully out of
date, but that's what's funno-transcript and I'll be a
different person, the companywill be different and AI will be
(40:20):
in a different state.
Kevin Kerner (40:21):
For sure.
Okay, great seeing you, Jacob.
Jacob Bank (40:23):
Yeah, thanks.
Kevin Kerner (40:23):
Kevin.
Hey everyone, I'm Kevin Kerner and this is Tech
Marketing Rewired.
I just had the chance to sitdown with Jacob Bank and I've
got to say this one was a realpleasure.
I've been watching Jacob'sYouTube videos for a while now
and I've always appreciated hisenergy, his curiosity and his
willingness to jump into the AIchaos with the rest of us.
In this episode, he breaks downthe real difference between AI
agents and traditionalautomations.
(00:21):
He also gives us abehind-the-scenes look at how
his team at Relay is usingagents in their own marketing,
and let me tell you, this isreally smart stuff.
Plus, jacob shares a story ofhow he finally cracked YouTube
and LinkedIn his own personalmarketing after a year of
testing, tweaking and totalpersistence.
And don't miss the AI rouletteat the end of this one.
It's a good one.
Let's dive in.
(00:41):
This is Tech Marketing Rewired.
Jacob Bank (00:50):
Well welcome, jacob.
I'm glad to have you on theshow.
I'm super excited it's going tobe a fun conversation.
Kevin Kerner (00:54):
So the first thing I wanted to do is just have you
introduce yourself, talk alittle bit about your journey
and then I definitely want youto say what you're doing now at
Relay.
Jacob Bank (01:04):
Yeah, I'll start with what I'm doing now and then
I'll go back to quickly how Igot here.
I'm currently the founder andCEO of Relayapp, which is a
platform that enables everyone,including non-technical users,
to build AI agents, automationsand workflows.
My road here was kind ofwinding.
Once upon a time, I was anartificial intelligence
researcher.
I thought I was going to pursuea PhD and become a professor.
(01:24):
One day, I ended up droppingout of my PhD to start my first
company, which was an AI-powereddigital calendar called Timeful
.
We got acquired by Google and Ispent six years at Google where
I led product for Gmail, googleCalendar and several other
collaboration and productivitytools, and so I basically spent
my whole career at thisintersection of AI and
collaboration and productivitytools, and that's what's taken
(01:45):
me to where we are now, wherethe stuff actually works.
Kevin Kerner (01:47):
finally, Well, I think we've now discerned that
you're a lot smarter than I am.
Far far from it, but I pretendto be on YouTube.
Yeah, that's really awesome.
Okay, I want to talk firstabout.
I want to.
You know, we have a lot ofpeople that are hearing all
about agents and agentic AI andall these things.
How do you describe or maybedefine what you see as AI agents
(02:09):
and maybe even the differencebetween that and agentic AI?
Jacob Bank (02:13):
It's easy to get lost in the terminology and I
don't think it matters that much.
So let me explain it how Ithink about it as a user of
these tools.
There's three basic ways that Iuse AI every day.
Number one is I go to a chatbot.
That's like ChatGPT or ClaudeProbably most people have tried
that by now.
Grok, pick your favorite.
You go talk to it and say likehey, I'm about to talk to Kevin
(02:36):
on Mighty and True, do someresearch about him for me.
So I show up prepared.
That's a chatbot.
You're in a window writing textmessages to an AI and it's
responding.
The second model is a co-pilot.
A co-pilot is you're in thetool that you normally use every
day and the AI is somehowhelping you do things.
So, for example, one of thefeatures I worked on on Gmail
was that as you're typing outyour email, the AI will suggest
(02:58):
what you might want to say.
Next, if you write code, youmay have used a tool like Cursor
that suggests what code youmight want to write.
Or if you've used Google Docsor Microsoft Word, they now have
co-pilots where they can helpedit your writing or suggest new
ideas.
So that's a co-pilot.
A co-pilot is you're in a tool.
You're not in a chatbot.
You're in a dedicated tool toproduce something and there's an
(03:19):
AI experience within that toolthat's helping you do it.
And then the third class of AItools I use every day are those
that are happening behind thescenes.
I'm not actively chatting withsomething in a chatbot.
I'm not actively using it in aco-pilot.
The AI is just doing work on mybehalf.
And that's where people havethis kind of technical debate
(03:40):
about what's an AI automationversus an AI workflow or what's
an AI agent.
But I think the fundamentalthing to keep in mind is that
there's this class of AI thatcan work on your behalf behind
the scenes on things thatpreviously would have required
human level intelligence.
For people who are reallyinterested in the technical
distinction, an AI automationwould be like a single step,
like generate a summary of thispodcast.
(04:02):
An AI workflow would bechaining multiple of those steps
together, like generate asummary of this podcast.
An AI workflow would bechaining multiple of those steps
together, like generate asummary of this podcast and then
create a post on LinkedInpromoting it, where we have
multiple different AI actionsthat are chained together into a
workflow.
And an AI agent is technicallya little different.
With an AI agent, you just giveit a goal and a set of tools
and say agent, go do work for me.
(04:24):
And so I would give the agent agoal.
That's like hey, I want topromote this podcast episode.
The tools you have access toare you can read transcripts,
you can write LinkedIn posts andyou can write Twitter posts.
Go, promote it as you see fit.
And so automation, workflow,agent.
The line between those threecan be very blurry and I find
the you know the debates aboutthe distinction between them on
(04:45):
Twitter and on LinkedIn kind ofbeside the point and overly
pedantic.
So the way I would think aboutan AI agent I wouldn't worry
about the difference between anagent and a workflow and an
automation that's just notpractically relevant for any of
us.
The thing I would think aboutis an AI agent is something that
can work on my behalf behindthe scenes, without me directly
chatting with it or using it asa co-pilot, and it can do things
(05:06):
with human level ofintelligence that I previously
would have needed a person aperson to do.
And that's a perfect examplelike promoting writing a really
good promotional post onLinkedIn about a podcast episode
.
Kevin Kerner (05:17):
Is there a reason why now it's getting so much?
Because these workflows havebeen around for a while and I
guess there's been ways to diginto AI in these workflows.
Is there a reason why now, ifwe're just in a hype cycle, or
is there is there a technicalreason why now they're becoming
more accessible or more peopleare just more interested in them
?
Jacob Bank (05:33):
Models are way, way better.
The models are way, way better,like, even basic things like
extract invoice information froma PDF barely worked two years
ago and now works flawlessly.
Things like write a LinkedInpost in my voice two years ago
didn't really work.
You got this like AI generatedslop with all the em dashes and
(05:54):
all the you know words thatpeople poke fun at that chat
should be to use, whereas nownow it's, the models can do it
very well.
So the models have improved, Ithink, along two dimensions that
I think are important foreveryone to know about.
One is their core quality at asingle task has improved.
They are way better atextracting information from PDFs
.
They're way better atsummarizing long form
(06:15):
transcripts.
They're way better atgenerating prose.
But you've probably also heardof this other kind of models
called reasoning models, andthose reasoning models are able
to do sort of multi-step chainof thought actions where they
have to make a plan about how toachieve something and then
deliver on it, and they've alsogotten way better at that.
And it's that second, that leapinto reasoning, that has
(06:36):
enabled people to shift fromvery well-defined AI,
automations and workflows intothings that feel a bit more
agentic.
And when I say more agentic, Ijust mean you're giving the AI
more latitude to figure out theplan of how it wants to
accomplish its goal.
Kevin Kerner (06:50):
I watch a ton of your videos and I always find it
funny when you're building someof the build with me and you
put something in the AI and itbrings back a result and you're
surprised how good it is.
You're like, whoa, wow, thatgot that thing.
Jacob Bank (07:03):
Honestly, this happened to me literally today,
where I was like I was doing asession today where we were
going to do some promptengineering and the idea was
we'd go into the session andthen it would get it wrong the
first time, we'd try again.
It would get it wrong.
The second time We'd try again,it would get it wrong the third
time We'd try again.
And I wanted to show peoplekind of here's my progression of
how I test and iterate to makesure I get a prompt.
(07:24):
That's really good and thesession kind of kind of blew up
because I just did it reallywell the first time for all my
test cases and you know I foundsome different.
You know I found some ways tokind of push it into areas that
it wasn't quite able to do.
I gave it harder and hardertasks.
But I'm like I'm in this everyday and every day I wake up
surprised by how good somethingis.
Kevin Kerner (07:45):
Wow, that's wild.
So you can see the improvement,maybe not day to day, but week
to week.
Jacob Bank (07:53):
Definitely week to week, and so I would encourage
everyone.
I'm not one of these people whothinks all human work is going
to be irrelevant by the end of2025.
And so I always try to presenta pragmatic, realistic view of
what AI is capable of today.
But I would really discount theopinions of people who say it's
all hype, it can't do anythingLike.
(08:13):
Those are people who are nottrying hard enough, because if
you find the right use cases andyou find the right tools and
you put in a bit of time togetting it good, it's really
amazing.
And one of the things that I'mtrying to push back on is a lot
of times people will say is likeI tried to get Chachi BD to
write me a blog post and itwasn't good, and I'm like, well,
what did you ask it to do?
And you say, like I gave it.
(08:34):
You know one sentence of theblog post I wanted to write.
Did you give it any context onwhat key points you wanted to
make?
Did you give it any of yourpast blog posts so it knows your
style?
Did you tell it anything aboutyour ICP and the kind of
language those people use?
Did you give it any contentfrom customer calls about issues
that might be top of mind.
Like imagine if you asked ajunior in college intern from a
one sentence prompt to justwrite a blog post.
(08:55):
They're not going to writesomething good based on that
information and so you reallyhave to think of like the mental
model right now for me that Ifind most helpful is imagine a
really, really bright sophomorein college intern.
That's like a good startingpoint and but that changes every
week.
If we talk again in threemonths, it might be.
Imagine a really good new gradand if we talk again in three
(09:18):
months, it might be.
Imagine a really good contentmarketing professional who's
been in the in the game for fiveyears.
Because it's the pace of the.
The evolution is incredible.
Kevin Kerner (09:25):
Yeah, the people that don't get really good at
that soon are going to be kindof left behind or they're going
to have to learn really fast,you know.
The other thing I was thinkingabout in terms of the maturity
of your tool and other tools isI don't know if you agree with
this is that it seems like theability to dig into other
applications.
Other applications are sort ofretrofitting to, and it used to
be API calls and you know itstill is to some degree, but man
(09:46):
, it's getting so much easier todig into other applications and
the more you build thatecosystem out of all these
different things that can beconnected with and have this
agent rules engine inside of it,it just it seems like it could
be really powerful.
I would put that thing ineverything you know as much as
you can get it into it's hardwork.
Jacob Bank (10:03):
But here's my current take on what it takes.
For an AI agent to be valuable,it needs three things.
Number one it needs to connectto your actual tools, because if
all it does is spit out sometext that you then have to
reformat and copy paste like,it's just not going to be able
to do work on your behalf if itcan't deeply connect to your
tools, both to read data andunderstand context and also to
take action.
(10:24):
Number two this is something Ifirmly believe the AI has to
show you a plan of what it'sgoing to do before it does it,
Especially in these early days,as we're all just building trust
and trying to figure out whatcan it do, what can it not do,
how it shows you the plan in thesame way that a great new
employee says hey, Kevin, thisis the task you gave me.
Here are the five things I'mgoing to do, and I'm going to
(10:44):
get this on your desk byThursday.
Does that look right?
And then you say, yes, no,we're like oh, also check this
source or add this step here.
And then number three is as theAI is executing, things will
come up and it'll need help it.
But, like, tell me if you wantsomething different.
And so if you can connect toall the tools, if you can be
(11:08):
very transparent and trustworthyabout what your plan is and if
you can loop in the human in theright way, as you're doing, I
think that's how you build auseful AI agent.
And on that first point that youwere making about tool
connections, there's two basicapproaches.
One is what's called an APIthat you mentioned, and this is
basically a special set ofcapabilities that software tools
offer for other software tointeract with them.
(11:29):
So you know, Salesforce willexpose an API that lets other
tools look up records and createrecords and search over records
.
But many, many tools don't havean API or don't have a complete
API, and so for those tools,you basically need to just go in
the browser and mimic a humanclicking around.
Now, in my experience right now,I have yet to succeed with a
(11:52):
browser clicky tool.
So there's Operator from OpenAI.
Manus got a lot of hyperecently.
There's Cloud Computer Use.
I would love to see use caseswhere you're using it for real,
but I have not successfully usedOpenAI Operator for anything
yet.
But that doesn't mean you canwrite it off forever.
I mean right now I'm buildingall of my workflows based on
(12:13):
APIs, with some light webscraping, but I totally can
believe that six months from now, that I would be able to fold
in Operator to some of theseagentic use cases that I use,
that's cool.
Kevin Kerner (12:23):
Operator to some of these, these agentic use
cases that I use, that's cool,but you and you haven't seen any
end applications cutting offaccess or limiting access, as
they want their own AI tools.
Jacob Bank (12:31):
The one case that I have seen is that LinkedIn has a
pretty locked down terms ofservice around what automated
actions you can take, and so Idon't know the details of it,
but they got into a little bitof a dust up with Apollo and
Seamless and a couple others acouple weeks ago, and so I think
, in general, most platformsbenefit from being open and
(12:54):
having good APIs, and so Iexpect that most companies will
invest more in their APIs andnot less, as AI agents become
maybe the primary thing.
That's interacting with thesetools like making an AI agent
click around in your browser isvery inefficient.
Like we have way more efficientways to give other computers
access to data.
Now you will see some providerslike my hypothesis is that
(13:17):
LinkedIn sees tools like Apolloas competitive with Sales
Navigator, and they're violatingtheir terms in a way that makes
them have a competitiveadvantage, and so there will be
some players that try to likehold on to their advantage by
through litigation or throughterms of service, but like I
just don't think that's going tobe the way the world is moving,
in the same way that in theearly shift to the web, like
(13:38):
there were some publishers who,like really tried to hold on
tight and say Google, you can'taccess my information.
And openness openness, kind of,kind of won, and so I expect
the same to happen in the future.
Kevin Kerner (13:47):
Doesn't that always happen?
It happened with email,happened with Napster and
Spotify and those things.
It happens in every theconsumer always wins.
Jacob Bank (13:55):
That's above my pay grade, but that's my guess.
Kevin Kerner (13:57):
Well, I will quote you on.
I won't quote you on that, Okay, so I want to dig into some of
the use cases because I've seensome of the stuff that you've
done and, of course, a lot ofthe people who will be listening
to this have are marketers ormight be maybe this marketing
ecosystem, like marketing salesand product.
Can you give some examples ofuse cases using agents that
might be helpful for people Likehow would you use an agent in
(14:18):
those functions?
Jacob Bank (14:20):
Yeah, so this is our tool Related App.
But I don't want to make thisan ad for Related App.
I want to show you kind of thegeneral things that we're using
it for and I'm just going towalk you through, kind of step
by step, many of the use cases Iuse AI for.
So here is one on marketresearch.
So these APIs are changingreally quickly Anthropic Croc
(14:42):
Perplexity and so we set up anAI agent that basically tracks
the changes to their websitesand lets us know when we have a
new model that we need toresearch or if they're
deprecating an old model.
So AI is really really good atchecking a website every day and
tracking changes.
We use AI for our communitymanagement.
So we have a Slack communitywith several hundred members and
(15:03):
we have a bunch of things thatwe want to do.
We want to send them apersonalized welcome message
when they join.
We do kind of communitychallenges from time to time so
we want to tally up their pointsand send them leaderboard
scoring.
So we have a bunch of assistivekind of community management,
community management tools.
We do a lot of AI poweredcompetitive research.
We follow the blogs of all ofour competitors.
(15:24):
We follow the YouTube channelsof all of our competitors.
We follow the pricing updatesof all of our competitors, and
so these typically like eitherthey run on demand like every
time Zapier posts a new YouTubevideo, we get a little summary
of it sent to us or you can runthem once a week or once, I
think.
Our competitor pricing analyzerruns once a month.
It's just like, hey, look atthe pricing pages of our major
(15:44):
competitors and just let me knowif anything's changed in the
last month, because pricingchanges tend not to happen on a
daily basis.
We have a partner program, andso we have a workflow that
screens applicants.
So our partner program isdesigned for agencies and
consultants that help peopleeither with automation or AI or
even function specific likemarketing people either with
(16:05):
automation or AI or even, youknow, function specific like
marketing, and so what we do iswe have a Google Doc with our
program guidelines, and then weautomatically look up the
person's URL and research themonline with perplexity and say,
based on what you found outabout this person and based on
our program guidelines, do youthink they're a good fit for our
partner program?
And then we have a human in theloop step where we just confirm
to double check.
The AI says here's, you know, Ithink Kevin's a great fit,
(16:30):
here's why he runs thismarketing agency, blah, blah,
blah.
And then a human in the loopconfirms it.
Here's another thing that isreally important.
Ai is great at synthesisSynthesis across.
For example, in this use case,I always record all of our
customer calls.
I probably do between 10 and 20customer calls in a given week
(16:50):
and I want to get aggregatedinsights from all of them.
So I have this AI agent thatevery Saturday it looks at all
the customer calls that I did inthe past week and it creates a
single synthesized report thatbreaks down.
I can actually show you theprompt for this one, because
it's kind of cool.
It breaks down.
You know, for each customercall, carefully look at their
industry, their role, their typeof business, what are their key
(17:12):
use cases for AI and automation.
What do they like about RelatedApp?
Where do they struggle withRelated App?
Where do they struggle withRelated App?
What other tools have theytried?
Kevin Kerner (17:18):
And then it synthesizes it into a single
report Digging into Gmail,finding the notes from any
meeting probably prompted bysome tag in the Gmail itself.
Jacob Bank (17:30):
Exactly so.
It just looks for the yearmeeting recap.
Kevin Kerner (17:34):
And then it's just sending it to Slack as a
consolidated note.
Jacob Bank (17:38):
And you can send it to Slack.
You can send it to Slack, youcan send it to email, you can
create it in a Google Doc,whatever works for you.
We have a bunch of kind oflifecycle marketing stuff.
So we send personalized emailsto new users.
We send personalized emails tonew paying customers.
We do outreach to churned usersand we kind of look into, like,
what they were doing in theproduct automatically and try to
(17:59):
get a sense of like, ah, didthey churn because they ran into
an issue or because their usagetailed off over time?
And then I review those emailsbefore they go out to make sure
they're accurate.
We have a bunch of SEOworkflows.
You'll see this folder's got alot of stuff in it.
We create programmatic pages.
We refresh all our blog postsautomatically from time to time
(18:20):
to make sure they're still up todate and accurate.
We automatically publish blogposts for new YouTube videos.
We automatically create postsfor new features that we launch.
Kevin Kerner (18:33):
Is the agent writing to your CMS, your
website?
Jacob Bank (18:36):
Yeah, we use Sanity.
We use Sanity as our CMS and it.
It outputs posts directly there, usually again in draft mode.
So then I go in and edit alittle bit and and hit publish.
Um, we also do a bunch of AIpowered social media marketing.
So whenever I make a YouTubevideo, we automatically
transcribe the recording.
Use that to generate theYouTube description.
Use that to generate theLinkedIn post.
(18:58):
Use that to generate theTwitter posts and the that to
generate the LinkedIn post.
Use that to generate theTwitter post and the Blue Sky
post.
I have a bunch of writingassistants that if I write in a
little after this call, I mightsay, hey, just had a great call
with Kevin.
We talked about this marketinguse case, this marketing use
case, this marketing use case,and then it'll expand that into
a full LinkedIn post for me withmore detail.
I can also feed it thetranscript of our call if I want
(19:19):
to.
I track all my performance onLinkedIn automatically.
So I think this works.
Every month yeah, every monthit looks at all my LinkedIn
posts for the last month andthen it automatically fetches
how many comments and reactionsthey got and it puts out a
tracker and it lets me know.
I can actually show you mytracker because this is pretty
fun.
It shows me like, um, oh shoot,sorry, I clicked on the wrong.
(19:45):
The wrong one I was.
This was the AI writingassistant.
I wanted to show you theperformance tracker, so I'm
showing.
Here is a spreadsheet.
It basically has the url of thepost, the post title, the date
I posted it, the day of week,the time of day, the post
content and then the performance, and then you can look at like
wow, it turns out all of my topsix posts were on Friday,
saturday or Sunday, sosomething's going on about my
content that happens to beresonating on on weekend.
(20:08):
So that was.
That was a very kind of liketactical and detailed, detailed
view, zooming out our major usecase in every marketing function
that I'm aware of.
We have a ton of use case inlifecycle marketing, email
marketing, community marketing,partner marketing, social media
marketing, traditional productmarketing, and we use it for
content creation, we use it foranalysis, we use it for
synthesis, we use it forresearch, we use it for updates,
(20:31):
and basically what this hasmeant is I am the marketing team
for our company, me plus.
Ai is the marketing team, and Istill have a very active role
to play in setting up theseagents, training these agents,
guiding them, giving themfeedback, because, as you've
seen, like a big part of ourmarketing strategy is based on
my relationship with ourcustomers, and I do all the
(20:52):
YouTube videos myself, I answera bunch of the support messages
myself, and that's only possiblebecause I have AI doing all the
busy work sort of behind thescenes.
For me, it's incredible.
Kevin Kerner (21:00):
It's just absolutely incredible.
That was the one thing I wasthinking is, when you're showing
me that stuff number one itreplaces a lot of other tools.
I was thinking Tapleo, you knowwhich is, or any sort of social
media creator tool, schedulertool, whatever.
It replaces a lot of thattechnology because you're just
building an agent space to do it.
It's kind of like citizen-leddevelopment.
Jacob Bank (21:18):
You're doing it yourself versus going to it,
yeah, and this approach won't beright for everyone.
So relayapp is a veryhorizontal tool.
You can think of it as like aspreadsheet.
Like every single company inevery single role uses
spreadsheets right, but you usethem for vastly different use
cases at vastly different levelsof sophistication.
Some people prefer to use aspreadsheet for modeling their
finances, some people prefer touse a dedicated financial
(21:42):
management tool, and so and foreveryone.
For me, I love the flexibilityof working with horizontal tools
because it lets me customizethings just the way I want to
and it gives me much morecontrol.
But it does take more effortbecause you have to tinker with
things a little bit.
You have to experiment.
So me.
So what I would encourageeveryone listening to do is
explore both use case specifictools and horizontal tools, and
(22:05):
my guess is we're going to endup with some blend of both.
You're not going to use Relayas your CRM.
You couldn't even use Relay asyour CRM.
It's not a theorem.
It doesn't have the right datastorage.
Like you got to use a CRM asyour CRM you.
You can definitely use Relay toaugment some of the
capabilities that might bemissing in your CRM, like
logging certain kinds ofcalendar events or doing a
certain kind of enrichment.
Kevin Kerner (22:22):
Well, you also had some emails in there.
They're going out through Gmail, but it's just amazing to see
that that's mind blown on someof this stuff.
Jacob Bank (22:31):
Can I show you one other thing that's super cool.
This is something that weliterally just built today.
So we've had a lot of successwith this combination of
LinkedIn posts and then liveevents like webinars.
This has been a.
It's a motion that really workswell for us, because the
LinkedIn post gets a lot of topof funnel traction and the
webinar really teaches peoplehow people to the webinar Like I
(22:58):
looked at Zoom webinar, Ilooked at Luma, I looked at all
these dedicated tools and theyall just felt kind of heavy and
kind of wrong for me.
Like cause.
I wanted this feeling ofpersonal interaction.
I didn't feel like you weregiving your email to some
faceless, you know newsletterwriter, and so I eventually I
tried to do like a public Googlecalendar.
That didn't really work, and sowe decided what if we built our
own custom landing page forevents and we used a tool called
(23:22):
Lovable, which folks haveprobably heard of?
Lovable, bolt and VZero theseare probably the three most
popular natural language-basedweb app creators and I want to
show you this tool that we wereable to build in about two hours
in Lovable.
It's a totally custom websitethat says here's when the
Relayapp live sessions arecoming up.
It shows you the list ofupcoming sessions.
(23:44):
There's one tomorrow, there'sone March 20th, there's another
one on April 9th.
You can click to expand to seeother future sessions.
You can select a session andenter your email.
Hit, sign me up.
And now.
This is lovable on the frontend and on the back end, it's
using a relayapp workflow toautomatically add that email
(24:06):
address to the calendar event,automatically adding that email
address to our loops emailmarketing tool, automatically
sending a personalized email.
And then this is something thatour head of product, who is
working this out just today Ithink this is a great touch Like
you can watch previous sessionsand that links right to our
YouTube channel where all of ourprevious sessions are posted.
And this is crazy because noengineer was involved in this.
(24:29):
It's totally custom and exactlyfits the need and the use case
that we have.
It took two hours to set up.
All like no code tools.
Lovable on the front end andthen relayapp on the back end.
It actually works.
It actually works.
We are actually going to usethis in production and send that
link to thousands of people.
Kevin Kerner (24:46):
Amazing.
Yeah, I've been on the no-codething for a few years now.
We built our I think I messagedyou about us using Glide Like
we use Glide to the max man, it,man.
It is such an easy tool to useand it's all spreadsheet based,
so I'm really interested to seehow Relay can work with it,
because we can get data in andout of it.
Jacob Bank (25:02):
One funny thing Go ahead, oh go ahead, Go ahead.
Kevin Kerner (25:05):
Well, I was going to, I was going to, I was going
to, but Glide is a tool that's alittle bit more cautious about
access, I think.
But I was going to, and whatwe're trying to do on the Mighty
and True side is find peoplethat can build in these no-code
tools for our customers.
Like, the makeup of a marketeris different now.
It seems like it's this sort ofcreative growth, entrepreneur
(25:27):
role, but you have to be able touse or learn these tools, which
are pretty accessible.
You can easily learn them.
I'm curious what you think themakeup of a like.
If you were a marketer, likewhat would you?
Or if you were going to hire amarketer or replace you in this
stuff?
If you were a marketer, or ifyou were going to hire a
marketer or replace you in thisstuff, what would you want them
to do?
What would their makeup be?
What type of person is that?
Jacob Bank (25:45):
So first, I don't know, because it's been hard for
me to know what marketer tofind, which is why I've been
doing it myself.
But let me tell you some of thethings I would think about.
First, I would think about whatmarketing strategies are likely
to work better in the futureand which marketing strategies
are likely to work worse in thefuture.
So, if you're used to writingSEO optimized blog posts for
(26:08):
really competitive keywords oof,I'm not sure how future-proof
that's going to be Like AI isjust going to get really, really
.
It's going to be hard todifferentiate there.
Ai is going to get really goodat writing the content and the
companies that already have thedomain reputation and the topic
authority are going to winanyway.
If you're really good atwriting cold emails, like ah,
are cold emails going to work ina world where AI generated cold
(26:28):
emails are a dime, a dozen?
I don't know.
And so what marketingstrategies I think are likely to
work in the future?
I think marketing strategiesthat enable the customer base to
build a real relationship withthe team and the builders.
I think those are going to bedurable into the future, because
people want to buy a productthat's from people they trust
and that they know.
(26:48):
And that's why I actuallyexperimented with fully
automating my YouTube videos andI got them pretty good.
It could click around thescreen, it could do the
voiceover, it could generate thescript.
But I got some feedback thatwas like hey, we're missing you
in the video, because they seeme on YouTube, they see me on
LinkedIn, they see me doingcustomer support, and so things
(27:10):
that are based on building trustand a relationship with the
team.
So certain kinds of socialmedia, certain kinds of content
creation and then also kind ofcommunity led channels, because
in a world where there's AI youknow AI generated content
everywhere it's kind of likewhen I need to find a plumber
for my home, I can't really goto Yelp anymore, because I just
don't really trust Yelp anymore.
It's so flooded with reviews Idon't know if they're meaningful
(27:30):
.
So what do I do?
I ask my four neighbors I saydo any of you have a good
plumber?
That you've worked?
And I think that's the world.
Software is going to go the way.
Software is going to go too.
And so I'd want a marketer whoreally has great instincts in
building community, buildingword of mouth, building
advocates, cultivating advocates.
So that'd be one thing I wouldlook for.
Are they aligning themselves tothe kinds of marketing channels
(27:50):
that I think are likely to workdurably in the future?
That's number one.
Number two are they totallyunafraid of playing with new
tools and learning new things,like and in fact, their first
instinct should always be isthere a tool that could help me
with this?
And you know, there are peoplethat we've worked with that are
like that, and there are peoplethat we work that are not like
that.
They're like I know my playbook, I know my way of doing things,
(28:11):
I'm doing it.
So I really need people whosefirst instinct is let me try to
find the best tool that'savailable right now, and are
always updating their mentalroster of like here are the
tools that are great, here'swhat AI can do, here's what AI
can't do.
And then number three is I needsomeone who's a little bit more
(28:33):
of a generalist, because I thinkcompanies are going to be
smaller in the future.
Cursor 150 million ARR with 22people.
Bolt 40 million ARR with 22people.
Bolt 40 million ARR with 17people.
Lovable 20 million ARR with 15people.
I might be messing up the exactnumbers, but, like, spirit is
definitely right.
(28:53):
There are just going to besmall, small companies with
player coach generalists thatare harnessing many AI agents to
do a lot, and so those are kindof the three things I'm looking
for.
I don't really care if you'vepreviously written a blog post
that ranked on Google.
I don't really care if you'vepreviously run an amazing PPC
campaign.
I want to see that you canlearn the tools that are going
to enable you to do the rightforward thinking thing.
Kevin Kerner (29:12):
It's a certain amount of curiosity and
fearlessness and desire toautomate and sort of systems led
thinking that I think is goingto win out in the end.
This is awesome.
We could go on.
I want to ask you reallyquickly, so we have a few
minutes.
Here is your marketing story.
You mentioned some of it like alot of this has been
founder-led and those sort ofthings, but you have this
(29:33):
amazing video story.
What do you think's real?
How have you really done whatyou've done with the brand by
yourself?
Where have you seen success?
I know some of it's been in theYouTube and on the video side
of things.
I'd love to hear just quicklytalk about that success.
Jacob Bank (29:47):
Let me tell you a quick story, and it has been
failure after failure afterfailure.
It's been really, really hard.
So the book that really shapedmy thinking in the early days of
the company was the mom test.
I highly recommend it toeveryone.
If you're in the early userresearch phase or you're looking
for your first five or 10customers, that book basically
outlines a methodology by whichyou can talk to people and get
(30:08):
real feedback, Because nicepeople like Kevin if I say, hey,
Kevin, I've been pouring myheart and soul in this thing for
the last six months what do youthink?
What are you going to say to me?
You're going to say I love it socool.
Let me know when it launches.
So there's a methodology tokind of asking questions in a
way that will give you the truthabout whether you're onto
something useful.
And then I was able to get sortof our first five-ish, 10-ish
(30:29):
customers just kind of throughmy network and using that
interview-based methodology.
Then the book that reallyshaped my thinking for the next
phase was a book called Traction.
It was written by I think hisname is Gabriel Weinberg, the
founder of DuckDuckGo.
And that basically outlines amethodology that says there's 20
or 25 different channels, fromlive events to paid ads to
(30:50):
social media and what you shoulddo is, at every phase of growth
, you should experiment with thefew channels that you think are
likely to work at that phase ofgrowth, and then you're gonna
have to rerun this exercise manytimes as you reach different
phases of growth.
So at the first phase of growth, we were going from, like you
know, the zero to 10 customers.
The channels we picked wereintroductions through my network
, finding people who had thisproblem on Reddit and
(31:11):
interacting with them, and doinga little bit of like outbound
on LinkedIn, based on people wholiked adjacent products, and
those got us from one to 10users.
But I quickly realized, like,oh shoot, those are not actually
gonna get us for one to 10customers.
Going to individual Redditthreads is not gonna get us from
10 to 100.
Like that's not gonna work.
And so, from 10 to 100, wethought what could work for us?
(31:31):
Partner marketing, because weintegrate with a bunch of other
tools Airtable Notion,smartsuite, et cetera.
So if they list us in ourmarketplace, if we do a webinar
with them, if we do a jointtemplate with them we get some
traction.
We built some basic bottom offunnel SEO blog posts.
We tried a bunch of stuff thatdidn't work that I won't even
mention here.
We tried outbound, we tried afew other things that totally
(31:52):
didn't work.
And then those tactics got usfrom 10 to 100.
But we were like, oh shoot,those are not going to get us
from 100 to 1000.
We sort of tapped out what wesee as the opportunity and
bottom of funnel blog posts.
We've kind of seen what theceiling is with partner
marketing not going to get usfrom 10 to 100.
And so we rethought and thought, okay, what are the things that
are going to compound with ouruser base and get us from 100 to
(32:13):
1000?
And we decided it was going tobe a combination of social media
.
It was going to be acombination of social media,
video content and then kind ofcommunity and partnership
facilitation of word of mouth.
And I've been posting two tothree times a week on LinkedIn
and YouTube for the last yearand had no results.
(32:34):
No results for the first ninemonths.
I can show you the chart of ourYouTube channel.
So here's our YouTube channel.
I'll kind of describe it andI'm not saying we have some
amazing channel.
Now we're closing it on 5,000subscribers, which is decent at
our pace.
But if you go to the chart ofthe historical subscribers, I'll
(32:56):
show you the last year Ourchannel kind of just turned one
year old and I'll show you thelast year Our channel kind of
just turned, just turned oneyear old and I'll show you the
subscriber count.
Sorry for navigating through thewhole YouTube UI here.
You can see here.
If you look at the last year,it was just flat, completely
flat.
From March to basicallyDecember Instinctively got up
(33:19):
like six or 700 subscriberspainful, painful, painful.
And then all of a sudden wekind of found the thing that
clicked.
In February my videos got a lota bit better.
I was cross promoting themeffectively from LinkedIn.
We got a bit of word of mouth,the algorithm started picking us
up and now in the last monthwe've gone from 1000 to 5,000
subscribers, and LinkedIn is asimilar story.
(33:39):
Let me show you this.
Did your?
Kevin Kerner (33:41):
volume of videos.
Jacob Bank (33:44):
Nope, nope.
Volume didn't go up, it waspretty much two a week.
Kevin Kerner (33:48):
Great for you for sticking that out.
That is unbelievable.
Jacob Bank (33:51):
Some people would just go oh well, video's not
working.
Got Gotta let it go, that'sgreat to see and this is where
you know you really have totrust your marketing or
entrepreneur intuition.
Sometimes the right answerskeep going and sometimes the
right answer is switch, like Iknew.
Outbound was not working and wehad to switch.
So look at this on LinkedIn Iwas posting three times a week
(34:11):
this whole time Nothing, nothing, nothing, nothing, nothing,
nothing, nothing, nothing,nothing, nothing, nothing,
nothing.
I had one post that hitDecember 31st, 23rd, the day
before Christmas, and then Ikind of figured it out and now
I've got 5 million impressionslike 4.7 million of them were in
the past 90 days, and I haveyou know 800,000, 750, 600.
(34:33):
And, and it goes, it goes down.
Kevin Kerner (34:35):
Amazing.
I mean, that's.
That's one of the, that's oneof the most clearly product led
growth examples I've ever seen.
It's really fantastic.
Jacob Bank (34:42):
And so I don't know what the lesson from that is
exactly.
I think it's a combination offind a channel that you believe
in, that you think is going towork for you, and then just be
ridiculously persistent and then, once it pays off, double down
on it.
So now LinkedIn is working fortop of funnel, youtube is
working for mid funnel, and sonow I can just focus on those
two channels.
Kevin Kerner (35:02):
That's awesome.
What a great, a great story.
Okay, there's so many thingsI'd want to ask you beyond that.
We're running up against time,so I wanted to do one more thing
.
I haven't tried this before ona podcast.
This is the first time I'vedone it on this, on this podcast
series.
So I put a question, intoperplexity you would probably.
I'm not going to tell you whatthe prompt was, because you're
such a prompt expert, but it wassomething around.
Hey, I'm meeting with Jacob,his founder.
(35:23):
You know we're talking about abunch of different things here.
Give me one question I shouldask him that is provocative, and
so I'm going to hit send onthis and see what it gives back
to me.
Hold to me, hold on a second.
I should probably just grab myscreen, but I'll just give it to
you.
Oh gosh, if you could go backto your time at Google and undo
one decision you made as aproduct leader, what would it be
(35:44):
and how do you think it wouldhave shaped your entrepreneurial
journey differently?
That's completely unedited.
Jacob Bank (35:47):
It's actually a great question and I have an
exact answer for it.
It wasn't a product decision,but it was a way I operated that
really handicapped me for thefuture.
I was the product lead of Gmailfor six years and I ended that
time with 800 Twitter followersInteresting.
That's what was I doing.
What was I thinking?
(36:07):
A?
I was way too insulated fromour real users when I was in
that position and I was lookingat all the data and I was behind
the scenes and I was, you know,interviewing small groups of
users and making decent productdecisions.
But if I look at the bestproduct leaders, they are the
public face of their product.
They are on Twitter, they're onLinkedIn, they're making demo
videos, they're doing interviewsand they're directly
(36:30):
interacting with users and, forwhatever reason, google culture
doesn't really value that orincentivize it or encourage it,
because and I can see why likeGoogle doesn't want a bunch of
loose cannon PMs talking aboutstuff on Twitter because you can
end up with regulatory issuesand PR disasters and stuff.
But when I look back at it, Iwas like what was I thinking?
(36:51):
I was in, I had this platformwhere we had 2 billion users and
I could have interacted withthem more, both by way of
getting their feedback andmaking the Gmail product better,
and then I would have had thatskill set and that personal
brand that I could have broughtthrough to my startup, because I
now am up to whatever 20,29,000 followers on LinkedIn.
I've still got nothing onTwitter, by the way, but I had
(37:14):
to learn all of this reallypainstakingly from scratch, and
the only thing that actuallyspurred me to do it is my
company's future depended on it.
So I was like, okay, I betterlearn Now.
It's not optional, it'srequired.
And so what I would recommendto everyone in any role, whether
you're at a big company or asmall company or agency or
consultant.
In the way the world is moving,where people want to build
(37:35):
authentic relationships with thebuilders or creators of the
products that they're using,like, you got to get out there
and you got to just put yourselfout there.
Make tweet, make LinkedIn posts, make YouTube videos, host
webinars.
Don't be afraid of lookingsilly.
That was the biggest mistakefor my time at Google.
Kevin Kerner (37:50):
Amazing.
That's an amazing insight.
Great question by perplexity.
I take no credit, but man,that's amazing.
Yeah, and there's a certainamount of courage.
You need to put yourself outthere.
I mean, I've been doing thisfor a long time, so it's like I
think the same thing.
It's like, golly, I was doingall this stuff.
What was I doing the whole time?
Because it was behind thescenes working.
It's crazy.
Yeah to you, and I'm such a fanof.
(38:12):
I think part of what makes yourbrand work is you're so
enthusiastic and so willing tolike share with the community.
And it's real.
You know you really want to dothis stuff.
So it's just, I feel, veryfortunate.
Jacob Bank (38:22):
I've always been a productivity nerd.
I absolutely love this stuff,and I think the best service I
can provide is to show the real.
Show the reality, becausethere's so many demos out there
that are very curated, and so Ithink you've you've seen a
couple of my my live sessions.
Like I make mistakes all thetime, I mess up my prompt, I
(38:43):
don't, I don't attach the rightinformation, but that's the
reality and so, and so I justwant to show people yeah, I make
mistakes too, and here's how Iwork through it when I do,
because it's a when, not an if.
Kevin Kerner (38:53):
Yeah, it's fantastic.
Well, there's a bunch of waysto get ahold of you.
What's the best way to followwhat you're doing in this space?
Jacob Bank (39:02):
LinkedIn is the best beginner way.
I post pretty broad contentabout how to use AI to help you
get stuff done.
None of it is too Relayappspecific, really, it's just
general content on ways you canuse AI.
That's the easiest.
And then, if you're ready to goone step deeper, I post very
detailed YouTube videos 45minutes or an hour about how to
(39:22):
set up an AI agent for yourself.
So those are the two best waysto follow along.
But please feel free to DM me.
You know ping me directly, dmme, email me at jacob at
RelatedApp Like.
I would love to hear your usecases, your feedback and any
ideas of how we can make AI workeffectively for you.
Kevin Kerner (39:37):
That's fantastic, and I've been through the
YouTube videos, the builds, withyou.
They're very accessible.
So if you're a marketer whowants to get into the stuff and
you're thinking I can't do this,it is really easy.
It's not hard to do.
You just need to give it a tryand you do a great job at
walking people through it errorsand all all the way, so it's
fantastic.
Yeah, Jacob, thank you so much.
I am really happy to havetalked to you.
(39:57):
We'll have to do a recap a yearfrom now to see how this all
turns out.
Jacob Bank (40:00):
I think the fun thing is yeah, a year from now,
I'm sure two thirds of what I'vesaid will look painfully out of
date, but that's what's funno-transcript and I'll be a
different person, the companywill be different and AI will be
(40:20):
in a different state.
Kevin Kerner (40:21):
For sure.
Okay, great seeing you, Jacob.
Jacob Bank (40:23):
Yeah, thanks.
Kevin Kerner (40:23):
Kevin.
Popular Podcasts
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.