Transcript
WEBVTT
1
00:00:02.540 --> 00:00:02.740
Yeah.
2
00:00:05.140 --> 00:00:09.110
Hey everybody Logan with sweet fish
year first of all, whether you are a
3
00:00:09.120 --> 00:00:12.860
regular subscriber of GDP growth or new
to the show. Thank you so much for
4
00:00:12.860 --> 00:00:17.930
hitting play on this episode a little
bit about what to expect over the next
5
00:00:17.930 --> 00:00:21.050
several weeks we're going to be
replaying some of our best episodes
6
00:00:21.050 --> 00:00:25.760
from a deep dive we did earlier this
year, all about a B. M. Or account
7
00:00:25.760 --> 00:00:29.730
based marketing. If you are new to be
to be or you're trying to get your A. B.
8
00:00:29.730 --> 00:00:34.240
M. Efforts off the ground or really
gain some traction in what you've
9
00:00:34.240 --> 00:00:38.560
started to do with a B. M. Within your
organization. You are going to love
10
00:00:38.570 --> 00:00:42.140
these episodes. So over the next
several weeks expect to see two
11
00:00:42.140 --> 00:00:46.920
episodes a week here in your BdB growth
feed and they will all be focused
12
00:00:46.930 --> 00:00:51.530
around A B. M. From our account based
marketing. Deep dive. All right, let's
13
00:00:51.530 --> 00:00:58.200
get into the episode. Today I'm joined
by a guy yell if who is the ceo of
14
00:00:58.200 --> 00:01:01.820
intel um eyes and we're gonna be
talking about website personalization.
15
00:01:01.820 --> 00:01:04.739
This is kind of a category that I found
out is in A. B. M. Though. I haven't
16
00:01:04.739 --> 00:01:09.360
covered in any of the episodes yet. So
I was excited to talk to guy about his
17
00:01:09.360 --> 00:01:12.840
company. What it does as well as just
the category of what we can do with
18
00:01:12.840 --> 00:01:16.830
personalizing a website which I think
is any any digital marketing plan. Any
19
00:01:16.830 --> 00:01:21.610
kind of marketing plan in the 2020 is
right is like the main hub of all
20
00:01:21.620 --> 00:01:24.610
marketing that we do. We're all trying
to get traffic back to our website
21
00:01:24.610 --> 00:01:27.930
where we have more control over the
experience. So it makes sense that we
22
00:01:27.930 --> 00:01:33.930
want to talk about a B. M as it relates
to your own website. So guy, I'm so
23
00:01:33.930 --> 00:01:37.210
excited to have you on the show today.
How are you doing? I'm doing great dan,
24
00:01:37.210 --> 00:01:40.450
honored to be here. I've been listening
to the podcast throughout the month,
25
00:01:40.450 --> 00:01:46.230
very excited to have this conversation.
Fantastic dive in. So to get started, I
26
00:01:46.230 --> 00:01:50.710
want to talk about broadly, what is
website personalization in case the
27
00:01:50.710 --> 00:01:55.540
audience doesn't know and why is it
important? Website personalization is
28
00:01:55.540 --> 00:02:00.170
giving different visitors, different
experiences as consumers. We've all
29
00:02:00.170 --> 00:02:05.550
been trained by yahoo amazon netflix
and a bunch of others to expect Our
30
00:02:05.560 --> 00:02:08.410
digital experience, our journey with
the companies with whom we're
31
00:02:08.410 --> 00:02:12.910
interacting to be tailored to us
website personalization is an approach
32
00:02:12.910 --> 00:02:17.460
where you can meet in a B2B context,
each of your prospects where they are
33
00:02:17.460 --> 00:02:21.730
in their funnel with you to recognize
that a first time visitor is really
34
00:02:21.730 --> 00:02:24.660
different than somebody has already
asked for a demo is really different
35
00:02:24.660 --> 00:02:29.970
than somebody who's already an existing
customer. The goal of it is more
36
00:02:29.970 --> 00:02:33.730
revenue more leads to sales or more
customers. Ultimately, it's about
37
00:02:33.730 --> 00:02:38.410
driving more money and it's one of many
techniques to go do that like the
38
00:02:38.410 --> 00:02:41.380
broader umbrella of a B. M or the
broader umbrella of conversion rate
39
00:02:41.380 --> 00:02:46.760
optimization. I've seen website
personalization out there for while I'd
40
00:02:46.760 --> 00:02:51.000
probably say the first experience I had
with it was probably five years ago and
41
00:02:51.000 --> 00:02:53.890
I had seen it. And what had happened is
there like, well which one, which
42
00:02:53.890 --> 00:02:57.280
bucket do you fit in? We wanted taylor
and it was literally like we want to
43
00:02:57.280 --> 00:02:59.880
tailor the website to you. And I was
like, okay, I'm like, I don't really
44
00:02:59.880 --> 00:03:02.710
fit in any of these buckets but I'm
curious to see how this works, click
45
00:03:02.720 --> 00:03:06.490
and then of course the homepage changed
and customize the website. I was the
46
00:03:06.490 --> 00:03:08.770
first time I realized that it was
happening though. I'm sure it was
47
00:03:08.770 --> 00:03:12.160
happening long before that and
generally you probably don't know about
48
00:03:12.160 --> 00:03:16.110
a lot of websites are probably being
customized. Do you right now as you're
49
00:03:16.110 --> 00:03:19.610
browsing around the web or when you get
back on the web sometime today. The web,
50
00:03:19.620 --> 00:03:23.470
some websites probably been customized
guy. What percentage of websites do you
51
00:03:23.470 --> 00:03:27.040
feel like you're just being customized
all the time now and people just aren't
52
00:03:27.040 --> 00:03:31.630
even aware of it. It's a great question.
The last time I looked at data from
53
00:03:31.640 --> 00:03:35.890
either built with their data. Nice to
the sources. We tend to look at The top
54
00:03:35.890 --> 00:03:42.110
10,000 websites. It was something like
a 3rd 30 to 40% top 100,000 was about
55
00:03:42.110 --> 00:03:47.390
half that it seems like look we as
marketers have been a marketer for the
56
00:03:47.390 --> 00:03:52.910
last decade, spent 100 and $20 billion
every year just in the us driving
57
00:03:52.910 --> 00:03:57.430
people to websites. But we failed to
turn them into customers 98% of the
58
00:03:57.430 --> 00:04:03.060
time. Like in what world is that? Okay?
And so we see this as something that we
59
00:04:03.060 --> 00:04:08.000
all increasingly expect as consumers
and in our business lives it has a
60
00:04:08.000 --> 00:04:12.430
great positive impact. I'm going to say
for a lot of websites, 2% would even be
61
00:04:12.430 --> 00:04:16.839
pretty dang good. Right? I've heard
anywhere from like if you're, if you're
62
00:04:16.839 --> 00:04:20.320
in between the 2 to 5% conversion rate
on your website depending on what
63
00:04:20.329 --> 00:04:24.460
industry of course that you're in, like
you're usually doing pretty good. Like
64
00:04:24.460 --> 00:04:28.180
even talked to ask chris walker and
linkedin recently. So he's like, we got
65
00:04:28.180 --> 00:04:32.300
to get beyond the 0.5% conversion rates.
I'm like, so what is good? He's like,
66
00:04:32.310 --> 00:04:36.030
oh, you know between two and 5% I'm
like okay, it's not like 10 or 15%
67
00:04:36.030 --> 00:04:41.390
though, it would be highly uncommon to
see that. And I think if you're not
68
00:04:41.390 --> 00:04:45.360
sure what website personalization is, I
love if you go to intel um, eyes dot
69
00:04:45.360 --> 00:04:49.700
com, they have this one button that I
found was wonderful because it
70
00:04:49.700 --> 00:04:53.760
illustrates what they do and what it is.
It's like a yellow button that kind of
71
00:04:53.760 --> 00:04:58.340
hovers over the website. If you click
it, it says, see what is optimized here,
72
00:04:58.340 --> 00:05:01.240
You click the button and all of a
sudden it throws like these blue dotted
73
00:05:01.240 --> 00:05:05.670
boxes around every section of their
website that has been customized for me,
74
00:05:05.670 --> 00:05:09.780
the website visitor. So it's very
exciting to see all the things that you
75
00:05:09.780 --> 00:05:12.490
can customize of course as a marketer.
I'm like, well it's probably a lot of
76
00:05:12.490 --> 00:05:15.260
freaking work because now I have to
figure out what would I serve to
77
00:05:15.260 --> 00:05:18.630
everybody and how would I think about
it differently. But that will have to
78
00:05:18.630 --> 00:05:22.060
wait for that part of the conversation.
A little later in the interview for now,
79
00:05:22.060 --> 00:05:26.050
I want to talk about like, what, how is
this related to A B. M. Like how is
80
00:05:26.050 --> 00:05:29.960
this serve somebody who's walking
through an A. B. M. Campaign to tailor
81
00:05:29.960 --> 00:05:35.050
their website to their audience. We
think they are directly related. So A B.
82
00:05:35.050 --> 00:05:38.910
M. Is saying, as you've covered
previously on the podcast, sales and
83
00:05:38.910 --> 00:05:41.610
marketing are not going to be
coordinated. So the marketing is
84
00:05:41.610 --> 00:05:46.270
focused. Their spears that they're
sending outbound on the exact accounts
85
00:05:46.270 --> 00:05:50.080
and individuals of those accounts that
matter to sales. Great. Let's say you
86
00:05:50.080 --> 00:05:53.960
hook one of them and managed to get
them to your website. You want to speak
87
00:05:53.960 --> 00:05:58.680
to them differently than you would
speak to others. We encourage all the
88
00:05:58.680 --> 00:06:03.060
folks with whom we work to treat every
page in particular. Their homepage like
89
00:06:03.060 --> 00:06:06.100
a landing page or you think about all
the landing pages we've all created
90
00:06:06.100 --> 00:06:10.400
over the years for our ads. What if
your homepage where that personalized.
91
00:06:10.400 --> 00:06:15.210
What if every page was 1-2 then go meet
them where they are in their journey
92
00:06:15.210 --> 00:06:19.630
with you to recognize that you're
speaking to different roles at
93
00:06:19.630 --> 00:06:23.580
different sized companies that are in
different places in their journey and
94
00:06:23.590 --> 00:06:27.370
that you can create a more bespoke more
relevant experience that will move them
95
00:06:27.370 --> 00:06:31.640
through the funnel more quickly. Some
website personalization helps bring a
96
00:06:31.640 --> 00:06:35.560
whole bunch of that to life. Often to
your earlier point about content. We're
97
00:06:35.560 --> 00:06:38.610
using content. You already have you've
already created the personas, You've
98
00:06:38.610 --> 00:06:41.980
already created the case studies. If
you can show that visitor, the one
99
00:06:41.980 --> 00:06:46.560
that's relevant to them, they're more
likely to engage with you more makes a
100
00:06:46.560 --> 00:06:51.430
lot of sense. But my question is like
let's say I've been interacting with
101
00:06:51.430 --> 00:06:55.930
people on linkedin. I share a link to
my maybe a blog post, a case study
102
00:06:55.940 --> 00:07:00.730
published recently and they're visiting
to it from linkedin. How does it know
103
00:07:00.730 --> 00:07:03.040
that they're in my target account?
Because there's probably a lot of
104
00:07:03.040 --> 00:07:06.770
traffic coming from linkedin but some
of them are outside my I. C. P. Some of
105
00:07:06.770 --> 00:07:11.310
them are my target accounts but could
be you know and some are specifically
106
00:07:11.320 --> 00:07:15.360
the exact people I want to build a
relationship with. How does it know?
107
00:07:15.840 --> 00:07:24.290
Great question. We've typically seen
three ways this will happen one Often
108
00:07:24.300 --> 00:07:28.990
you've already put you tm parameters or
other Euro parameters in the link for
109
00:07:28.990 --> 00:07:33.920
your own analytics. You can use those
to better personalize the experience,
110
00:07:33.920 --> 00:07:38.900
We automatically parse all of them. Uh,
and use that in machine learning to
111
00:07:38.900 --> 00:07:42.670
find correlations or you can set up
rules, right? If they're in the target
112
00:07:42.670 --> 00:07:47.540
account list, I want to show them this
different experience. 12, you can deny,
113
00:07:47.540 --> 00:07:52.220
minimize the traffic. Maybe maybe you
didn't tag the link or maybe they came
114
00:07:52.220 --> 00:07:56.170
from somewhere else. You might use one
of the demonization vendors out there
115
00:07:56.170 --> 00:08:01.660
that say, oh you know what dan showed
up from the Sweet fish Media office and
116
00:08:01.660 --> 00:08:05.350
we know who he is. We can personalize
that way. And three if they visited the
117
00:08:05.350 --> 00:08:09.410
site before and maybe they felt that a
form You can go match the market a
118
00:08:09.410 --> 00:08:15.160
munchkin or some other way to reconnect
back to that profile to know who it is.
119
00:08:15.170 --> 00:08:19.810
So you can personalize by industry by
role, by size of company by stage and
120
00:08:19.810 --> 00:08:23.520
funnel or whatever other lens you would
like to use on that personalization
121
00:08:23.530 --> 00:08:27.660
does that resonate them? So I think so
I think so is three different ways. one
122
00:08:27.740 --> 00:08:32.400
I give something like an televise the
information through the link. You earl,
123
00:08:32.440 --> 00:08:35.760
the Perma link that comes with the U.
Tm programs and all that kind of stuff.
124
00:08:36.240 --> 00:08:40.700
Um, so I can essentially say, hey, I'm
going to send people to my home page,
125
00:08:40.700 --> 00:08:44.200
but I know in this particular setting
where I'm sharing the link that would
126
00:08:44.200 --> 00:08:47.950
be in this particular industry. So
while I'm sending to my homepage, it's
127
00:08:47.950 --> 00:08:50.130
going to customize it for the industry
because I have a little bit more
128
00:08:50.130 --> 00:08:52.500
control over that. Maybe I know where
I'm sharing the link. Maybe I'm
129
00:08:52.500 --> 00:08:57.060
shooting the link than email out to a
group and I haven't targeted already.
130
00:08:57.440 --> 00:09:01.170
So there's that I think the second one
I heard was just through cookies
131
00:09:01.180 --> 00:09:06.610
because they visited maybe they visited
particular blog post in the past and
132
00:09:06.610 --> 00:09:10.750
because of that I can tag them now in
the system to know like okay this
133
00:09:10.750 --> 00:09:14.790
person's in this camp. So next time
they come back to the home page maybe
134
00:09:14.790 --> 00:09:19.390
through a retargeting ad I can
customize it knowing what I know about
135
00:09:19.390 --> 00:09:22.220
them because they visited this one blog
post. Maybe they found it through
136
00:09:22.220 --> 00:09:26.280
search or something and then the last
one is through the Crm because they
137
00:09:26.280 --> 00:09:29.890
filled out a form at some point we can
customize it based on what we have in
138
00:09:29.890 --> 00:09:32.850
the Crm about them which could start to
customize it based on where they're at
139
00:09:32.850 --> 00:09:38.350
in their journey. Is that right? Almost
exactly. I'm totally with you on the
140
00:09:38.350 --> 00:09:42.620
first one and in fact you can use those
U. T. M. Parameters, chances are you've
141
00:09:42.620 --> 00:09:45.620
done it for yourself. It's not for
untelevised. You've already done it for
142
00:09:45.620 --> 00:09:49.330
your own analytics will reuse them and
to your point and you can do it by
143
00:09:49.330 --> 00:09:52.450
industry, you can do it down to the
account, you can say it is JetBlue
144
00:09:52.450 --> 00:09:55.830
showing up and so I want to personalize
the specific, I want to show JetBlue
145
00:09:55.830 --> 00:09:59.820
their account rep on the page I want to
say hey joe or jane is ready to talk to
146
00:09:59.820 --> 00:10:05.110
you. The third one totally with you.
The crm system. Absolutely. The second
147
00:10:05.110 --> 00:10:10.150
one. That's a great idea. We've seen
that done more in B two C. Where you'll
148
00:10:10.150 --> 00:10:16.180
pay a blue kind experience. They've got
huge cookie pools out there. They're
149
00:10:16.180 --> 00:10:20.580
looking across sites to go identify,
you know, demographics or behavior
150
00:10:20.580 --> 00:10:23.670
about somebody. The thing I was
mentioning that I didn't do a good job
151
00:10:23.670 --> 00:10:29.580
describing was to pay someone like some
of the folks we think about a clear bit
152
00:10:29.590 --> 00:10:33.980
or six cents or demand base where
they've got a big database of this I. P.
153
00:10:33.980 --> 00:10:37.660
Address is this company's office. So
it's not about the cookies, it's
154
00:10:37.660 --> 00:10:40.910
actually about the I. P. Address. And
so you know, they showed up from
155
00:10:40.910 --> 00:10:43.850
JetBlue's office in wherever their
headquarters is. Let's pretend as new
156
00:10:43.850 --> 00:10:47.890
york that's their office. I know that
somebody from JetBlue and I can
157
00:10:47.890 --> 00:10:51.740
personalize the site accordingly. So
it's like a reverse. I could look up
158
00:10:51.750 --> 00:10:54.460
actually it's exactly that does that
resonate,
159
00:10:56.040 --> 00:10:59.430
yep. Absolutely. And that makes sense.
And a lot of a. B. M. Tech companies
160
00:10:59.430 --> 00:11:03.620
are certainly using the I. P. Address
quite often. I imagine that's gotten
161
00:11:03.620 --> 00:11:07.690
hard in a remote work environment where
people are working from home a lot that
162
00:11:07.700 --> 00:11:12.800
that tactic quite working as well as it
used to. But I sure as we go back, but
163
00:11:12.800 --> 00:11:16.300
as many people go back to work that'll,
that'll probably work a lot better. The
164
00:11:16.300 --> 00:11:19.140
cookie ones interesting though is I
found that that was probably at least
165
00:11:19.140 --> 00:11:23.370
for me, what I've done with websites
before is based on the cookies. I think
166
00:11:23.370 --> 00:11:26.040
one of my favorite plug ins that I've
used just for Wordpress and this is
167
00:11:26.040 --> 00:11:31.850
like a kind of like $100 or $3200 kind
of month thing called opted monster can
168
00:11:31.850 --> 00:11:38.040
actually, you can customize the Cts,
the pop ups or the C. T. A. Forms which
169
00:11:38.040 --> 00:11:42.570
ones are showing based on previous
pages, hit or loaded to it. I'm
170
00:11:42.570 --> 00:11:45.290
assuming they're doing it with cookies.
It's probably the only logical
171
00:11:45.290 --> 00:11:50.250
explanation I can get to because they
wouldn't have access to my crm. But
172
00:11:50.740 --> 00:11:54.210
that's probably the only reason why I
would think that's possible. As I've
173
00:11:54.210 --> 00:11:57.880
seen a few other people like that do it
interesting and it does seem like there
174
00:11:57.880 --> 00:12:01.490
are folks who make a whole business out
of it. You know, looking across sites,
175
00:12:01.490 --> 00:12:06.510
we purposely don't. I spent the last 15
years and ad tech and we aren't doing
176
00:12:06.510 --> 00:12:12.070
that cross site, but the matching in
any way. But within a site dandy, your
177
00:12:12.070 --> 00:12:16.660
point, you can infer an awful lot based
on somebody's behavior friday if they
178
00:12:16.670 --> 00:12:21.120
go to a page with a particular case
study, you can infer size of company
179
00:12:21.120 --> 00:12:24.170
and industry, if they go to a
particular product, you can infer what
180
00:12:24.170 --> 00:12:27.920
product they're interested in. Often
you'll have personas on your site. You
181
00:12:27.920 --> 00:12:33.900
can then, you know, infer from their
behavior which persona they are and
182
00:12:33.900 --> 00:12:37.560
remember that and speak to them
differently across the site using
183
00:12:37.560 --> 00:12:41.330
content you already have. Or if they
went to that case study, show them
184
00:12:41.330 --> 00:12:43.950
another one from the same industry on
their home page the next time they show
185
00:12:43.950 --> 00:12:47.960
up, you don't have to create a whole
new one. You can tailor the experience
186
00:12:48.540 --> 00:12:53.690
to create a more personalized, account
based approach. So you're using cookies
187
00:12:53.690 --> 00:12:59.210
to some degree and changing the site
live absolutely within the site. When I
188
00:12:59.210 --> 00:13:03.760
heard cookie, I was thinking back to ad
tech days Cross. No, no, no. It's like,
189
00:13:03.770 --> 00:13:08.140
I've heard some people call on site
retargeting. It's like ads, but they're
190
00:13:08.140 --> 00:13:12.290
all on your site. Like you have full
control over it, right? And maybe some
191
00:13:12.290 --> 00:13:16.060
people caught, gosh, I've run into a
few small startups that do this Now.
192
00:13:16.070 --> 00:13:20.700
They've tried to position it is on like,
on site retargeting, but really it's
193
00:13:20.700 --> 00:13:24.890
just changing things off based on what
pages people visited before. Okay,
194
00:13:24.900 --> 00:13:27.910
that's kind of cool that it's like the
site suggesting to you, even if it
195
00:13:27.910 --> 00:13:30.740
knows nothing about you. As soon as you
start clicking on links that start
196
00:13:30.740 --> 00:13:34.280
suggesting the experience for you and
that's a different depth that I have
197
00:13:34.280 --> 00:13:40.620
not heard or thought about before, you
spot on and if it does nothing about
198
00:13:40.620 --> 00:13:45.690
you, it can still do a better than I
know nothing about your job of deciding
199
00:13:45.690 --> 00:13:49.370
what to show you why. Because even if
you show up and we know nothing about
200
00:13:49.370 --> 00:13:49.750
you
201
00:13:50.840 --> 00:13:55.450
through reverse, I look up to location,
we know city state country, we know
202
00:13:55.450 --> 00:14:00.050
mobile desktop tablet. We know is it
your first visit a repeat visit. We
203
00:14:00.050 --> 00:14:03.470
know if you came from another site when
site where you want just through the
204
00:14:03.470 --> 00:14:07.310
refer if you were clicked on an ad or
an email to your point about the U. T.
205
00:14:07.310 --> 00:14:11.030
M. S earlier, we know what messaging
you saw, we know what targeting was
206
00:14:11.030 --> 00:14:14.960
used. You can use all of those. Even if
we've never seen anybody who looks
207
00:14:14.960 --> 00:14:19.930
exactly like you, we can use machine
learning to infer. Well I've seen
208
00:14:19.930 --> 00:14:23.950
people who share some of the attributes
with dan. Let me go show dan something
209
00:14:23.950 --> 00:14:27.770
that's more likely to get him to fill
out the form, we want them to fill out
210
00:14:27.770 --> 00:14:32.260
to talk to sales or to do the uh you
know uh free product trial signing,
211
00:14:33.140 --> 00:14:37.520
it's that resident. Absolutely. So I'm
thinking through, it's almost like
212
00:14:37.530 --> 00:14:40.710
putting something like this on your
website makes sense in any case. But it
213
00:14:40.710 --> 00:14:44.370
certainly makes sense for your A. B. M.
Campaigns mainly because of the the I.
214
00:14:44.370 --> 00:14:48.480
P. Look up or because you can tap into
other data sources that can then change
215
00:14:48.480 --> 00:14:52.690
the site for what you know is your
target account. But even if you don't
216
00:14:52.690 --> 00:14:56.760
know it's your target account it can
still start affecting and changing the
217
00:14:56.760 --> 00:15:00.860
conversion rate on your site quite
drastically by customizing the whole
218
00:15:00.860 --> 00:15:06.460
experience. I couldn't agree with you
more and often will find you know 98%
219
00:15:06.460 --> 00:15:10.670
of the traffic to the site is not the
target account list. And so to be able
220
00:15:10.670 --> 00:15:14.340
to have that spectrum from you know
what I'm doing 1 to 1 marketing. I know
221
00:15:14.340 --> 00:15:20.570
this is joe or jane at Acne Co and I'm
going to go name Agnico on the site.
222
00:15:20.580 --> 00:15:24.910
I'm going to go show them their
competitors logos in the logo farm, I'm
223
00:15:24.910 --> 00:15:29.420
going to show them an industry specific
case study all the way through to. I
224
00:15:29.420 --> 00:15:34.890
have no idea who this is and being able
to do that 1 to 11 too few. One too
225
00:15:34.890 --> 00:15:39.460
many all in the same experience on the
same website we find that helps
226
00:15:39.840 --> 00:15:43.490
maximize revenue for B two B marketers.
Tell me a little bit more about the 1
227
00:15:43.490 --> 00:15:46.430
to 1 experience because most of what
I'm thinking about when it comes to
228
00:15:46.430 --> 00:15:50.250
work website personalization is like a
one too few type of approach. How does
229
00:15:50.250 --> 00:15:57.030
it become 1 to 1? So if you knew for
example in that email or add somebody
230
00:15:57.030 --> 00:16:02.290
clicked on that that was targeted
specifically to dan Sanchez and then I
231
00:16:02.290 --> 00:16:08.290
wanted to name your company. I wanted
to put your rep on the site. I wanted
232
00:16:08.290 --> 00:16:12.620
to push you away. I'll make it up.
You're a large enterprise. I don't want
233
00:16:12.620 --> 00:16:16.240
you to go to self sir. I want to talk
to a human that works for my company.
234
00:16:16.250 --> 00:16:21.790
Right? I can go change each of those
experiences now. I can. Most of us I
235
00:16:21.790 --> 00:16:24.830
think are used to thinking, okay, well
that's a B M. I'm going to do that with
236
00:16:24.830 --> 00:16:29.090
rules. I'm going to do if this then
that statements we agree completely and
237
00:16:29.090 --> 00:16:33.960
think you can do even better. For
example, what's the right way to say?
238
00:16:34.440 --> 00:16:40.640
Acne Co in the headline of the page. Hi
Agnico. Welcome back Acne Co. We think
239
00:16:40.640 --> 00:16:43.280
you're great company. Agnico, here's
your competitor. Agnico, right? I'm
240
00:16:43.280 --> 00:16:47.360
making those up. You can use machine
learning to say this is a B. M.
241
00:16:48.540 --> 00:16:51.410
But I don't know exactly the right way
to do it. Right. And you actually get
242
00:16:51.410 --> 00:16:54.640
to not have this sort of religious
debates internally, like what's the
243
00:16:54.640 --> 00:17:00.560
right way to say Agnico in the headline?
You can use Ml paired with the rule,
244
00:17:00.560 --> 00:17:04.099
right? If they're on my target list, I
want to name them in the headline, but
245
00:17:04.099 --> 00:17:08.079
then which version of the headline? Use
I'm able to do that. Using machine
246
00:17:08.079 --> 00:17:12.060
learning to do a better job of figuring
out what version to show each visitor.
247
00:17:12.069 --> 00:17:17.410
You can apply that principle throughout
the whole experience for the elements
248
00:17:17.750 --> 00:17:21.800
that are a BME and for the elements
that aren't like, what do I want to
249
00:17:21.800 --> 00:17:26.770
call the action to be or what, what
sequence of things do I want on the
250
00:17:26.770 --> 00:17:29.370
page, You know, do I want to talk about
the product first or about the
251
00:17:29.370 --> 00:17:33.900
customers first or about the industry
first. You can then use ml to
252
00:17:33.910 --> 00:17:39.190
personalize those other elements in
parallel to the abbey, any sort of 1-1
253
00:17:39.200 --> 00:17:43.340
elements. Does that resonate? Gosh,
it's like you put two things together
254
00:17:43.340 --> 00:17:47.120
that I normally don't put together in
personalization and almost like this
255
00:17:47.120 --> 00:17:50.810
multi variant testing that the machine
learning is doing. And I always kind of
256
00:17:50.810 --> 00:17:53.460
figured that those was kind of like you
either do one or the other because
257
00:17:53.460 --> 00:17:56.470
they're really difficult to do together
unless I guess you have machine
258
00:17:56.470 --> 00:17:59.990
learning driving the whole experience
and can do more than just plug in
259
00:17:59.990 --> 00:18:05.140
insert name and can actually fiddle
with the actual copy around the
260
00:18:05.150 --> 00:18:11.380
personalization. So it opens up a whole
world and must be interesting to see
261
00:18:11.380 --> 00:18:14.390
how many different variations of a web
page you can load if you have that kind
262
00:18:14.390 --> 00:18:18.240
of thing going on almost like your
website. So leaving living breathing
263
00:18:18.250 --> 00:18:22.030
thing. Um, still you have to put
parameters on it, right? You have to
264
00:18:22.040 --> 00:18:25.350
give it like a character, limit what it
can do for a headline before it starts
265
00:18:25.350 --> 00:18:29.970
to throw off the site design, right?
And what you said resonates so strongly
266
00:18:29.970 --> 00:18:34.590
and we so often hear these are two
strategies to budgets, two teams, and
267
00:18:34.590 --> 00:18:37.840
ultimately they're both in service of
making more money. And so I'm totally
268
00:18:37.840 --> 00:18:43.620
with you on sort of, the mindset shift
and on the variations you can load, I'm
269
00:18:43.620 --> 00:18:47.850
with you. In fact, our average customer
over the last two years tried 177
270
00:18:47.850 --> 00:18:52.330
different ideas, had they done that
with traditional split testing, that
271
00:18:52.330 --> 00:18:53.770
would have taken a quarter century,
272
00:18:55.240 --> 00:19:01.130
but using Ml they didn't do 177 they
optimized across the 78 million
273
00:19:01.130 --> 00:19:06.340
possible combinations out of those 177
which is a mere mortal. Humans can't do,
274
00:19:06.340 --> 00:19:09.210
it would take too long. And the things
themselves to your point about
275
00:19:09.210 --> 00:19:14.750
character limits totally with you. We
ultimately, these are humans coming up
276
00:19:14.750 --> 00:19:20.020
with ideas. Humans getting more time to
do this stuff, we're uniquely good at
277
00:19:20.020 --> 00:19:23.300
doing. We're uniquely great and
empathizing with our prospective
278
00:19:23.300 --> 00:19:28.310
customers were uniquely great at
creatively coming up with ideas to
279
00:19:28.310 --> 00:19:31.260
control them further through the funnel
to get them to become customers.
280
00:19:31.270 --> 00:19:36.330
Machines are particularly good at
acting at 1 to 1 level of precision
281
00:19:36.340 --> 00:19:40.500
listening and responding 24 7,
effectively auto updating these rules
282
00:19:40.500 --> 00:19:44.510
all the time every few minutes. And so
yes, a human puts the character limit
283
00:19:44.510 --> 00:19:47.760
by saying this is a brand safe thing I
want to put on my site. Machine
284
00:19:47.760 --> 00:19:50.250
learning can do a better job than we
can of figuring out what do I show to
285
00:19:50.250 --> 00:19:54.600
him and then to your last point you can
then overlay rules with the machine
286
00:19:54.600 --> 00:19:58.090
learning, right? You can say these five
headlines. I show them my best
287
00:19:58.090 --> 00:20:01.980
customers. These five headlines, I
showed them to the folks, I really
288
00:20:01.980 --> 00:20:04.610
don't want to talk to my sales team, I
want them to go to Self sir. And then
289
00:20:04.610 --> 00:20:10.050
let the machine learning within those
ma'am now it's part of the conversation
290
00:20:10.050 --> 00:20:13.400
that was really looking forward to,
let's dive into machine learning. And
291
00:20:13.410 --> 00:20:17.520
just because I've always had a
hesitancy with machine learning knowing
292
00:20:17.520 --> 00:20:20.680
that like oh my gosh, I would always
think like machine learning was for the
293
00:20:20.680 --> 00:20:26.010
big guys, the amazons, the Expedia's
that people with millions, hundreds of
294
00:20:26.010 --> 00:20:29.400
millions of page views that can split
tests, I mean the machine can take over
295
00:20:29.400 --> 00:20:32.490
because it has so much data to play
with and maybe that's something I've
296
00:20:32.490 --> 00:20:37.300
been told, maybe incorrectly that it
takes a lot of data for machines to
297
00:20:37.300 --> 00:20:40.920
learn. But you were telling me just on
this protocol, that might not be the
298
00:20:40.920 --> 00:20:45.250
case. So tell me why, why doesn't it
take so much data? Or is that is that
299
00:20:45.250 --> 00:20:48.310
total misconception? Where does that
misconception come out of? And how much
300
00:20:48.310 --> 00:20:52.300
data does it actually take for a
machine to learn uh even in a split
301
00:20:52.300 --> 00:20:58.460
test combination where you have 150
headlines to test then I I think the
302
00:20:58.460 --> 00:21:01.950
question is fought on, I think we've
all been trained on that. You read
303
00:21:01.950 --> 00:21:07.600
about deep learning in the press. That
particular approach to machine learning
304
00:21:07.600 --> 00:21:12.400
was built for google level data to go
figure out how do you understand, you
305
00:21:12.400 --> 00:21:16.560
know things like what's in an image and
to take that approach, you need that
306
00:21:16.570 --> 00:21:21.000
amazon Expedia level of data, you can
take a bunch of very experienced
307
00:21:21.010 --> 00:21:24.150
machine learning folks and create a
different approach. I'm very fortunate
308
00:21:24.150 --> 00:21:28.170
to work with two co founders who were
the two most senior engineers that led
309
00:21:28.180 --> 00:21:32.720
a several 100 person team that used Ml
to personalize the yahoo homepage for
310
00:21:32.720 --> 00:21:36.250
more than half a decade when it was the
most popular page on the internet that
311
00:21:36.250 --> 00:21:40.920
led to some insights on how to do this
with very little data. So those 78
312
00:21:40.920 --> 00:21:45.440
million possible combinations versions
of the page we tell our customers hey
313
00:21:45.450 --> 00:21:51.460
in every scenario using MLB faster than
split testing. But a good rule of thumb.
314
00:21:52.040 --> 00:21:55.670
I was in a day, you're getting 1000 a
day. That's a good balance between
315
00:21:55.670 --> 00:21:59.860
trying all of your ideas at the same
time and getting answers quickly and
316
00:21:59.860 --> 00:22:03.900
that's limitations for us, mere mortal
marketers, not the limitations of the M
317
00:22:03.900 --> 00:22:08.000
L. It will be faster and how do you do
that? It truly is a different approach
318
00:22:08.010 --> 00:22:11.540
like there 30 or 40 people in the world
that are really good at this kind of ml
319
00:22:11.540 --> 00:22:15.950
we feel very fortunate to work with a
bunch of them and it yeah, I'm with you.
320
00:22:15.950 --> 00:22:20.760
That is a hard science problem to solve.
It is counterintuitive. Um part of
321
00:22:20.770 --> 00:22:24.970
being able to solve that is actually
choosing to solve a different problem,
322
00:22:24.980 --> 00:22:27.700
recognizing ultimately it's about money.
We're trying to make more money, we're
323
00:22:27.700 --> 00:22:31.680
trying to drive more leads to sales to
get more customers. We're used to split
324
00:22:31.680 --> 00:22:35.930
testing where you say I need a big pile
of data, I'm gonna try to things and
325
00:22:35.930 --> 00:22:38.490
I'm gonna wait until there's a
statistically significant difference
326
00:22:38.490 --> 00:22:42.060
between the two of them. So, you know,
more data, I get to learn faster. I
327
00:22:42.060 --> 00:22:46.840
gotta wait if instead you say that's
the wrong problem to solve to make more
328
00:22:46.840 --> 00:22:50.990
money. I'm gonna Instead every few
minutes try to get better at predicting
329
00:22:50.990 --> 00:22:54.960
human behavior every few minutes. I'm
gonna try to be better than I was a few
330
00:22:54.960 --> 00:22:58.660
minutes ago at predicting human
behavior. If you frame the problem that
331
00:22:58.660 --> 00:23:02.950
way, you actually can take a totally
different approach and drive a whole
332
00:23:02.950 --> 00:23:07.160
lot of lift a whole bunch sooner
because you're not making a decision
333
00:23:07.160 --> 00:23:11.070
forever. If you're making a decision
forever. Like with split testing, you
334
00:23:11.070 --> 00:23:15.070
gotta get really good at making sure
it's not the wrong answer. You need,
335
00:23:15.070 --> 00:23:18.180
you use statistical significance for
that. If instead you're saying I just
336
00:23:18.180 --> 00:23:21.480
need to be better, there was a few
minutes ago, it's a very different
337
00:23:21.480 --> 00:23:26.490
problem to solve and that allows you to
learn every step of the way it allows
338
00:23:26.490 --> 00:23:29.720
you to learn off a whole lot less
traffic because your goal again is not
339
00:23:29.720 --> 00:23:33.170
to be perfect to be better than you
were a few minutes ago. The net result,
340
00:23:33.540 --> 00:23:37.530
you actually get results sooner. You
make more money along the way and you
341
00:23:37.530 --> 00:23:38.860
personalized for each visitor.
342
00:23:40.040 --> 00:23:43.660
That's interesting. So essentially it's,
it's just making choices faster. You're
343
00:23:43.660 --> 00:23:51.030
lowering the threshold from 95, or
whatever Accuracy to a lower threshold
344
00:23:51.040 --> 00:23:53.650
in order for it to start making
decisions sooner. So it's taking groups,
345
00:23:53.660 --> 00:23:57.910
maybe a little cohorts of 20 or 30
watching it being like, Okay, based on
346
00:23:57.910 --> 00:24:01.370
just the small test, we're thinking
it's going to be more like this, which
347
00:24:01.370 --> 00:24:04.900
means it can start test ruling in
different factors a lot sooner. Is that
348
00:24:04.900 --> 00:24:09.480
about right? It is very much the right
mentality. It is taking small groups
349
00:24:09.490 --> 00:24:14.350
every few minutes. It's updating itself.
The counter intuitive thing, static is
350
00:24:14.350 --> 00:24:20.160
actually not the goal. It's not that it
raised or lowered stat. Sig it said, I
351
00:24:20.160 --> 00:24:24.460
need to do I need to refine my
prediction every few minutes. Static is
352
00:24:24.460 --> 00:24:28.310
important for us. The marketers, it's
an important mile marker along the way
353
00:24:28.320 --> 00:24:33.640
to be able to say, do I want to bank
this? Do I want to use this in my email
354
00:24:33.650 --> 00:24:36.500
because I learned something new. Do I
want to use this in my ads. Do I want
355
00:24:36.500 --> 00:24:41.220
to use this in direct mail. Do I want
to use this in redesigning my website?
356
00:24:41.230 --> 00:24:45.960
But for the machine learning, it
actually doesn't care? Its goal is not
357
00:24:45.960 --> 00:24:49.400
to reach that. Sig its goal is to
continuously improve. And it turns out
358
00:24:49.400 --> 00:24:53.160
that if you do that, you actually reach
that six sooner. So it's not that it
359
00:24:53.160 --> 00:24:57.830
lowered the bar, solve the different
problem. That's interesting. I don't
360
00:24:57.830 --> 00:25:01.960
quite understand the math behind it,
but that's why I'm not a computer
361
00:25:01.960 --> 00:25:04.790
scientists dealing with these kinds of
things. So I'd like to take more into
362
00:25:04.790 --> 00:25:08.960
and understand the uh kind of the math
behind what it's doing because I'm
363
00:25:08.960 --> 00:25:12.110
still it's still using some kind of
calculations in order to get the
364
00:25:12.110 --> 00:25:16.450
learning right, has to use has to it's
a computer. So it has to use some kind
365
00:25:16.450 --> 00:25:21.920
of statistical variance to know you are
totally right and the difference is
366
00:25:21.930 --> 00:25:25.540
it's incrementally getting better
rather than like here's our final
367
00:25:25.540 --> 00:25:30.430
answer um happily, that's produced on
average, across all of our customers.
368
00:25:30.440 --> 00:25:37.500
46% lift across everybody over the last
two years. Looker used us to drive 44%
369
00:25:37.500 --> 00:25:44.670
more downstream. Stage two qualified
leads drift 300% 322% more SQL s
370
00:25:44.740 --> 00:25:50.060
Tableau 40% more digital transactions
using this approach. These are well
371
00:25:50.060 --> 00:25:54.440
known brands. Not only get a huge
amount of traffic to be able to do that
372
00:25:54.450 --> 00:26:00.870
is enabled in part by framing the
problem differently this way, wow. So
373
00:26:00.880 --> 00:26:04.050
if you're a small player and not one of
these huge brands, how do you get
374
00:26:04.050 --> 00:26:07.370
started? What do you even know? Do you
just have to wait until you're a huge
375
00:26:07.370 --> 00:26:11.780
brand to be able to test this out? It's
a great question. Two parts of the
376
00:26:11.780 --> 00:26:12.860
answer. one.
377
00:26:13.940 --> 00:26:18.900
If you have the resources to devote to
it, I humbly submit, go do this right
378
00:26:18.900 --> 00:26:23.380
away because you get to skip learning a
whole bunch about stacks, checking this
379
00:26:23.380 --> 00:26:25.730
thing every day, bugging engineering.
Every time you find a winner, right?
380
00:26:25.740 --> 00:26:29.110
All the things you have to do with
split testing, you get to focus more
381
00:26:29.110 --> 00:26:33.940
time on empathizing with your customers
and coming up with creative ideas to
382
00:26:33.950 --> 00:26:38.830
turn prospects into customers. That's
great. If you don't have the resources
383
00:26:38.830 --> 00:26:43.950
to go after it then humbly submitted.
The way we normally think about a B. M.
384
00:26:43.960 --> 00:26:49.400
With rules. Great. Go give that a shot,
go try them out. Right. If you're
385
00:26:49.400 --> 00:26:53.470
getting If you don't have the resources
internally or if you're getting on and
386
00:26:53.470 --> 00:26:56.930
I'll make it up 10 a day. Right? If
you're getting 10 visitors a day, data
387
00:26:56.930 --> 00:27:00.820
driven marketing will be hard. You
should use your intuition. Absolutely.
388
00:27:00.950 --> 00:27:05.510
So you're recommending just sticking
with rules for now in doing the split
389
00:27:05.510 --> 00:27:08.400
test, doing those kinds of things and
working your way up to the point where
390
00:27:08.410 --> 00:27:11.710
you can afford some machine learning
tools, but do you have any like machine
391
00:27:11.710 --> 00:27:17.100
learning tools specifically that are
maybe basic but can still show you the
392
00:27:17.100 --> 00:27:21.780
power of what machine learning can do
in website personalization. I wish I
393
00:27:21.780 --> 00:27:27.160
had a better answer to that To me. It
is primarily around the resources you
394
00:27:27.160 --> 00:27:33.310
have the attention, you can give this.
And if you don't have those or you have
395
00:27:33.310 --> 00:27:37.970
like 10 a day on traffic, then I would
suggest using if this then that or
396
00:27:37.970 --> 00:27:43.140
straight split testing. Straight baby
testing. I transparently don't don't
397
00:27:43.140 --> 00:27:48.830
have a great answer on like what's an
inexpensive way or or to do Ml happily
398
00:27:48.830 --> 00:27:54.120
Ml When correctly applied, he's a big
force multiplier for us, right? Like
399
00:27:54.130 --> 00:27:57.140
none of the marketers with whom we work
would have time to create 78 million
400
00:27:57.140 --> 00:28:02.450
different ideas. No human would And
most of them did not create 177
401
00:28:02.450 --> 00:28:05.990
different pieces of new content. They?
Re used a bunch of content they already
402
00:28:05.990 --> 00:28:09.830
had on their site, they tweet headlines,
they changed calls to action, but like
403
00:28:09.830 --> 00:28:12.590
a whole new case study, they didn't go
to that. They used the stuff they
404
00:28:12.590 --> 00:28:17.970
already had and were able to get more
bang for the buck off the investments
405
00:28:17.970 --> 00:28:22.200
they've already made using Ml And again,
if you don't have the resources to go
406
00:28:22.200 --> 00:28:26.810
do that, you know, do straight split
test or try try rules would be my
407
00:28:26.810 --> 00:28:31.370
humble suggestion. Fantastic. I don't
think as you were talking to mike, I
408
00:28:31.370 --> 00:28:34.640
think this is this is kind of a hack
and maybe a shortcut way of using
409
00:28:34.640 --> 00:28:37.860
machine learning or using someone
else's machine learning tool and then
410
00:28:37.870 --> 00:28:40.990
kind of applying it to a different
setting. But I do find that it works a
411
00:28:40.990 --> 00:28:45.430
little bit. One thing I've done is take
advantage of Facebook's algorithm
412
00:28:45.440 --> 00:28:48.850
because Facebook's really good at
figuring out which add an image
413
00:28:48.850 --> 00:28:53.150
combination in which text is going to
work now is not personalized, but you
414
00:28:53.150 --> 00:28:57.740
can still find a test 150, literally
150 different headlines with a bunch of
415
00:28:57.740 --> 00:29:01.140
different images go out and spend a
couple 100 bucks on facebook and it'll
416
00:29:01.140 --> 00:29:03.540
tell you which one works better and
then you just go and put that on your
417
00:29:03.540 --> 00:29:06.580
home page or your landing page because
it's probably gonna be better than the
418
00:29:06.580 --> 00:29:09.860
ones before, even though it's in a
different context, so it's not a true
419
00:29:10.240 --> 00:29:15.590
Fit, but it's probably going to be
better than the bottom 50% of what you
420
00:29:15.600 --> 00:29:21.790
put out there to test dan. That is a
great idea. I'm gonna share that with
421
00:29:21.790 --> 00:29:27.430
others if that's okay with you. Uh, it
resonates a lot. It's not an accident
422
00:29:27.430 --> 00:29:30.130
that the thing we're talking about
doing on websites feels a lot like what
423
00:29:30.130 --> 00:29:33.540
you just described in ads because I
spent the last 15 years in antec right?
424
00:29:33.540 --> 00:29:38.390
We, it made no sense to us at all did
facebook or google will say, hey, give
425
00:29:38.390 --> 00:29:41.890
me five search creative, I'll figure
out the right ones to show to whom I'll
426
00:29:41.890 --> 00:29:46.270
show the better ones more. I'll show
the worst ones less. Nobody done that
427
00:29:46.640 --> 00:29:50.910
on websites. The bottom of the funnel
where this money is actually made. Okay.
428
00:29:50.910 --> 00:29:54.840
So we did. And what you're talking
about is a proxy for, how do I get that
429
00:29:54.840 --> 00:29:58.110
research in the end? I'm not going to
personalize on my site using that
430
00:29:58.110 --> 00:30:00.890
approach. Right. I'm gonna do winner
take all the show everybody the same
431
00:30:00.890 --> 00:30:05.730
experience. But I will learn right.
It's I, I can to your point, you are
432
00:30:05.730 --> 00:30:10.210
effectively running a multi variant
test on more traffic than you get to
433
00:30:10.210 --> 00:30:13.860
your website because you're paying for
the impressions. That's it. If you're
434
00:30:13.860 --> 00:30:17.490
going to run a multi variant test to go
find a winner, take all winter to put
435
00:30:17.490 --> 00:30:21.230
on your site. That's a thoughtful way
to do it. At least. That's what I
436
00:30:21.230 --> 00:30:24.360
usually tell customers who want to run
facebook ads and I'm like, I don't
437
00:30:24.360 --> 00:30:26.890
expect a lot of conversions on the
front and think of it as like market
438
00:30:26.890 --> 00:30:32.600
research. Right. And then maybe we go
get some great zero later. Great lead.
439
00:30:32.610 --> 00:30:36.640
A great amount of leads later. It's
been a fascinating conversation guy. If
440
00:30:36.640 --> 00:30:41.460
people want to learn more about this
topic of website personalization
441
00:30:41.460 --> 00:30:45.780
machine learning and how you can apply
to a B. M, where can they find you
442
00:30:45.780 --> 00:30:50.090
online and your company in order to
learn more. You generously shared
443
00:30:50.090 --> 00:30:54.530
earlier in televised dot com. Uh, it
will have a bunch of content around
444
00:30:54.530 --> 00:30:59.040
this. We've done a bunch of webinars
taught a couple of course, is really
445
00:30:59.040 --> 00:31:04.180
tried to talk to us as fellow marketers
about how to get the most out of m. L
446
00:31:04.190 --> 00:31:10.610
uh in your marketing efforts. And for
me um G Y A L I F on twitter, my handle
447
00:31:10.610 --> 00:31:14.050
first initial last night. G Leaf look
forward to continuing the conversation
448
00:31:14.050 --> 00:31:18.310
there and dan, thank you so very, very
much for having me too. Fantastic.
449
00:31:18.320 --> 00:31:23.370
Thanks again for joining me on GDP
growth. One of the things we've learned
450
00:31:23.370 --> 00:31:28.090
about podcast audience growth is that
word of mouth works. It works really,
451
00:31:28.090 --> 00:31:31.690
really well actually. So if you love
this show, it would be awesome if you
452
00:31:31.690 --> 00:31:35.720
texted a friend to tell them about it.
And if you send me a text with a
453
00:31:35.720 --> 00:31:40.010
screenshot of the text you sent to your
friend, meta. I know I'll send you a
454
00:31:40.010 --> 00:31:43.410
copy of my book, content based
networking, how to instantly connect
455
00:31:43.420 --> 00:31:50.640
with anyone you want to know. My cell
phone number is 40749033 to 8. Happy
456
00:31:50.640 --> 00:31:51.450
texting