Why You Should Be Using AI In Your eCommerce Business

welcome to the ecommerce podcast with matt edmondson a show that brings you regular
interviews tips and tools for building your business online [Music]
hi and welcome to the ecommerce podcast with me your host matt edmondson all of
this week's notes and links can be found on our website at ecommercepodcast.net forward slash
now ai artificial intelligence it is becoming
more and more popular in the e-commerce industry and for good reason in today's show you are going to
learn why you should be using ai in your business and how it can help you increase sales and grow online in
essence can ai help us deliver e-commerce
wow that's the question a lot of people are still skeptical about using ai in
their business which i get maybe it's witchcraft or maybe it's
something beyond understanding i don't know but we've got a great guest today sheniff and he's here to show us that it
is in fact a valuable tool that can help you and i increase sales online and grow
our online web stores so stick around and let shaniff explain a little bit
further don't go anywhere hey there are you a business owner here
at orion digital we know firsthand that running an ecommerce business can be really hard work
as the online space gets more competitive it is becoming even more challenging to stay ahead of the curve
we totally get it so we want to help you succeed by offering a wide range of services from fulfillment marketing
customer service and even coaching and consulting just so that you can do what matters
most save yourself the time and the money and let us handle the day-to-day tasks this way you can run your business
without having to worry about the boring stuff so what do you say are we a good fit for each other come check us out at
oreodigital.com and let us know what you think
thanks for joining us on the ecommerce podcast it is great that you are here
now whether you are just starting out or if like me you've been around a while a bit of an e-commerce dinosaur as they
say our goal on this show is simple it's it's a podcast that is all about helping
you deliver e-commerce well how to make your own e-commerce stores bigger and
better and to do that every week we bring two things one great show sponsors
who we're going to help you so do check them out but also amazing experts who we
get to interview we get to chat with uh we get to dig into their stories you know their
ideas their principles the things that are going to help you and me grow online
and today we get to talk to shane f dahanini he is the co-founder and ceo of
aptio he's a data scientist bit of a geek software engineer
that helps he's got a passion for helping businesses make better decisions using data we all need data let's just
face it now prior to aptio saneff was the lead engineer and head of analytics
at tap commerce which was a new york city-based ad tech startup which was
acquired by twitter so i guess elon musk now owns it right
most recently he was a data scientist and software engineer at twitter itself
yes he was where he worked on a variety of projects across advertising and analytics and
machine learning so prior to twitter price tapconos shaniff was a software
consultant at booze allen so you know what he's got a good resume right he knows what he's talking about he's got a
bachelor of science in computer science from the houston university university the university of virginia and an mba
from nyu stern school of business the man is qualified to talk to us about ai
and he is going to tell us why we should care and what are some of the things that we should be thinking about so without further ado here's my
conversation with the phenomenal shannon thanks for joining me on the ecommerce podcast great that you're here i've been
Meet Matt
looking forward to this conversation since our pre-call because um we are going to be talking about all
things a i but one of the things that got me excited in our pre-chores the
fact that you actually run a small ecommerce business because for years we've seen these as two separate
things you know you can't have ai and be a small business because you needed big data i think was the official thing
so we've got a lot of good stuff to get into so thank you for being here i really appreciate it just for the good
folks listening whereabouts in the world are you matt thanks so much for having me i am in new york city midtown manhattan for
those of you who know much about new york midtown manhattan and are you uh are you a new yorker by trade are you are you an
import into new york city much like most of us i am an import but
i have been here for more than years now so whenever i talk to anyone who is uh who's anywhere else i'll say i'm from
new york but uh was originally born outside of uh washington dc in the suburbs of northern virginia oh wow so
not too far just down the road just down the road that's a quick train ride these days not
too bad not too bad at all why it's great that you're here all the way from new york city uh i have to admit i've i've not been to
new york since i think was the last time i was there mainly because of covert
yeah and i am i remembered we i did a trip with my boys i have three kids two
boys and a daughter and i did a trip with my two boys and we uh went
to various places of which new york was one of them we went for a few days they really wanted to see the city and we had
such a great time i really really like new york it's like the fourth time i've been there i really love it so i get why
people stay there um you know what they get sort of drawn in and unhooked is that is that is that
what's happened to you you've just got hooked you could never leave i just got hooked you know i came here for business school and then i i
actually had to leave for about a year and a half for work and i couldn't stand it so i came back and you know it's just been one of those
things where you know new york life uh it's it's a tough living but it's also
really really engaging and really exciting and so i'm still here and i don't know how much longer but i
anticipate i'll have ties to new york for a while yeah yeah i get it i do i get we i suppose we have friends in
london who live in london and i every time i come out of london i go i just don't understand why they live in london
then i go to london and go i totally understand why they live in london it's a it's a really fascinating thing yeah
yeah the biggest deal yeah yeah so um so ai let's get into ai um
What is AI
and talk about that so let's start right at the beginning what do we mean when we say phrases like a.i
and why should i pay attention you know i that word and that phrase has become such a buzz word these days it
almost doesn't mean anything um so my background is in sort of data science and machine learning so i'll break it
down for you really quickly hopefully not boring any of these ai means a lot of stuff today it can
mean anything from uh like kind of trying to predict the weather all the way up to having a big sort of
sophisticated game engine that can beat the world's best uh player in a particular sophisticated game so it's
this really all-encompassing term generally what you're hearing about in ai when it comes to business
is more along the lines of predictive analytics so trying to forecast or
predict a particular interesting thing that somebody cares about and that's really uh if you think about
it it's kind of advanced statistics you know you take a bunch of data you take a bunch of numbers you see what those
numbers essentially relate to in a metric you care about and then you generalize that pattern and that's
generally what we're going to be talking about at least what i'll be talking about today uh when i refer to ai
yeah no totally because like you say every everything has now got the phrase ai involved in it i can buy ai headsets
like jeremy there's all kinds of things um i can get a uh something that cuts my grass using ai i
mean hey i'm mows my lawn now um you know there's barbecues they've got ai in them i did
this just everywhere um and i i don't know if you're old enough yeah i don't know how old you are uh
shane if compared to me i remember the movie ai coming out you know the spielberg one
yeah so i've grown up i've grown up i've sort of lived with this sort of rapid
introduction of this phrase ai so um i appreciate the explanation because like you say it just means
i don't think it really means a whole great deal these days doesn't mean anything now so you really have to kind of hone in and and and
narrow it down when you're talking about it today so why predictive analytics why that
Why Predictive Analytics
specific aspect why is that interesting yeah so we you know we we work a lot with
e-commerce stores and one of the things that they're having trouble with today is obviously driving sales from their
customers acquiring new customers and one of the most important things that
ecommerce source have to do is create highly targeted marketing they have to create campaigns that resonate
with their customers and without a lot of data to guide you you know you might be relying more on
intuition or perhaps your experience which is great but when it when it comes to creating
these campaigns being able to predict what your customers are going to do next makes it
a lot easier for you to create a campaign that's going to speak to them so when it comes to predictive analytics
if you know what maybe a customer is going to buy next or if you know that a customer is likely to never come back
unless you give them a discount code or you give them an offer then you can tailor your messaging a little bit more
appropriately and so that's why we focus on this particular area of ai it helps
our brands and our customers uh grow more sales by personalizing their marketing
so you this idea that um [Music] you can use predictive analytics to
figure out who's not going to come back and shop with you again for example was one of the the categories that you
mentioned how what how uh is it witchcraft or how how
How Does AI Know
does that should i mean well how do you how do you know that how does ai know that yeah
it's really as a data scientist i find this stuff really fascinating um i'll start off by saying this each individual
person can be wildly unpredictable it's really hard to know what a person is going to do but if you get a lot of
people together it's really really easy to figure out what's going to happen on on average across that group of people i
can we use this principle i've used this principle across many sort of jobs twitter my former startup this one
you can use this principle to maximize your you know your gains across different areas so what we do
is we let stores plug into aptio our startup once we do that we get all sorts
of information about their customers everything from you know where they live to their names so we can
predict their gender all the way down to like what did they purchase five years ago six years ago and then what do they go on to do after that so you can think
of it as a giant data set and the ai basically just takes a snapshot
in time of everybody and then they go on to see it goes on to see what did those people do
you know a year later or two years later and it takes another snapshot and it does the same thing what do those people
go on to do so it takes all of these snapshots in time and it generalizes across all the
patterns that it finds and then once it learns the patterns you can actually apply those patterns to predict what
people are going to do tomorrow a year in advance two years in advance and if you have enough data you actually
can start to do this with very high levels of accuracy so now if you're creating campaigns
based on let's say a group of people who are likely to never come back you're sending them a different message than a
campaign that's targeted to maybe a more smaller group of people who are likely to buy your second best-selling product
or third best-selling product and so basically like i said earlier you get a lot of data and you crunch the numbers
and you kind of find what uh what patterns there are uh that are associated with the metric that you care
about okay well one of the things that i mentioned earlier um
is when we've talked about ai on the show in the past i've i've always come away
with sort of two overriding thoughts one is this sounds really exciting and the other one is i don't know if my ecommerce businesses
have enough data to make this a viable thing right so
like you say one person is unpredictable but you you put a lot of people's data together you can start to predict with
some accuracy and i suppose the more data you have in theory the more accurate that data is
going to be um and so when you talk about big data or when you
talk about having you know enough data to get some information out of it
How Much Data Do You Need
what is that i don't know if there is a number for it but what is it where do i need to at what point does my business stop being
too small to being something that actually can start to think about ai
we generally find that customers or brands that have let's say at least
at the bare bare minimum let's say historical purchases can generally start
to make use of ai we ask our customers to generally have at least a thousand historical customers or historical
orders and it's a little bit of an arbitrary number but what we found is that that around
that order of magnitude generally starts to create meaningful recommendations meaningful predictions
if you've only got five or ten customers you know you're not really getting enough information you're probably just starting out you're probably not really
sure what you're trying to do and so you got bigger you know bigger problems or bigger sort of challenges that you're
trying to solve first once you get up to a couple of hundred customers you can use ai to a certain extent to try to
figure out what's going on next and then after a couple thousand customers uh you get pretty accurate so the more data you
have our customers who've got let's say historical orders will certainly have more accurate uh
predictions than somebody who's got a couple thousand but it's one of those things where it just gets better and better and better and you're generally
going to want some sort of minimum about that maybe a thousand like i said yeah for our stores okay
so which is not actually a huge amount i mean it might be a huge amount of people who are just starting now go man a thousand customers sounds like an
absolute dream um but actually uh if you're selling sort of smaller
products at least a thousand customers um with at least a thousand lines of order data
should be fairly doable so actually ai and the science behind it then
Is AI Becoming More Accessible
is it starting to become more accessible to the smaller e-commerce companies
i do think it is and it's not like the smaller e-commerce companies are hiring ai engineers to go in and do this for
themselves i think what we're starting to see is you've got services apps like ours tools like ours
that are coming around to do certain things and those tools are incorporating generally best practices
for ai into the services that they offer so you're not again you're not sort of
getting um you know someone to come in and crunch the numbers for you in order to come up with let's say the best
product to recommend to someone but you are starting to see these apps that are uh developed by folks who have a
background in machine learning and data science who can do certain tasks for you and those those tools are using ai in
the background and i'm starting to see a lot of uh stores and brands consolidating or actually aggregating
lots of these tools into their store using them all in collaboration in conjunction with each other to sort of
improve their business this basically now means that you as a marketer you don't need to have maybe the same level
of advanced analytic skills that you might have had to before these tools existed now it just means you need to
know how to use these tools and understand what they're doing so you can grow your business a little bit uh more
efficiently so does that mean uh and and let me start off by saying quite
frankly it's a beautiful thing that i don't have to be as you know in tune with my analytics as maybe i used to be
um but the
so does that mean uh learning how to use these tools um because the the machine
the algorithm that is doing the data interpretation for you it's spitting out some numbers at the
other end so i'm the number of the data that it's giving you at the other end is that
What Is Targeted Marketing
making it so that what i should be focusing on is very um targeted marketing is saying basically the there
are these customer types these customers are in this um you go away and figure out how to market
that group i can i'm telling you who's in it for for our case that's exactly right so
we'll essentially you know look at all of your data and we'll output something that says this group of customers will
likely respond to this offer this group of customers will likely respond to this offer different tools might actually do
things for you differently so one of the things we're building is now once you've got this group of customers you still as
a marketer need to take that group we'll you know sync them we'll sync them to your tools automatically like for
example clayview or mailchimp or whatever but you still have to go into your email tool and set up the campaign
now one of the things that we're working on is being able to automate that step so you've got your group of people who
are likely to respond to an to an offer click a button and we'll set up the campaign for you using a different type
of ai using something like maybe natural language processing where we can start to predict what's the best headline or
the best sort of subject line to send to a customer and then all you have to do as a marketer is go and tweak it and send it other tools are doing different
things you know other tools might be saying have have your customers fill out a question quiz we'll collect that data
we'll feed it into a system that learns what sort of preferences your customers have and then we'll come up with maybe
an on-site product recommendation for your customers and that recommendation gets better and better and better over time so you as a marketer then don't
even need to worry about customer data or groups of customers you can just have this automated system select the best
product for a customer who's looking at your website right now and so different ai systems are doing different things
in our case what we like to do is provide a tool for the human marketer because the marketers are the ones who are driving all these sales they're the
ones who are very creative they're they're having trouble kind of trying to figure out what campaigns to create
they're not they don't want to spend hours in excel and so what we're doing is we're say we're basically saying man
plus machine will help you drive sales the ai is a tool that you can use to figure out how to make better campaigns
going forward and it's the data that is coming out of ai now i mean you say you know the more
What Is Reasonable Accuracy
data you have the more you can start to predict with reasonable accuracy and what's happening what does
that mean though reasonable accuracy when it comes to the data um how how
reasonable i don't i didn't ever envisage asking this question but how how reasonable is
that data the data well i'll say as a there's a couple of really interesting things to
dive into here as a data scientist the machine learning engineer most of my time is spent on
kind of grunt work cleaning up data making sure that someone's name is capitalized properly making sure that
you know there's no missing sort of information um so the the question you had actually has kind of a hidden
question inside of it which is data can be kind of unclean and we need to spend a lot of time cleaning it now let's say
you you clean it you get a lot of data you know how how useful is it in predicting what's going to happen next
there's a couple of ways to measure this there's the sort of in-depth ai way where you've got all
sorts of technical metrics to talk about how accurate your you know your ai model is these are things like
precision recall area under the curve roic things that most non-practitioners
wouldn't understand so when we sort of start to talk about the the efficacy of
our system we generally talk about it in terms of what it can do for you for your marketing and so what we try to
do is we say you know maybe you were running a campaign before that had less than one return on ad spend maybe
you are targeting a very broad interest group let's say you're a fashion company starting a broad interest group you know
maybe maybe women between the ages of and with just a lot of dresses
you come in you say look maybe we could start to be maybe we could start to use ai to
get more precise so you come in you say okay we're going to use ai to predict who's most likely to buy our specific
products you start running a campaign against them maybe your reach is a little bit smaller but your return on ad spend
suddenly goes from less than one to maybe one and a half or two and suddenly you become profitable
by targeting the right people at the right time so from our perspective our ai helps you
run those sort of more profitable campaigns we've generally taken customers from less than one row as to
above one particularly especially after the facebook sort of issues now with ios privacy updates we're able to sort of
get people back to those original pre-pre-numbers but we're also able to do things like target your
emails a little bit better so now instead of your open rates being maybe thirty percent forty percent we can get
them up to sixty percent and maybe your click-through rates go from five to ten percent to twenty percent and your
conversion rates go higher so what we generally try to do is frame the frame the accuracy of a metric of an ai system
within the context of what our customers are trying to do which is create more accurate campaigns and drive more sales
now matt that being said i am happy to jump into the technicals and talk all about that stuff if you want but i suspect i suspect you might go to sleep
so no no no i i part of me would actually love to go there shane if i'm not gonna like um
but it's i i i get that it's mathematics which is way way above what i learned at school is you know is how you analyze
this data um there's a couple of things that you mentioned there that i that i want to get into one you talked about cleaning
data um two you talked about um better facebook ads campaigns um
and three you talked obviously about email and how i i can help with conversion
rates and open rates which are all great things to say i want to understand how that works so we're going to get into that in part two
Show Sponsors
of the podcast but first we're going to take a quick second to hear from this week's show sponsors
did you know that nutrition is one of the keys to maintaining the energy you need to drive your business forward
vegetology creates incredible unique supplements in an eco-friendly ethical
and sustainable way that feed your body with the precise nutrients it needs
we're not just making you healthier we're helping to protect our planet too our products are vegan friendly and
approved by the vegan and vegetarian society plus they're gluten free so they fit perfectly into any lifestyle they
also contain no artificial colors or flavors making them good for your taste buds too you can feel good about your
food choices with our healthy natural supplements we have something for everyone whether you want to boost your
immune system or just get more energy every day and we're always working on new ingredients so that we can provide
even better products in the future so what are you waiting for get started now by heading over to vegetology.com
[Music]
hey there are you a business owner here at oregon digital we know firsthand that running an e-commerce business can be
really hard work as the online space gets more competitive it is becoming even more
challenging to stay ahead of the curve we totally get it so we want to help you succeed by offering a wide range of
services from fulfillment marketing customer service and even coaching and consulting just so that you can do what
matters most save yourself the time and the money and let us handle the day-to-day tasks this way you can run
your business without having to worry about the boring stuff so what do you say are we a good fit for each other
come check us out at oreodigital.com and let us know what you think
well a big shout out and a big thanks to our show sponsors i'm back here with the legendary shanique talking about all
things ai trying to understand without getting too into the theoretical physics
of it um uh my uh i've not told you this and if my eldest son he's at st andrews at the
moment doing a degree in theoretical physics so i'm sure he would have a whole bunch
of questions for you around this he does all that kind of modeling and it just wrecks my head um before
uh before you you were talking about how you can um you know we're talking about the reasonable reasonableness of the
data how accurate is the data and you kind of flipped it on the head and say well actually what you want to see is you want to see an increase in the roads
in areas like facebook advertising so a lot of people
especially because of ios is now getting a row as of less than one
when doing outreach to cold audiences and with their facebook ads so how does how does your system or how
Target Interest Groups
does ai tell me to target an interest group on facebook
using data from previous orders i mean if you can sort of join up that triangle for me i'd appreciate it
yeah we uh you know you you use the phrase uh flipped it on its head so we kind of flip the concept on its head we
no longer think about broad level interest groups at least for the customers who are using
aptio and the reason is we now know exactly what each customer that has ever
purchased from you is likely to do next and we understand which customers are likely to come back
which customers are likely to spend a lot in the future uh which customers have spent a lot in the past
so now what you can do as a marketer is you can for example take a group of your best customers let's say
maybe your biggest spender is your most loyal customers and maybe maybe a group of people who are forecasted to spend a
lot in the next six months take these groups of people send them over to facebook and create a
look-alike audience off of them and then when you start running new customer acquisition campaigns to that look-alike
audience what you're doing is you're actually providing facebook with some of the
insights and signals that it has lost post ios updates and you're doing that by providing providing facebook with a
highly highly precise targeted list of people who you want facebook to find now maybe facebook can no longer say okay i
can exactly find the people i can find exactly the group of people who look like this
like i could two years ago but by you providing a list of people who are similar to each other in terms of the
behavior of what they've done on your site facebook can sort of you know start to account for its loss of previous data it
still has information about these people and so i can start to find people who look kind of like your best vendors kind
of like your biggest customers and when you're doing this facebook can now more precisely target
new customers who will look like your best vendors and so it's a simple it's actually quite a
simple uh process and you could have done this before aptio but you might
have had to spend a lot of time in excel trying to crunch numbers on who your best customers were maybe you were able
to forecast who your best customers in the future would be but you were to do this sort of upload a
csv to facebook refresh the csv every couple of weeks with ai and automated systems
our system will find these people for you automatically send them to facebook automatically keep it up to date automatically and then you just use a
facebook acquisition campaign that's optimized using your brand's messaging your brand's you know value props
and what we found is actually this is one of the one of the strategies that works probably the best across all of our
customers every single campaign we run for customers and we're still small so i'll say every campaign so far
maybe not every campaign going forward has gone from negative one row s to positive one row as using new customer
acquisition strategies like this and so you're really just helping facebook recoup some of the data that it lost um
you know after the ios privacy updates by giving them better groups of customers to target so it's no longer
sort of about broad-based interest groups it's a little bit more about lookalikes and clusters of people who
look similar and using those sort of methodologies to find customers who are likely to spend more in the future
that's that is as i would say in england bloody clever and i think the um
remarkably simple but remarkably clever and i i i totally get that just for those who are listening that don't
understand what you mean when you say a lookalike audience on facebook it's probably worth just just given a quick understanding a quick
explanation of what you mean by that that's a really good point i should have defined that
so obviously facebook but other systems like even google ads they have this concept of allowing you
to upload a list of customers or a list of anybody really you can provide email addresses or phone numbers names
and you can upload this list of customers to facebook ads and then you can create basically an audience that
will be used to target these customers now one of the greatest things about facebook is the ability to find
similar groups of people to a to a pre-existing group of people or a group of people that you have just
uploaded as a list for example facebook calls this uh look alike audiences i
believe google ads calls it similar audiences concept is you provide a list of people
who for whatever reason you care about and then you uh create sort of this virtualized list of people who look like
them and when you what what the goal of doing that is is when you're creating ads you
can start to target ads to this virtualized look-alike group of people
with the ad messaging that you want to provide so that's really all it is it's just a a group of
a group of people around the world around the you know the facebook ecosystem or the google ads ecosystem
that facebook has you know deemed to be very similar to the group of people that you uh as a marketer have provided a
list for yeah that's this is really clever and so you're using ai to in
effect generate that list aren't you you're using the data analysts and other analytics or the analysis probably right
um to go away and go right here's the list of people you want to go find a look-alike audience for on facebook or
similar audience on google um okay so i get how it works with facebook
Email
how does it work with um you mentioned email and having better open rates on on email so how
does the system help there email is one of my my favorite channels
for e-commerce even across i think you mentioned my little tiny e-commerce store email is just a fantastic channel
but you're not generally using email to acquire new customers obviously somebody has to go to your website and put in
their email address before you can email them so what email is really good for is driving sales from people you already
have a relationship with but you know matt you probably get lots of emails every day i know i get at
least or most people are seeing you know hundreds and so the problem with email
is you have to cut through the noise you have to figure out how to make sure somebody sees your message opens it
clicks it and then goes on to make a purchase so when it comes to email the name of the game is
um it's all about personalization it's all about making sure that you're sending the right message to someone so that you
know your subject line and your preview text incentivize them to open it and then you're once they've opened it the
you know the headings the product images incentivize them to make a click then once they make a click they're on
your site and then your conversion rates come into play but it's all about getting them to care about that initial
sort of uh look into what the email is so all this goes to say you have to
personalize your emails you have to send sort of the right message or the right offer to customers
one of the best ways to do that is to just show them the products that they care about the most so for example maybe
you have products in your store and sure you might have a best seller but
not everybody's going to care about that best seller or maybe they've already purchased that bestseller and they don't want to buy it again but they're
interested in other products you as a marketer you can't go through and figure out you know out of the hundreds of
products you have what's each individual customer going to buy next but an ai system can do that it's really good at
doing that so what you can do you can just say okay ai system figure out what each
individual customer is likely to buy next or what they're interested in sync that to their customer profiles and then
when you go to create an email just pull that in pull that information in dynamically so that it fills in for each
customer automatically that's really one of the ways you can use email and if you do a campaign that contains these sort of
personalized emails and you run this campaign once every week or once every two weeks you actually start to be able
to generate revenue from a predictable stream you know a couple of our cosmetics brands are doing this sort of
email once a week or once every two weeks and they're generating very very consistent sales from it um and in fact
as they get more data you can start to see that the trend the amount of money they make from each email goes up over
time so that's one way you could do it um but you also have more traditional strategies you know you still have
people who you know might never come back you still have those people who might be at risk of of uh only buying once
our ai system can figure out who those people are and when you've got people like that you offer a discount letter
and so you say you know maybe a three email sequence where you say on the first email hey we've rolled out a bunch of new products come check us out and if
they don't respond to that the second email will offer a discount along with some maybe personalized product
recommendations and if they don't even respond to that the ml which actually tends to drive the most number of sales
is very sort of in your face it says hey we miss you we want you back take off straight to the point and a discount
ladder like that can actually do wonders in recovering people who are at risk
you've got sort of or different strategies like the two that i just mentioned when it comes to email as well as sms that you can
start to implement the key again is sending the right message to the right customer and in
order to do that you just have to know what your customers are you know you have to know a little bit about your customers and that's where the ai can
help again very smart isn't it it's like and as you're talking i'm thinking
in in essence what i don't want to do is send somebody that's used to buying blue shoes uh information on red shoes i i
just don't want to do that because that's just not going to help me um and using ai to simplify that entire
process uh and so okay i start to see how that works with email um the the third thing that you mentioned that i
Cleaning Data
just want to clarify a little bit on you said the bulk of your time is cleaning data
and you're capitalizing somebody's name um that sounds remarkably painful to me so
if i um but is is this cleaning of data well first of all what do you mean by
cleaning of data and secondly is this something that that i'm gonna have to do as a store owner to get the best out of
the um the ai at the other end it's uh you know it's a it's a
well-known secret these days that if you go into data science as an ai engineer or machine learning engineer you're not
spending a lot of time doing really interesting sophisticated modeling or math or any sort of things like that
what you're doing is you're spending a lot of time writing writing systems writing code creating processes to make sure that the
the data that your ai is learning from is reasonable so you know i've spent in
my career lots of time making sure data is reasonable now that doesn't mean that when a new brand signs
up i'm going into their data and sort of manually cleaning it and updating it what it does mean though is we have a
system a series of processes that you know before we start making predictions
on what someone's going to do in a store before we start analyzing all the data we just make sure
that it's reasonable you know if if you don't have the zip code for somebody you have to make sure that that's
represented properly either as a properly associated missing value
or maybe a lot of times what you do is sometimes you infer the zip code based on sort of
the most popular zip codes there's all of these techniques for dealing with bad data which is missing data
malform data you know sometimes phone numbers are coming in as just a series of numbers and maybe your system thinks
that's just a very large number machines are really dumb actually that machine might think it's just a really large number instead of you know a piece of
text that it has to interpret as as differently and so you're spending a lot of time making sure that the
formatting of this data is up to date and when i say you're spending a lot of time that means upfront before you
create the ai system or as you're creating it and improving it you're just making sure you have checks
in place that can take care of missing values that can take care of standardized formatting issues and then
you basically have processes that go through and assume the data is good to go and those processes go in and start
making predictions so as i guess what i'm trying to say is as a data scientist
a lot of time is spent sort of understanding the data understanding where things could go wrong where things are malformed what data is coming in
from the from the e-commerce platform what data do we have what data do we not have understanding all that so that i
can properly format it and my assumptions about it are correct so that the ai system can go in and do its thing
afterwards very good very good i just yeah that's good i'm
glad it's more you than me because i said visions of me having to go through an entire database while that customer
surname is not capitalized i need to resolve that one and that one's got a missing postcode yeah that would just be uh that would just be
a nightmare um and it's interesting though the data has to be of a certain quality to get
reasonable information out um and uh you you mentioned that machines inherently are
done uh uh which i think is quite an interesting thing because we do think everything's getting smarter and smarter but
you've still got to put good stuff in to get good stuff out um what's that phrase garbage in garbage
garbage in garbage that's right yeah and it applies to ai doesn't it so it's especially true with ai especially true
that's really fascinating so excuse me you mentioned um
What is Hydronize
i think we've both mentioned now you have a small e-commerce business um hydronized that's right
uh just quickly what is hydronizer what what is it you're selling online
hydronize is a really interesting small store um we specialize in hydrogen
infused water bottles water machines and you might be thinking what the heck
is this guy talking about ho hydrogen infused water it doesn't make any sense um well you know i didn't know this
before i i acquired the store i didn't build it from scratch but when i acquired the store i learned that there
are some studies out there that say that that water infused with additional hydrogen ions might have some health
benefits and so now you've got these water bottles they're really popular in japan and in asia these water bottles that
basically you turn them on and over maybe seconds to a minute
they will sort of infuse additional hydrogen ions into the water those hydrogen ions will remain for
minutes and the idea is you drink the water um while they're there and so you know some folks are saying look this
will help you sleep better this will help with inflammation some sort of initial initial studies out there but hydronize sells these machines
you know we're based on shopify we're a shopify store we've plugged into klavio we've plugged into postscript so a lot
of the same tools that our customers on aptio are using and we are you know we're trying to figure out sales in the
same way that our customers are and i will say having this store has helped me really empathize with my customers on my
software side um on the aptio side so that's us in a nutshell yeah well that i mean you know
we go talking about the product aside i i think it's fascinating that actually you're you're both doing ai and
e-commerce you're seeing it from both ends of the the stick now and which fascinated me when we talked initially
Does Hydronize have enough data
and so you've got this small store um does it have enough data to
to justify the the process that you've been talking about doing the whole ai sort of
um analysis yeah yeah it's a really small store but fortunately we have about
historical customers and so we are able to start to use a lot of the same
techniques that i talked about earlier in terms of you know automatically segmenting
customers who might never come back automatically segmenting your best customers identifying which products
your customers are likely to buy next now one of the things that i didn't quite mention earlier that it's worth
mentioning now is ai really works best when you've got a large product catalog with us we only have a few products and
so what we can start to do is we can start to blast out every product
every customer now maybe it's not going to be as precise as using ai would be but there's also not a lot of downside
in doing that for us and so from our perspective what we're interested in doing is more of that sort
of on the email side as i mentioned maybe we don't need to get super sophisticated but on the facebook
advertising side and google advertising side we can start to use information about our existing customers to acquire
new customers kind of like what we talked about uh during the paid social section and we are doing that and we're
fortunate we've got sort of the enough customers to do that i would love to i bet you if we had you know x as many
customers we would be able to use all that additional data to drive even more sales it's kind of like the rich get richer you know yeah um yeah so we're
still a little bit poor in that area but we are starting to use some of the some of the techniques i mentioned on hydrodyes
and so it's uh again it's interesting that you you this is how you're using your e-commerce store is to sort of build out
your facebook audiences um and you're using that to drive sales in a more successful way i assume it's more
successful since you've plugged the ai yeah it is it is yeah i just don't want to make the assumption
that's helpful it's very helpful yeah especially after those i you know before the ios updates we didn't have to worry
about ai and segmentation audiences we just said broad interest group health you know health based we were able to
target people of a certain age demographic it tends to be that folks who buy these water bottles tend to be above years old they're more
concerned about their health after the ios you know updates none of that worked anymore so now we have to do more
precise audience targeting and that seems to help us on hydrogenates i guess um
Can AI be used for anything
just sort of thinking about this audience target uh malarkey and
we've been obviously been talking about e-commerce here but i'm assuming actually
um it can be used for anything so say charities that have had over
donations from various people can use the same ai to analyze who's making donations to then try and predict
who would be likely to make a donation in the future right that's right you know i won't i won't
make a blanket statement but i i think there's a lot of similarities um e-commerce is really really good for
this type of stuff because again human behavior is very predictable you can start to create patterns i believe
charities would probably be be good for this as well there might be some industries where you can't do this for example
appliances furniture they might not be what we actually do work with a couple of furniture companies but any store or
organization where you're only sort of generating one one transaction per customer may not be optimal for this and so you
kind of do have to understand your business case and how ai can play into that um but in your example charities i
think would would probably do really well with a technique like this and i'm a huge proponent of sort of donations
and charities and so i'd love to work with charities that might be interested in doing this yeah so really i'm just thinking about the the possibilities
because i mean obviously you've got customer data coming in but i mean one of the things you mentioned on your website is
you don't have many products which you sell hydrogenate you have one or two products um and so you can't take full
advantage of the whole um the whole sort of feature set i suppose
of ai and it's helping you with facebook but it's not necessarily what you need
on for for email so is it a case of the larger my product category uh my product listings the better my
emails are going to be with the ai that's exactly right what we found you know and what a lot of our brands have
found is that if you have a long tail of products that's where ai really starts to shine
it's similar to google if you think about the way they work there's all these sort of random websites
available online but google is able to find the ones that best match your search query even if it's a very arcane
or strange query ai is able to find those sort of strange or arcane products that are you know
hidden away in your product catalog that you as the marketer may never have thought to promote to a particular group
of customers and so the more products you have the more precise you can start to the more sort of customer data you
have and the more products you have the better ai is compared to humans at matching
those products and customers together that's really incredible yeah yeah and and so i'm i'm again i'm now thinking
about the furniture guy who's just generated you know his it's not like it's not like a supplement
company that's selling a supplement every two months is it this guy's selling a couch hopefully once every years i mean it should last i suppose at
least a little bit um so there are industries where ai seems to be better at and so from what
i'm understanding um it's going to be better if you've got over customers and it's going to be better if
you've got a reasonably sized product catalog that doesn't exclude you from having ai and using it
for certain benefits but you're going to get the most benefit if you meet those criteria have i
understood that correctly that's that's exactly right you know in our case we say a thousand it could be
Retraining AI
but somewhere around that area uh that's exactly right and then it's just a case of using
technology um which i assume aptio your company does and most of the other ai systems out there
it's not just a one-time analysis you're continually re-learning and re-adapting and reprocessing is that correct
that's actually a really good point um you know back when we worked this process of there's kind of two processes
when you think about ai there's the process of learning which is generalizing the patterns that exist in the data and then there's the process of
predicting so actually applying those patterns to some some expectation going forward this happens at every ai company
the training process needs to continuously happen so some systems are able to do it you know continuously our
system does it every few days but you need to be able to update the patterns because humans even though they
are predictable they they evolve and change over time so patterns will shift and change over time and so that's one
of the key benefits of ai you don't have stale information coming in and informing your marketing decisions
information from is not affecting information or activities from we
believe in continuously updating sort of the patterns that are learned and then applying that to activities going
forward now one of the key things one of the really good things is ai can learn changes in patterns over time if you
provide ai with with some sort of information about when something happened like what was the date
it can start to compare different things that happen across time and sort of make adjustments automatically
so yeah long answer to your question but that's you know the process that i mentioned is called retraining retraining the ai model we retrain it
pretty continuously and then we apply its patterns and predictions going forward on a continuous basis
yeah that's quite helpful because like you say i think the mother of all failure is assumption isn't it and so you're if you
once you've got one data set and you assume that's just going to be the same from now until time and moral
then you you've got a you've got a fundamental problem so it needs to adapt and grow and learn uh which is uh a beautiful thing i mean
in terms of processing i mean this is just a pure interest question now uh i
i'm curious how much processing power is is there involved in machine learning
i've i've kind of got pictures of in my head of you know huge super computers uh in a
room somewhere or is this something that most machines can handle now
Processing Power
so it's it's funny because um just like how ai is kind of a loaded term and it doesn't mean anything the processing
power that's required to do ai can vary based on the tasks
now let's say let's say you just need to forecast your sales for the next year you can do this on excel you know using
some simple modeling but let's say you're google building this this what's called a reinforcement
learning system a system that can beat the best chess master not even chess today the best alphago mod or the best go
player in the world a very large system like that or a system that tries to sort of
download the entire internet and understand all of the text so that it can create new text systems like that
are going to require large amounts of processing you know very large sort of uh data centers with hundreds of servers
um and so what you're learning what we're seeing today is as time goes on and the
sophistication of ai improves at the highest end you're always going to need these very large data centers
and servers that are doing massive amounts of processing but what's considered highest end ai
evolves over time so today on our laptops we can do something that you couldn't do back in or and so
i'm able to run ai models on my macbook pro and a lot of these models
are the same models that we are using in production it's just that they might i might have developed them on my local
laptop now in order for us to run them across all of our customers at scale
i will maybe run these models on the cloud so on a you know on a server that i'll rent from amazon or google and it's
not a really you know when you think about the models that we're running they're not necessarily sitting in these large data centers with tons and tons of
computers they might be some you know a small amount of hardware dedicated to running aptio's code
and maybe it costs us a little bit of money every month but it's really not something that's that's eating up a lot
of energy it's not something that's you know doing a lot of work that's just the type of ai that we run
obviously other companies like microsoft google amazon they're going to run ai systems that are much much more energy
intensive and hardware intensive and so again it just kind of depends on the task that you're trying to accomplish
nah okay very good and sorry that was just a slight diversion i was kind of curious um
listen shaneaf this has been fantastic uh i i feel like my head is a lot clearer now when it comes to which is
which is very very good um if people want to connect with you if they want to reach out to you what's the best way
Free Trial
that they can do that i appreciate that matt um so if anyone's listening we are happy to offer an
extended free trial to your listeners they can reach out to us on our website it's aptioapteo.com
it's a co uh feel free to reach out to me directly shanief at aptio.com uh
we're on twitter aptio ai uh linkedin aptio so feel free to come on by we would love to get everyone set up we are
starting to see really good results from a lot of our customers and so um you know time is being kind of a little bit
uh hard to manage now so the earlier the better but we'll get you guys all set up for the free trial and go from there
legend and we will of course put all of those links in the show notes i guess in closing um shaneef what's been the most
surprising you've got clients coming on board you've seen a lot of data coming through and a lot of stuff going on i
suppose what's what's what's some of the surprising things that you've noticed i'm kind of curious on that yeah
i'll just add a general on a general sense i just find it surprising how many different things are sold online like
Surprising Things
we've got people who are joining us who are selling just just products i could never have even conceived um
doll heads uh automotive parts obviously it makes sense obviously cosmetics but there's just some ran medical devices
like there's all sorts of things online that people are selling um and i suspect this trend is going to continue now
should i have been surprised at the variety of this maybe not but just seeing it in action and seeing our ai
handle all of this different sorts of data sets it has been uh impressive it's been exciting to see and
so that's one of the things um that's been really surprising to me and then the other sort of on the flip side of
things i find it um you know it's surprising how many marketers are are having trouble today you know there's
just all sorts of challenges to marketing and driving sales and it's um
there's no standard playbook so maybe one company is doing something that a different company just never knew about
and so there's a lack of information there's a lack of knowledge across all of these sort of industries and marketers within
these industries that i think can start to be standardized and because me i'm a data guy i'm an ai guy i came from the
world of technology i just assumed everybody sort of has the same playbook when it comes to marketing and that's not the case so i would say those are
those are sort of two surprising things maybe flip sides of each other maybe two totally different things
that i found to be surprising fantastic fantastic well shanef really appreciate
your time thank you so much uh for being with us and again we will link to shane in the show notes thank you again thank
you man appreciate it well a big thanks to my very special guest shaniff what did you think are you
convinced now about ai i think you probably should be really uh how how's it going to work for you what can you do
i think it's all good stuff to think about isn't it and of course if you were unable to take your copious amounts of
notes which i know you wanted to take you can get all of the links the notes and the transcript from today's show at
ecommercepodcast.net forward slash and that's available for free no email address none of that sort of stuff
just head on over there and grab them and you're gonna want them because if you're like me ai is such a big topic
and it's becoming bigger and bigger and bigger uh and with that i feel like if you're
like me i i feel like it becomes more and more of something that can't apply to me something that i can't get
involved with because it feels complex when you throw in machine learning and algorithms and
expensive software and all this sort of stuff but actually the thing i liked about shaniff makes it all super accessible so do dig
into it and if that's not enough next week we have rich brooks coming to talk to us about how to find your
remarkability for competitive advantage that's awesome isn't it how to find your
remarkability uh for your competitive advantage does that sound interesting is that wetting your whistle well if not
here's an excerpt from next week's show and hopefully that will help you think yeah this is gonna be awesome because
there's always going to be somebody out there with a cheaper version of what we're trying to sell people and you can't be racing to the bottom
when it comes to price so it's really important it does mean sometimes when we choose
of something to focus on when it comes to our own remarkability or what helps us stand out online that we are pushing
away potential customers but that's okay the the idea is to find the audience
that is going to be loyal to you to be acolytes for your business acolytes for your e-commerce store and really work on
being the best we can be for those specific people rich it's such a fab blog you're not
going to want to miss it i am looking forward to this conversation let me tell you now if you enjoyed this week's
podcast then i would appreciate it if you could rate the show on itunes or wherever you get your podcast from and
of course share it out to the wider world let the good folks in the wider world know about it because the more
folks that we can connect with the bigger the podcast the bigger the podcast the better the sponsors the better the sponsors the better the
guests the better the guests the better the content you know the story you know the scoop right it's a win-win for everybody
so you know help me to help you as they said in jerry maguire
i really should learn to do that better anyway as i said at the start all of the notes links and transcripts to the today's show are available online and
you can get them for free at ecommercepodcast.net forward slash that's it from me thank you so
much for listening i'll be back again next week so uh wherever you're whatever you're doing i hope you are learning and
growing your e-commerce business i hope like me you are delivering e-commerce well great to see you see you soon bye
for now
you've been listening to the e-commerce podcast with matt edmondson join us next time for more interviews
tips and tools for building your business online
[Music]