Connected Commerce: AI Won’t Replace Your Team. It Will Multiply It.

Hai Mag 00:00

Welcome back everyone to Connected Commerce Podcast. And I have a great guest today. It's a distinct pleasure for me to have Ritu Java. And she's not only the founder and the CEO of PPC Ninja, but also a great resource who always talks about like how to improve e-commerce, Amazon, how to leverage AI. So it's great to

Hai Mag 00:26

Have heard today, so welcome, Ritu.

Ritu Java 00:29

Thank you so much. Hi. I'm I'm so honored with that introduction and I've known you for such a long time and it's always a pleasure chatting with you and very happy to be here. I'm excited.

Hai Mag 00:39

Absolutely. So it's almost I think five, six years the first time we met

Ritu Java 00:43

Yes.

Hai Mag 00:44

in one of these conferences and we love it. By the way, Ritu just told me that she will be at Amazon Accelerate and she will be speaking with the Amazon ads booth and she will be there too. Eva is

Ritu Java 00:57

Yes.

Hai Mag 00:57

sponsoring Amazon Accelerate. we do some amazing things there. I'll be a speaker too. We'll talk about connected.

Ritu Java 01:04

Exciting.

Hai Mag 01:04

Commerce, AI, advertising. So if you are planning to join Seattle, come and find us. It's not very difficult. Ritu, hi. Not like very common names, you know. At least, you know, maybe it's more common

Ritu Java 01:14

Yes. I know. Yes.

Hai Mag 01:18

in some other places in the world. But

Ritu Java 01:20

Yeah.

Hai Mag 01:21

I think, you know, what I like about like your name, Ritu Java is like already like a brand, you know. You don't need PPT ninja, you know, you just use

Ritu Java 01:29

That is so

Ritu Java 01:30

cool.

Hai Mag 01:30

Ritu Java

Hai Mag 01:31

Incorporation, that's it. You know, like I think you should trademark your name. I I should do the same with me. It's not it's very unique, you know, it's good to have it. So go ahead, go ahead.

Ritu Java 01:38

yes. Very unique. Yeah, yeah, yeah. No, thank yeah. Mm-hmm. Yeah, no, good. Yeah.

Hai Mag 01:45

you know, today we talk about connected commerce because it's

Ritu Java 01:49

Mm.

Hai Mag 01:49

not only anymore doing manual work with Amazon. I mean, we were just talking before the podcast, like

Hai Mag 01:57

Things emerged a lot now. Like AI is not only a supplement, it is just the foundation. It's fundamental for success in e-commerce. Data is always very important. It's the foundation of being successful, understanding data. Now AI helps us a lot more to understand the data, analyze it, come with the n suggestions. So I would like to ask, you know, to you, like, you know, tell us your

Hai Mag 02:25

connected commerce story and let's take it from there.

Ritu Java 02:30

Yeah, that sounds great. so you know, I had made a rule for myself several years ago that if there is anything that I'm repeating more than three times and I know that I'm gonna use it again in the future, I just turn it into an automation. So back in the day there was many ways of doing that. You know, we used to have Zapier.

Ritu Java 02:49

That was a common one. there was another tool that I used, if this, then then this or whatever. And so it was many, many different types of tools that I kind of evolved from and then started making my own tools where you know, if there was anything that I thought, okay, I'm gonna use this, let me turn it into an app script. and I can talk a little bit about what these things mean because I know ultimately, you know, whatever I'm gonna say, if it doesn't help.

Ritu Java 03:13

Our listeners, it doesn't make any sense, right? So I'm gonna talk about App Script, maybe, because App Script is one of the most powerful tools that anybody that's in the Google ecosystem can use, like literally for free. It's sitting right there, and most people don't even know about it. It is one of the most powerful ways to automate pretty much anything on Google, right? So I started doing that, and now we have arrived at the AI era and

Ritu Java 03:40

Like my connected story is about ops plus AI. And I'm on a mission to kind of you know pretty much automate things with AI, but there's a trick to it that a lot of people don't know and they don't really know how to integrate everything, and so maybe I'll talk a little bit about that. So it's helped our business so much. just to say BPC Ninja is an advertising and creative services agency.

Ritu Java 04:04

So not only do we run ads for Amazon, Walmart, Credo, and many other channels, also Chat GPT ads, we do all of those on the advertising side, but we also have you know, AI readiness for the listing side of things. So we get our listings ready for Rufus, for Alexa, for other channels that you know, need to be able to understand the listing and help improve the discoverability of products because people are using

Ritu Java 04:29

AI more and more to find the things that they really want instead of typing keywords. You know, like there's been a transition from keywords to conversations and I myself don't use the Amazon search bar as much as I used to in the past. I generally just start my conversation with Alexa or Rufus as you want to call it because they just name change the name in the US other markets still call it Rufus but within Alexa I can ask very complex things

Ritu Java 04:56

Like if I want a supplement of some sort, I even give it my symptoms and my requirements and the fact that I cannot swallow big bills. I need something small and I need it to be gel like and this and that. I can describe it in so much detail and it finds me the right product and I don't have to waste my time with keywords, right? So that transition has happened and people are not really

Ritu Java 05:19

bought into the idea that, you know, AI is here to judge their listings and it's there to kind of pick up whatever you're putting out there and make that content be visible to people who are actually looking for it. And so our agency helps with the whole gamut of not only advertising it but then making sure that the listing is ready for advertising so that the funnel is like you know complete and it feeds into each other. these two sides feed into each other so that the funnel moves faster and faster.

Ritu Java 05:45

So just talking a little bit about the connected story, you know, every agency has a set of SOPs, right? Standard operating procedures. This is how we do it. You know, every week we open you know someone's account, one of our clients' accounts, and then we go through certain steps and we say, Yes, I've done this, I've done this, I've done there's a checklist. and we do all the tasks and and we make sure that

Ritu Java 06:07

You know, their account is, you know, been taken care of, it's been optimized, new keywords have been found, anything that needed to be negated, it's been negated, budgets have been optimized, things like that. So SOPs are no longer as powerful as they used to be, because now you have AI. So what is the equivalent

Hai Mag 06:24

Right.

Ritu Java 06:25

of an SOP on the AI side of things? I would say that is the skills library that you have, right? So skills are

Ritu Java 06:32

One of the most powerful ways to train your AI to do something very predictably. If there was something that you would do through a checklist or a series of steps, you can then convert that into a skill because AI can actually help with automating that stuff through many ways. Like one of the ways that we use it is through AppScript, which is Google's method to kind of

Ritu Java 06:55

run apps or run scripts on top of a database. The database that we use is called BigQuery, which is the back end where all the data comes streaming in from Amazon ads, from SPAPI, from rank tracking sources, and so on. So it's kind of a it's an ecosystem that works really, really well where AI plus backend the data layer as well as the the the AI layer, they kind of work together

Ritu Java 07:19

To you know, make a like a like a huge difference to the way things are done and human errors are being eliminated, you know, and more and more quality work is being produced from this kind of a system. and I think we have seen the results of that. people are finding that you know, they're able to catch more things than they ever were before, things that are, you know, hiding, you know,

Ritu Java 07:42

you know, like s slowly moving objects are being more easy to discover. For example, slow moving keywords that slowly bleed you over time, they're being discovered now through the processes that we've set up and all the skills that we've set up. So it's a really smooth functioning system that we've we're evolving. And at this point we have close to 85 skills in the agency that are doing all kinds of things for for us.

Hai Mag 08:06

Now that's very interesting because many years back I was intrigued with the idea of SOPs, although I don't believe them in a way, because each time you write the SOP, it's expired. In a way, like I mean, you need to

Ritu Java 08:21

Yes.

Hai Mag 08:22

update the SOP the next month because either Amazon changed something or you met with another customer and they have a different way of doing things. That's one thing. The second thing is you said

Hai Mag 08:34

A very something that is really right. Like if you are repeating yourself three times, then it means it has to be automated. then we thought in the past way is let's put an SOP instead of this repetition. At least everybody knows how to do this, but then it changes again. People don't know how to navigate all the SOPs, whether you write Google Docs, let's do Notion, let's do this.

Hai Mag 09:00

But at the end of the day, there is something not working with that. And now you are talking about skills. what I want to, you know, what I want to maybe elaborate you, I want you to elaborate a bit on an example. Like let's say there

Ritu Java 09:13

Yes.

Hai Mag 09:14

was an SOP before and now it's a skill now. And anybody from using cloud or codecs, open AI,

Ritu Java 09:21

Hm.

Hai Mag 09:22

by the way, they can both use the same skills. That is one of the things that I love.

Hai Mag 09:27

I create a skill

Ritu Java 09:27

Yes.

Hai Mag 09:28

on cloud, but I can now leverage it with codecs because we have a 20 people development team. They will work on codecs, but our strategy team and everybody else is working on the cloud. So the skills are also shareable. But what is a very good example of like an SOP which now works as a skill for you?

Ritu Java 09:49

simple very annoying kind of action or so p that takes up so much time is finding negative keywords very very time consuming and the thing is you cannot just automate it it's not just a rule that you can set because it can actually hurt you so we did like this really cool analysis of you know what happens to keywords that you negate if you purely base yourself on number of clicks

Ritu Java 10:13

So let's say your threshold is 10 clicks, 15 clicks, 20 clicks, whatever, or some sort of multiple of the clicks required to have like two sales, like a conversion rate metric, something like that. So let's say let's assume that you're willing to tolerate about 20 clicks before you kill a keyword, right? And you negate it, right? Let's say that's my you know SOP or whatever. In the past,

Ritu Java 10:35

That would be you would have a look back window, right? So let's say you say look back over 30 days. in the 30-day look back period, if you find any keywords that have you know X number of clicks, then you know evaluate them. Are they relevant? Not relevant? If they are relevant, then find out what's wrong with the listing, why is it not converting, or if they're irrelevant, then kill them, or if they're relevant but they're not profitable, then also kill them, you know. So that's something that

Ritu Java 11:00

An SOP tells you what to do. Go go check it, then find the thing, and then kill it, right? In the AI era, you can do a lot more. So for example, you can actually we actually split that out into four different skills, not just one, because the way we negate is quite complex and very sophisticated because we don't want to kill keywords that are slow movers and have a potential of converting after a certain number of clicks. You know, there's

Ritu Java 11:25

There's a time that needs to be taken on a keyword before it can be written off, right? Because otherwise you'll write off almost every keyword. You'll be left with just a handful of exact matches or very few keywords, right? So

Hai Mag 11:38

Exactly.

Ritu Java 11:38

the the way we're doing it now is through a bunch of skills, and I can share share one of those skills with you. So one of the skills is basically searching for irrelevant keywords only, it's only job.

Ritu Java 11:48

Is to look at the search term report and find the most irrelevant keywords with respect to the actual product. So it does an analysis of the product, it figures out what the product has and what the product describes itself as, and then it's it'll comb through the search term report and then find all the irrelevant keywords that are that that are no-brainer. Like you've got to kill them because they're not even relevant for the product.

Ritu Java 12:12

Previously, this would have been a manual step of checking each and every one of those to see is this relevant, is this not relevant, you know, is a sweet cooker this a negative for a rice cooker? Like someone needs to use their brains to determine whether that's negative, relevant, and so on. Now AI can help us with that. So that's one of the four negation skills that we have.

Ritu Java 12:35

which completely improves the quality, the reliability, and it reduces the burden on the human to have to do this themselves. So AI can do it in a second or two, whereas humans would take a long, long time to filter through so many keywords. And then when you get to an agency size with you know 70, 80, 90 c you know companies that they're working with, it's not that easy. It's not it is pretty time consuming to do that and do it well.

Ritu Java 13:01

But but when we try to integrate AI into our systems, we can do so much more, right? another example is you know, we want to give our clients something of value, like you know, from time to time we want to share some some reports, share something about their positioning you know, on Amazon or some some sort of share of shelf or some sort of information. Now, all that previously.

Ritu Java 13:25

was very time consuming. But now we can do it cheaply. Now it doesn't mean that we have unlimited capacity. We we do have to be careful about how much because tokens are also important, right? So we need to worry a little bit about how much we're spending in terms of time, in terms of tokens, but you can get a lot of rich information out of your data sources if you have planned your data sources properly. So making sure that your data layer is clean, it has

Ritu Java 13:51

data streaming in and all your skills are you know up to date, they're tight, they're not like loose AI skills. They're very they they've got to be efficient. Like the efficiency of a of a skill means that it needs to have some amount of script in it and some amount of the actual LLM process of the AI process. It needs to have both sides to it in order for them to qualify, at least in our system. So the 85 plus skills that we have

Ritu Java 14:16

they all follow that rule and we try to make things efficient both for ourselves as well as for our clients because ultimately our efficiency, you know, goes back to our clients to to keep them happy and to to make sure that their businesses are growing.

Hai Mag 14:30

Absolutely. So this A to five skills, that's already amazing. I have two questions around that. The first one is like, you know, you basically let's go back to your the negation example. you give the the rules to AI. I mean, this is the way you should do it, and then the AI is already having some type of you know, thinking style and it does the things. But I also see that.

Hai Mag 14:53

There is still some level of hallucinations happen and AI decides on its own that something is relevant, but it's not. Maybe it doesn't happen like 50% of the time, but let's say 5% of the time, meaning probably it still makes sense to have that PPC specialist. That's why like I I am being I a person working with AI for the last 20 years from my IBM Oracle Times till now.

Hai Mag 15:21

I still see the same thing. There is a level of mistakes happening too. So does it mean that like you let the AI to run the skills and you know kind of move move on with that? Or you have that human person still like there is somebody on top of everything and checks, for example, if all the negations are done properly, although the skills are perfect and it works, but sometimes it doesn't work. Do you have that experience?

Ritu Java 15:46

Right, right.

Ritu Java 15:48

Absolutely. So the thing is that's where the design of the skills come in comes in. Like you don't wanna design skills that have a large margin of error. Like you wanna make sure that your margin of error is close to zero. I mean, obviously it's not gonna be 100% zero, even with human judgment, right? Like even with a human looking at it, they might say, I don't know, it could be this, or maybe it's negative, maybe it's not negative. There's always that little one percent chance that.

Ritu Java 16:14

There's gonna be a judgment call that even a human cannot pass. But the the way the skill is designed does matter because if the skill is designed purely as an AI operation, it could fail. It could have a lot more hallucination. But if a skill is designed with what I was explaining earlier, a left brain and a right brain thinking in mind. A left brain requires calculations, mathematics, statistics that become the foundation

Ritu Java 16:41

For the right brain, the AI brain, to actually do a much better job of determining the output. So, like if you just throw a search term report at AI and say, go find me negatives, it's probably going to hallucinate, you know. So that's not the way to do it, right? There's there's a different way, there's a way to kind of design the skill so that the statistical calculations are done first.

Ritu Java 17:04

And then the AI comes in and looks at those and then it finds whatever is the most irrelevant in that set. right. So so it's it's not as simple as saying, AI is gonna do everything for me. The human is the one that's designing the skills so that it can help the human. And yes, humans are in the loop all the time. Like we're not getting rid of humans anytime s anytime soon.

Ritu Java 17:26

Without them

Hai Mag 17:26

Exactly.

Ritu Java 17:26

things could you know, just be derailed. So we definitely have humans at every stage of the process. it's just that the humans time is you know, time usage has been optimized, I would say.

Hai Mag 17:37

Absolutely. I mean we do it in a bit of a different way where we have our own software which is leveraging the skills via agents, kind of the same way, but it's kind of a a software which is packaged packed

Ritu Java 17:51

Mm-hmm.

Hai Mag 17:51

skills because of two reasons. And first of all, now with the software, the PP like in the past, a PPC specialist like to have a

Hai Mag 18:01

Really good PPC specialists, we were spending at least six to twelve months to train the person. And some of this training happens real re with real customers, obviously. But right now the time is limited to one or two months because you know if it the person knows everything, then the skills or our software, which is called Viria, is really helping. So there is a ramp up time, it's much shorter, and after 12 months.

Hai Mag 18:27

That PPC specialist is maybe two times better than a human, just the PPC specialist without the skills. So I can understand it. My other question was more about now the skills. Actually, I have now again two questions. The first one is now

Ritu Java 18:43

Mm-hmm.

Hai Mag 18:44

you mentioned 85 skills. is this proprietary to your company, or you also share these skills with

Hai Mag 18:54

I don't know, publicly or with customers, like how do you handle that part? because in the past I remember even people selling a bunch of SOPs for a thousand bucks or something like that, and nobody's buying it anymore, but maybe the skills people will buy, or

Ritu Java 19:07

Yeah. No one no no. Yeah.

Hai Mag 19:12

maybe it will be a elite generation engine. I don't know. Like, are you leveraging it in some ways, the skills?

Ritu Java 19:18

No,

Ritu Java 19:19

not at all, not at all. So skills are very, very customized, I would say, to an operation, right? And you can't just give out skills and hope that someone will buy it. because yes, they might get the idea from that skill and maybe improvise it and make it their own, but I don't think anybody's skills are like ready to go.

Ritu Java 19:38

as is. You know, there's always this element that people need customization. There's the desire to customize for my workflow, for my SOP, for my, you know, system. And so we don't sell our skills. It is proprietary and we use it primarily for our own clients, for our agency clients. And that's how we see the the best kind of ROI on on the time we spend on building these these skills. because you're right, there was a time where people were

Ritu Java 20:04

you know, selling skills or selling SOPs and they get outdated. You know, this this stuff needs to be maintained and you can't just create it and and leave it alone. Yeah.

Hai Mag 20:14

You know, there is

Hai Mag 20:14

this LinkedIn lead generation way now, which is like people say, Okay, this skill replaces this thing, and you know, and I

Ritu Java 20:22

Ugh.

Hai Mag 20:24

don't know, reply skill and I'll send it to you. By the way, I

Ritu Java 20:27

Ha ha

Hai Mag 20:28

like the I did a test the last 15

Ritu Java 20:31

Yeah.

Hai Mag 20:32

days or so. I actually put an AI agent, which is like getting all types of skills and whatever people are providing as knowledge, as valuable knowledge.

Ritu Java 20:41

Yeah.

Hai Mag 20:41

out of the 20.

Hai Mag 20:44

The valuable things that I have seen is zero. Like I haven't seen anything that really

Ritu Java 20:48

Zero. Yeah.

Hai Mag 20:51

worked out beyond basic understanding, because as you said, every customer

Ritu Java 20:56

Correct. Yeah.

Hai Mag 20:58

is different because the product is different, context is different,

Ritu Java 21:01

Yes.

Hai Mag 21:01

competition is different, so it's different. There is only one thing that I am not sure. I mean, it's okay if you don't have a a kind of a

Hai Mag 21:10

Full answer because nobody knows that answer right now. Big customers

Ritu Java 21:11

Mm-hmm.

Hai Mag 21:14

that we work with, they are concerned about their data security and compliance. And the problem is now cloud or open AI, if it is not run in a private environment, which only happens with Bank of America's and all the other big guys,

Ritu Java 21:30

Yes.

Hai Mag 21:31

at the end of the day, you are giving all the information to an open

Ritu Java 21:35

Yes.

Hai Mag 21:36

system, you know, like

Hai Mag 21:39

How do you feel

Ritu Java 21:39

Right.

Hai Mag 21:39

about that piece? You know, like to kind of avoid that. At least like because

Ritu Java 21:43

Right.

Hai Mag 21:44

we we internally we try like we don't use

Ritu Java 21:48

Mm-hmm.

Hai Mag 21:49

open systems in the development. I mean that's the only way

Ritu Java 21:52

Right.

Hai Mag 21:53

we can limit it, but I don't know how you think about that part.

Ritu Java 21:57

Yeah, yeah. And so if you remember we had mentioned earlier about App Script, right? So that's

Hai Mag 22:02

Right.

Ritu Java 22:02

our preferred style of you know, working on the data so that we can get anomaly detection done or alerts that come out from the the backend data and you know all kinds of recommendations, etc. That's actually happening on our Google ecosystem, so it is fairly secure. It is as secure as Google would want it to be. Where the AI comes in is in

Ritu Java 22:23

informing those skills so that those actions can take place. You know, so it's not like all the data is going to, you know, OpenAI or to anthropic. It's not like that. It is using these you know platforms for generating those skills so that we can run a lot of our stuff on App script or yeah within within our kind of control. Now some amount is inevitable because

Ritu Java 22:48

you know, there is the assessment of let's say we wanna look at a listing and we're trying to figure out what this listing is, but that is in the domain of public knowledge. Like it's already exposed on Amazon. So if if somebody is like scraping it or if a if an AI is actually looking at that, it could do it even without going through us. You know, so from from that point of view it's I would say it's fairly normal use. It's nothing extreme and and I I don't think

Ritu Java 23:14

These companies should be as paranoid as they are because otherwise they are getting left behind. I know that some companies have developed their own internal like bots, like a a GPT bot for themselves, but it's limited. It's got limitations and even their own employees are like, I don't wanna use this.

Hai Mag 23:32

I mean I think at some point there will be

Ritu Java 23:35

Yeah.

Hai Mag 23:35

systems that people will run locally in their cloud and it is happening. I'm seeing like it's it's happening,

Ritu Java 23:38

Locally Yes. Yeah, for sure.

Hai Mag 23:43

but it's not we are not there yet. I mean we already talked about this wonderful connection between the operations and how to run you know, Amazon and operations and replacing SOPs. So our you know, I think you know that's all great.

Hai Mag 23:59

as a final question, maybe, so you did that, you are doing it for the last year. what type of an outcome wa you had as a result of leveraging the skills now compared to the old style of doing it with SOPs? Was it more you reduce the number of people or you improved better results? Or what is the outcomes really?

Ritu Java 24:23

Yeah, absolutely. So for us, we're not letting go of any of our people. That was one of the decisions I made early on that because of AI, there's gonna be a lot of disruption in the world and I don't want to lose my team. Like my team is precious to me, so I'm not gonna fire anyone for AI. but I will use AI to help them do better. So for me, the biggest leverage is quality, like the quality of our

Ritu Java 24:45

data, the quality of our value to the customer, the quality of you know how we can reduce costs to for ourselves as well as for our customers. So those are the areas in which AI has actually helped us in improving our efficiencies, I would say. So helping the the bottom line for ourselves as well as for our clients, that's been the biggest leverage. In terms of people, we think that with our current team we can actually grow the the number of clients without having to

Ritu Java 25:11

hire a lot more. Like previously there was a hiring frenzy every couple of months it would like, like we need to hire more, hire more, hire more. But now it's like with our current team, we're not letting them go, but we will, you know, have them be able to do a lot more.

Hai Mag 25:23

That's absolutely great to keep the team and I really enjoyed this conversation, especially when it's about AI. I'm always excited. And I think

Ritu Java 25:29

Yes, me too. Yeah.

Hai Mag 25:33

we'll just continue. Maybe I'll just c you know call you again in a few months and we'll you know kind of continue to discuss AI.

Ritu Java 25:38

Yeah, sure. likewise.

Hai Mag 25:41

again for

Ritu Java 25:41

Yes.

Hai Mag 25:42

the audience, if you guys be if anybody w will be in Seattle for Amazon Accelerate, it seems both of us are speaking there about AI, connected commerce, how to do operations, advertising better. Please come and watch us and meet with us. And thank you very much, Ritu for being my guest today.

Ritu Java 26:01

Thank you so much, hi. This was great. Awesome. We'll see you around.

Hai Mag 26:04

Thank you.

Hai Mag 26:05

Bye.