00:00.031 –> 00:08.159
[SPEAKER_00]: We have a huge benefit of being small enough so we can roll out changes, experiment, fail, try again, fail, try again, win.

00:08.459 –> 00:24.475
[SPEAKER_00]: And this is what the formula of the companies, I make sure that people have time to dedicate to try stupid things because you only learn by trying things with the eye so that they can actually use it when the hard time comes in and the hard question comes in.

00:27.350 –> 00:29.731
[SPEAKER_02]: It’s the e-commerce master plan podcast.

00:30.271 –> 00:34.352
[SPEAKER_02]: Here to help you solve your marketing problems and grow your e-commerce business.

00:34.872 –> 00:40.414
[SPEAKER_02]: Cutting through the hide-off to bring you inspiration and advice from the e-commerce sector and beyond.

00:40.654 –> 00:42.655
[SPEAKER_02]: Here’s your host, Chloe Thomas.

00:46.025 –> 00:47.106
[SPEAKER_01]: Hello, and welcome.

00:47.146 –> 00:48.187
[SPEAKER_01]: It’s great to have you here.

00:48.247 –> 00:52.210
[SPEAKER_01]: Thank you for hitting play and choosing to listen to one of our inspiring guests.

00:52.671 –> 00:57.094
[SPEAKER_01]: In this episode, we’re catching up with Roman for the second time this year.

00:57.114 –> 01:03.499
[SPEAKER_01]: Yeah, we don’t do that often, but when I chatted to Roman for the last episode when he was on, he had so many projects going on.

01:03.519 –> 01:10.705
[SPEAKER_01]: I’m just saying such fascinating work that I thought it was worth getting him back on the show for an update and just find out what he’s working on, really.

01:11.626 –> 01:23.815
[SPEAKER_01]: What that transcribed, that turned into is one massive chat about how they are day-to-day using AI tools within the business car.

01:24.316 –> 01:33.863
[SPEAKER_01]: Across the teams, how they are embedding it across the teams, and we get a little bit philosophical about some of the future of the AI tools, but he also shares some really interesting

01:34.603 –> 01:38.368
[SPEAKER_01]: impacts of what it’s doing to people’s workload and enjoyment of work as well.

01:38.929 –> 01:47.641
[SPEAKER_01]: So lots of cool stuff in this episode, a few tablets and two, and his top tips answers are awesome as well, so make sure you stick around for those.

01:53.859 –> 01:55.621
[SPEAKER_01]: And now to introduce our special guest.

01:56.061 –> 02:00.725
[SPEAKER_01]: Roman Petrochenkov is the head of marketing, a media analytics at CarWow.

02:01.145 –> 02:08.992
[SPEAKER_01]: CarWow is a one-stop shop for changing your car, offering customers the ability to buy a new or used car or sell their existing vehicle.

02:09.472 –> 02:24.979
[SPEAKER_01]: Founded in 2013, they now operate in the UK, Germany and Spain, and in 2024, they grew 50% with nearly £3 billion worth of cars bought on the site, and £1.8 billion of cars listed for sale through the self-micar service.

02:25.339 –> 02:26.100
[SPEAKER_01]: Hello Roman.

02:26.600 –> 02:28.501
[SPEAKER_00]: Hello Chloe, very happy to be back.

02:29.101 –> 02:30.662
[SPEAKER_01]: Awesome to have you back on the show.

02:30.762 –> 02:31.522
[SPEAKER_01]: How have you been?

02:31.582 –> 02:33.203
[SPEAKER_01]: What’s been going on for the last year?

02:34.258 –> 02:38.201
[SPEAKER_00]: Well, AI is changing a lot of things, so we have to have to a lot to that.

02:38.781 –> 02:51.910
[SPEAKER_00]: And I think what is the craziest and the most complex thing is that if we used to have 10 things we’re working now, we have 50 things that we’re working on and all of them have to be worked on because you have multiple AI chats in behind that helping it do that.

02:52.874 –> 03:02.136
[SPEAKER_01]: I feel like, I mean, last time you were on, we got into AI, but I feel like we did the mirror scraping of the surface of AI.

03:02.156 –> 03:03.417
[SPEAKER_01]: You know, it was like, AI exists.

03:03.577 –> 03:07.978
[SPEAKER_01]: It’s quite useful for putting together proposals to massively summarize what we talked about last time.

03:08.418 –> 03:12.559
[SPEAKER_01]: So it doesn’t surprise me at all, the AI is right up there.

03:12.579 –> 03:21.921
[SPEAKER_01]: Have you, would you say car wow is ahead of the game on AI adoption or middle of the pack or behind?

03:23.172 –> 03:24.833
[SPEAKER_00]: I’m not going to say we’ll be behind.

03:24.873 –> 03:28.776
[SPEAKER_00]: We definitely, in the top, maybe 10, 20% of a car or it.

03:29.377 –> 03:37.502
[SPEAKER_00]: We have a huge benefit of being small enough so we can roll out changes, experiment, fail, try again, fail, try again, win.

03:37.783 –> 03:39.984
[SPEAKER_00]: And this is what the formula of the companies.

03:40.685 –> 03:46.169
[SPEAKER_00]: And especially in my team, like, I make sure that people have time to dedicate

03:50.992 –> 03:59.014
[SPEAKER_00]: because when you learn, you only learn by trying things with the I so that they can actually use it when the hard time comes in and the hard question comes in.

03:59.774 –> 04:05.935
[SPEAKER_01]: Got you, and so active, a bit like Google’s old 10% of your time to be spent on other projects.

04:05.975 –> 04:12.397
[SPEAKER_01]: You’ve got like choose the afternoon or something, go and do something silly with AI just to see if you can get something from end to end.

04:12.437 –> 04:13.057
[SPEAKER_01]: That’s sort of thing.

04:13.590 –> 04:21.674
[SPEAKER_00]: I think company white will have different practices, but in generally the whole company approaches we want to get people to try and experiment as much as possible.

04:22.294 –> 04:42.485
[SPEAKER_00]: But then when we realize if you just ask people to do something, like out of the outside of context anything, people find a choice paralysis, like, okay, what I’m supposed to do with this, you’re looking at this black box, you turn it left, it’s black, it turn it right, it’s still black, but you need to give a few examples, and then it comes down to the teams,

04:43.205 –> 04:56.030
[SPEAKER_00]: Sometimes it takes more than 10% of the time, but I find the most is you need like a couple of weeks of a lot of heavy usage and then you can down be down to like 5-10% of your time and experimenting.

04:56.450 –> 04:57.091
[SPEAKER_01]: Got you?

04:57.291 –> 05:03.033
[SPEAKER_01]: It’s finding those reasons to use it heavily for a couple of weeks to then get yourself up to speed.

05:03.493 –> 05:04.594
[SPEAKER_01]: And you say it’s

05:05.984 –> 05:07.885
[SPEAKER_01]: different parts of the company have got different approaches.

05:07.925 –> 05:14.368
[SPEAKER_01]: So can I ask which department is leading the push into AI?

05:14.628 –> 05:22.472
[SPEAKER_01]: You know, because I imagine in many companies, it’s like a marketing got really into it or the website team got really into it and then everyone else is following on behind.

05:22.533 –> 05:23.893
[SPEAKER_01]: I doubt it’s ever financed.

05:24.233 –> 05:31.117
[SPEAKER_01]: But is there one team kind of leading for you or are there several of you pushing the boundaries a bit more than others?

05:31.792 –> 05:33.073
[SPEAKER_00]: It’s such a good question.

05:33.253 –> 05:38.996
[SPEAKER_00]: Obviously, not affiliated, but there’s a head of marketing and it is going to say analytics is actually very much ahead.

05:39.716 –> 05:55.284
[SPEAKER_00]: What we have done is, with some point, we realize that the only way to push adoption is to sit down with people, like person by person, see what specific person would problems they have and then show them how they can solve those problems.

05:56.004 –> 05:58.686
[SPEAKER_00]: And though you can try to write like a huge instruction

06:01.307 –> 06:16.862
[SPEAKER_00]: When you have a small enough team, the most efficient way to get people in is to just sit with them, spend a couple of hours, and then two hours down the line, you see people’s eyes, just dig to, they’re like, I want a more of this, and then you immediately start getting like a person that’s human, it’s spreads and spreads.

06:17.483 –> 06:21.910
[SPEAKER_00]: To be fair, in caros specifically finances, definitely number two team in adoption.

06:22.210 –> 06:25.516
[SPEAKER_00]: And they’re driving adoption across the company a lot.

06:26.096 –> 06:28.781
[SPEAKER_00]: And they’re very much sharing examples on how they use it.

06:29.322 –> 06:31.946
[SPEAKER_00]: I’m actually very, very proud for the team we have here.

06:32.631 –> 06:51.344
[SPEAKER_01]: I suppose there’s some element of AI works best where you’ve got hard and fast facts to feed into it and in finance there should be hard and fast facts to, you know, definite black and white territory of, you know, this is a yes, this is a no and so on and so forth.

06:51.444 –> 06:52.005
[SPEAKER_01]: And actually,

06:53.191 –> 07:04.637
[SPEAKER_01]: for all, you know, let’s not get into the attribution debate again, but apart from how do you attribute the actual nuts and bolts and the data coming into the world of analytics is also pretty steadfast isn’t it?

07:04.657 –> 07:13.021
[SPEAKER_01]: So it would make sense that those two teams are able to take it a bit further because you’re dealing with more facts than maybe the copyrights are dealing with.

07:13.883 –> 07:18.567
[SPEAKER_00]: I’m sent, I would say, there are two parts for me that work a lot.

07:18.987 –> 07:20.168
[SPEAKER_00]: One is the harness.

07:20.769 –> 07:34.380
[SPEAKER_00]: So if all the AI tooling that you have in front of you is just chat, so you can do a lot of chatting, but then you try to ask it to go and change your Google Ads campaigns and it’s like, well, it just cannot, it’s a chat.

07:34.400 –> 07:35.881
[SPEAKER_00]: So I need to do some MCP in it.

07:35.961 –> 07:37.883
[SPEAKER_00]: It becomes really technically really complex.

07:38.463 –> 07:44.428
[SPEAKER_00]: But let’s imagine you have a chapter that is specifically designed for Google like boom, it has a harness, understanding how it works.

07:45.129 –> 07:50.054
[SPEAKER_00]: And that is, for example, for finance, cloud provides Microsoft Excel connection.

07:50.354 –> 07:52.576
[SPEAKER_00]: So you can just ask it to do your pivot table.

07:53.277 –> 08:02.165
[SPEAKER_00]: And though most people in finance know how to do the pivot table, everybody in e-commerce know how annoying it is sometimes to make specific filters with specific semifz.

08:02.685 –> 08:08.910
[SPEAKER_00]: And this is where Claude is amazing, but it just takes all of that complexity out of the question.

08:09.771 –> 08:13.374
[SPEAKER_00]: And the second thing is your logical thinking.

08:13.434 –> 08:21.901
[SPEAKER_00]: So if the person is kind of wired to do, you know, like, logical, almost coding, like, experience.

08:22.581 –> 08:32.726
[SPEAKER_00]: They get it a little bit quicker, because they frame their questions in the way that the eye is just going to get their quicker, but it doesn’t mean that they’re smarter in any way.

08:32.786 –> 08:41.150
[SPEAKER_00]: It’s just copyright or immediately spots the eye slope text, and like, well, it doesn’t actually, it does mean the other thing that I need to do, it doesn’t replace me in any way.

08:41.410 –> 08:46.453
[SPEAKER_00]: There’s nobody’s going to argue that the club text is anywhere close to actual human being writing that.

08:47.053 –> 08:51.419
[SPEAKER_00]: But then you can use AI to research a topic, provide ideas, or summarize what other people do.

08:51.960 –> 08:56.206
[SPEAKER_00]: But that requires you to have a little bit of like how to automate my work.

08:56.226 –> 08:59.291
[SPEAKER_00]: And this is the technical capability, rather than incorporating capability.

09:00.134 –> 09:12.437
[SPEAKER_01]: You mentioned the AI tools there and I feel like this is a not a plateau point but a sudden load of work that many companies are going through with their AI tools.

09:12.597 –> 09:24.160
[SPEAKER_01]: Is it starts off, you know, Jenny and Accounts has got a log into Claude and Kevin in marketing has gone, has got an open AI account on the go and someone’s racking up some

09:30.341 –> 09:44.579
[SPEAKER_01]: And then there comes a point where you suddenly realize you’re duplicating all this work because everyone is working in different tools without one kind of hub and then someone is using a tool where all the information gets fed back into the central LLM and then oh god nightmare.

09:45.159 –> 09:47.542
[SPEAKER_01]: So I feel like everyone’s heading towards a point of

09:48.443 –> 09:52.970
[SPEAKER_01]: AI tool, consolidation, and corporatization, dare one say?

09:54.132 –> 09:57.577
[SPEAKER_01]: How are you handling that within CarWow?

09:57.677 –> 10:03.586
[SPEAKER_01]: Is it as everyone logging into the same Claude portal or are there different ones set up for different teams?

10:04.449 –> 10:07.150
[SPEAKER_00]: We have several of them set up for different teams.

10:08.071 –> 10:10.952
[SPEAKER_00]: We have shared skills and shared context with people to use.

10:11.492 –> 10:18.736
[SPEAKER_00]: But I think the problem with AI tooling from where I said is that people think of AI as a tool that does a thing.

10:19.036 –> 10:20.897
[SPEAKER_00]: For example, everybody knows Photoshop.

10:21.117 –> 10:27.867
[SPEAKER_00]: it’s a tool to change images and there’s like a tool like final card that changes in the video and stuff like this.

10:28.347 –> 10:36.719
[SPEAKER_00]: Whereas AI is a tool you want it to be and what my concern is like when you try to say well everybody in the company should use it the same way.

10:37.520 –> 10:43.041
[SPEAKER_00]: It comes just all because you can’t because people in different departments use it in absolutely different way.

10:43.461 –> 10:50.423
[SPEAKER_00]: It’s almost like a piece of metal that you can forge in any form and then the fighters will make swords.

10:50.863 –> 10:53.664
[SPEAKER_00]: And then engineers are going to make small CPU units.

10:54.064 –> 10:59.145
[SPEAKER_00]: And then I don’t know, people who are doing arts are going to make a really beautiful statues.

10:59.185 –> 11:00.925
[SPEAKER_00]: And like this is what AI is.

11:01.206 –> 11:04.646
[SPEAKER_00]: Depending on your purposes it will be absolutely unrelated to each other tools.

11:08.007 –> 11:10.648
[SPEAKER_00]: But this is where the real complexity comes in.

11:10.988 –> 11:14.749
[SPEAKER_00]: Do you try to solve this of everybody using it the same?

11:14.909 –> 11:17.450
[SPEAKER_00]: Or do you give everybody freedom but then how you control it?

11:17.870 –> 11:24.812
[SPEAKER_00]: And I feel this is like a constant debate internally which we don’t know which one is better because each one of them comes with the trade-offs.

11:25.685 –> 11:33.308
[SPEAKER_01]: Yeah, I feel like, um, with, like, so I can thinking too much there, sorry for the pauses, this is.

11:33.529 –> 11:43.313
[SPEAKER_01]: So, um, I completely get me coming from about different things being, different for, for, for, for the best tool for different teams, you know, your example of finance with their Excel spreadsheets.

11:43.353 –> 11:46.734
[SPEAKER_01]: Yeah, Claude is obviously going to be the best one to go for.

11:46.774 –> 11:52.297
[SPEAKER_01]: If marketing a deep in Google sheets, they’re probably going to want to go into Gemini and Nano and all of that stuff.

11:52.897 –> 12:01.419
[SPEAKER_01]: And I think there’s, I think Claude became a bit of a game changer with its folders and, you know, the different structure it put around things.

12:02.199 –> 12:08.921
[SPEAKER_01]: But I feel like it’s another stepping stone and in a couple of years time we’ll all be logging into something else.

12:09.441 –> 12:17.343
[SPEAKER_01]: Because there’s a lot of tools out there now that do that same document capture and folder thing that Claude does.

12:18.003 –> 12:20.024
[SPEAKER_01]: But enable you to access everything.

12:20.744 –> 12:23.086
[SPEAKER_01]: you know, all the models available in the marketplace.

12:23.126 –> 12:30.611
[SPEAKER_01]: And I can see that becoming more of a corporate apart from the fact that, you know, the Claude time to Microsoft.

12:30.631 –> 12:42.779
[SPEAKER_01]: But, yeah, I can see that more of a corporate play, we’ve got this hub where we can feed in Gemini and OpenAI and Claude and anything else that comes along in the next two years.

12:43.140 –> 12:44.000
[SPEAKER_01]: And we’ve immediately got

12:49.404 –> 13:03.291
[SPEAKER_01]: in some government somewhere that suddenly decides we can’t have access to it anymore to LLMX, then we’ve still got all the rest of our stuff in this third party piece that then we’ll we can then use it across other AI stuff.

13:03.791 –> 13:06.092
[SPEAKER_01]: Do you see that coming down the line as well?

13:06.192 –> 13:08.313
[SPEAKER_01]: Or am I am I in a little island of my own here?

13:08.373 –> 13:09.214
[SPEAKER_01]: And that thinking?

13:10.145 –> 13:12.548
[SPEAKER_00]: I think it’s very much depends on the company size.

13:12.808 –> 13:30.626
[SPEAKER_00]: Like big companies will try to take a control over it because it exponentially grows and it creates so much noise when it’s uncontrolled that it becomes more expensive to take care of the noise rather than to actually, like you don’t have enough incremental benefit to cover for a month of noise that people with AI.

13:31.587 –> 13:35.049
[SPEAKER_00]: For the smaller companies, I think it will come down to a different thing.

13:35.569 –> 13:43.453
[SPEAKER_00]: I don’t know how about everybody else, but as of today, I don’t really feel much of a difference between models, like they’re all roughly do the same thing.

13:43.954 –> 13:50.297
[SPEAKER_00]: Obviously there’s not a banana, which is amazing for images and Claude’s image generation is really really far away from this.

13:50.697 –> 13:55.220
[SPEAKER_00]: But if I’m writing some kind of like research proposal, I can do different things.

13:55.260 –> 13:55.800
[SPEAKER_00]: It doesn’t really

13:57.901 –> 14:00.042
[SPEAKER_00]: What really changes it is the harness.

14:00.602 –> 14:04.264
[SPEAKER_00]: Like sharing skills and cloth is just superior to Gemini.

14:04.444 –> 14:05.745
[SPEAKER_00]: And that’s why it’s so big.

14:05.765 –> 14:07.366
[SPEAKER_00]: And that’s why everybody talks about it.

14:08.046 –> 14:12.128
[SPEAKER_00]: And the real question is, is harnessing going to all equalize?

14:12.168 –> 14:14.669
[SPEAKER_00]: And you can do all of these things across all of the models?

14:15.149 –> 14:20.512
[SPEAKER_00]: And if that’s true, the question will come down to cost, who can actually run the model at the cheapest level?

14:20.852 –> 14:22.493
[SPEAKER_00]: Because right now they all subsidize.

14:22.533 –> 14:26.575
[SPEAKER_00]: And if you all will pray, we’ll pay the amount of money that actually costs.

14:26.875 –> 14:31.140
[SPEAKER_00]: And then it also would be using, yeah, all of us will just say, well, it’s a thing from 20 years from now.

14:31.160 –> 14:34.984
[SPEAKER_00]: Like, you know, it’s going to take so much time for it to actually be normal, like flying cars.

14:35.284 –> 14:39.929
[SPEAKER_00]: They exist, but about five of them, you know, but we don’t know what’s going to happen with a cost factor.

14:39.949 –> 14:44.234
[SPEAKER_00]: But I think harness around it, I predict is going to be all equalized.

14:44.494 –> 14:47.117
[SPEAKER_00]: You can take a lot of skill put it in Gemini and be fine with it.

14:47.831 –> 15:05.493
[SPEAKER_01]: Yeah, I think you’re probably right there because, and the point you make about the cost of it is from an ecological and sustainability standpoint, I would like those costs to come in sooner or later because we are burning through an awful lot of the earth resources for no good reason at all, you know, creating.

15:06.254 –> 15:14.599
[SPEAKER_01]: And sorry, no one who likes doing this, but creating endless pictures of your friends looking like dogs is not a good use of the world’s energy and resources.

15:15.260 –> 15:18.782
[SPEAKER_01]: But if we had to pay for that, then we would use it more sparingly.

15:19.342 –> 15:20.403
[SPEAKER_01]: Get off that soapbox, okay?

15:21.144 –> 15:31.871
[SPEAKER_01]: But I think, you know, as business is we are, we don’t want to become too dependent on it because you could suddenly find yourself in a scenario when the prices do start going up to where they should be.

15:32.811 –> 15:49.607
[SPEAKER_01]: economically if nothing else, where suddenly our costs get a lot higher, and it’s I can’t help but think that for the ease of life of all of us and less of the yo-yo effects, then it would be good if if that happens sooner rather than later, but maybe I’m just being no commudging with that.

15:49.947 –> 15:50.908
[SPEAKER_00]: I honestly don’t know.

15:51.228 –> 16:03.177
[SPEAKER_00]: I think what can happen is that the AI cost will actually not going to get that much lower and it’s actually going to climb up and that is also the removal of this image generation and it will basically focus on a specific thing.

16:03.637 –> 16:06.680
[SPEAKER_00]: It’s more natural to assume that it actually will go come down.

16:07.100 –> 16:15.106
[SPEAKER_00]: But I always need to remind myself that back in the days everybody invested in Uber because the driverless taxes were just one two years away.

16:15.626 –> 16:25.469
[SPEAKER_00]: And now we’re like 15 years into this conversation is still not quite at the point that Uber had to switch completely to this strategy and being like one, I have to provide a lot of different things.

16:26.010 –> 16:32.912
[SPEAKER_00]: So how many years are there until AI will be at the point where it is cheap, affordable and efficient to run?

16:33.492 –> 16:41.695
[SPEAKER_00]: I honestly not experienced enough to know that, but I definitely know that at the current price if you’re small and medium sized business,

16:42.335 –> 16:46.459
[SPEAKER_00]: you can build so many things for basically their cheap.

16:46.719 –> 16:50.242
[SPEAKER_00]: You couldn’t even imagine to afford like five years ago.

16:50.943 –> 17:05.015
[SPEAKER_01]: So do you find that you and your team are mainly using AI to analyze data to write reports or to build technology that you couldn’t afford or couldn’t build before?

17:05.616 –> 17:07.837
[SPEAKER_01]: Coding words or number crunching, basically.

17:08.693 –> 17:20.360
[SPEAKER_00]: I think what I’m really kind of concerned, honestly, is that when I say, yes, I use it a lot for coding because for coding it’s amazing, and then I use it a lot for analysis because it’s also amazing.

17:20.680 –> 17:27.944
[SPEAKER_00]: That also comes with an extent that I have more than 10 years experience in both of them and I can verify every single input and output.

17:28.424 –> 17:34.168
[SPEAKER_00]: So maybe it’s just as good in everything, but I can’t verify output of everything else or prompt it right away.

17:35.028 –> 17:39.331
[SPEAKER_00]: but also it comes down from where the quality really matters and where the quality doesn’t matter.

17:39.811 –> 17:50.497
[SPEAKER_00]: If you need to get a number to have an estimation of how many app users you have, but you just want to have a distribution of which different marketplaces drive the conversions or like leads to your business.

17:51.137 –> 17:53.398
[SPEAKER_00]: The risk factor on that number is very low.

17:53.819 –> 17:55.780
[SPEAKER_00]: So even if you get it a bit wrong,

17:56.628 –> 17:58.070
[SPEAKER_00]: It’s not really that big of a deal.

17:58.110 –> 18:02.436
[SPEAKER_00]: As long as it’s not drastically wrong, but the chance for it to be drastically wrong is not really big.

18:03.217 –> 18:11.648
[SPEAKER_00]: Whereas if you’re creating an investor deck or the board level meeting and you’re using AI for that, you probably should reconsider that approach you taking, because

18:12.249 –> 18:14.690
[SPEAKER_00]: It’s very inconsistent that what it gives to you.

18:15.011 –> 18:20.894
[SPEAKER_00]: And this is what I have in my team and say, look, I want you to use AI to build solutions that are deterministic.

18:20.934 –> 18:25.656
[SPEAKER_00]: So you can use AI to write code because you can verify code and then code produces exactly the same result.

18:26.176 –> 18:30.759
[SPEAKER_00]: You can use AI for marketing, for example, to scrape information about your competitors.

18:31.439 –> 18:32.600
[SPEAKER_00]: And that is a low risk.

18:32.900 –> 18:40.004
[SPEAKER_00]: It’s always nice to know what they’re doing, but you should not use AI to drive your Google Ads campaigns because one day it’s going to be really expensive.

18:40.904 –> 18:42.129
[SPEAKER_00]: because it’s going to just something wrong.

18:43.159 –> 18:45.680
[SPEAKER_01]: Yeah, there is that context element, isn’t it?

18:45.720 –> 18:53.744
[SPEAKER_01]: He’s like, if you’re in a board meeting, or a marketing strategy a wayday, and some of you guys, oh, what’s our customer lifetime value?

18:53.764 –> 18:54.945
[SPEAKER_01]: Or what’s our attention number?

18:54.985 –> 18:57.286
[SPEAKER_01]: A quick and dirty answer is completely fine.

18:57.906 –> 19:07.931
[SPEAKER_01]: But if you’re then building out the budget model for five years based on a customer lifetime value from per channel, you need that to be, to the letter precise, don’t you?

19:08.091 –> 19:10.532
[SPEAKER_01]: Absolutely, absolutely spot on.

19:12.873 –> 19:18.436
[SPEAKER_01]: There’s no rule you can set for that, is there for which solution to use?

19:18.476 –> 19:21.177
[SPEAKER_01]: That’s just something you have to get through experience.

19:21.838 –> 19:26.100
[SPEAKER_00]: 100% and this is where the harness, which I’m talking about, comes into play.

19:26.880 –> 19:29.682
[SPEAKER_00]: Because what AI is really good is when you give it

19:31.283 –> 19:33.885
[SPEAKER_00]: It tasks to build you something that is not AI.

19:34.225 –> 19:42.652
[SPEAKER_00]: So for example, you want to build a dashboard or automation or some sequencing that then can work without your subscription or any tokens, whatever.

19:43.372 –> 19:44.013
[SPEAKER_00]: It’s amazing.

19:44.373 –> 19:53.320
[SPEAKER_00]: But currently, if you take for example marketing, there is not so many platforms that allow you to build, for example, a harness to generate consistent creatives because again,

19:53.900 –> 20:00.222
[SPEAKER_00]: It’s just not there, it’s not economics, it’s not proven to build it efficiently or like a good automation.

20:00.642 –> 20:01.943
[SPEAKER_00]: But there are some examples.

20:02.123 –> 20:08.865
[SPEAKER_00]: So for example, Google Ads Script is a simple cheap dirt automation that any business can run.

20:09.825 –> 20:11.746
[SPEAKER_00]: And it’s so small and it’s so annoying.

20:11.766 –> 20:13.767
[SPEAKER_00]: You need to know JavaScript, which is annoying.

20:14.168 –> 20:18.290
[SPEAKER_00]: You need to know how it works, which is not like the addition of the whole tool is just horrible.

20:18.870 –> 20:24.434
[SPEAKER_00]: And then if you go to Cloud Chat and ask it to generate your code, it will confuse everything around.

20:24.934 –> 20:29.597
[SPEAKER_00]: But you can give it enough, it’s good skill, that will basically give it all the elements.

20:29.677 –> 20:32.498
[SPEAKER_00]: And boom, you have a very cheap automation for your Google has campaigns.

20:32.558 –> 20:33.799
[SPEAKER_00]: If you need to know zero coding.

20:33.819 –> 20:35.580
[SPEAKER_00]: And this is exactly what I’m talking about, harness.

20:35.900 –> 20:38.402
[SPEAKER_00]: If you give it enough instruction, it can put it in there.

20:39.062 –> 20:53.114
[SPEAKER_00]: It’s amazing, but what you shouldn’t do, you shouldn’t actually run campaigns directly from MCP and say, I’ll just up my budget for a couple of campaigns and come to more, I was like, oh yes, I did it all campaign, sorry about that, I will revert it now, but you will have a million pound bill from Google Ads.

20:53.735 –> 21:06.485
[SPEAKER_01]: So it’s, it’s used the AI to create the code to run the non-AI programs like Google scripts that are just the non-thinking programs that just do the same thing, the same thing, the same thing.

21:07.106 –> 21:10.669
[SPEAKER_01]: Ro ask goes above x, they do y in the area.

21:10.709 –> 21:14.672
[SPEAKER_01]: So the kind of the reliable non-thinking, I’m saying yes.

21:16.053 –> 21:21.914
[SPEAKER_00]: Like another example is link it in, like I’m posting link it in Lately and trying to get some impressions and trying to get stuff.

21:22.415 –> 21:26.375
[SPEAKER_00]: My first interaction was obviously I’m just asked for to write a link it in post.

21:26.615 –> 21:35.458
[SPEAKER_00]: And then I realized that every post is exactly the same because the boarding it uses the average out caught like order of words is exactly the same.

21:35.878 –> 21:36.978
[SPEAKER_00]: So now I don’t do that.

21:37.098 –> 21:43.119
[SPEAKER_00]: I ask it to criticise me, maybe create a better hook, maybe give me some advices,

21:46.140 –> 21:46.400
[SPEAKER_00]: work.

21:46.460 –> 21:53.146
[SPEAKER_00]: When was such a two-manual human input with all of my human mistakes and all imperfections of my English?

21:53.967 –> 21:59.332
[SPEAKER_00]: I actually got significantly better outcome because people like reading human human-generated content.

21:59.986 –> 22:07.513
[SPEAKER_01]: I do just a tangent slightly away from me commerce, but the new report this post is AI Slop button on LinkedIn.

22:08.114 –> 22:16.341
[SPEAKER_01]: I would love to know whether they’re actually feeding that into the algorithm or whether it’s just to stop people messaging them and going, stop putting AI Slop following them.

22:16.742 –> 22:19.744
[SPEAKER_01]: Is it a PR campaign or is it actual data they’re capturing?

22:19.804 –> 22:21.206
[SPEAKER_01]: I suspect it’s both.

22:22.407 –> 22:22.547
[SPEAKER_01]: Right.

22:22.567 –> 22:23.847
[SPEAKER_01]: Let’s not go down that tangent.

22:24.168 –> 22:39.217
[SPEAKER_01]: One thing you said right at the beginning when I ask you how things are these days was that embracing AI heavily in your department means that rather than working on 10 projects at a time, you’re now working on 50 projects at a time.

22:40.998 –> 22:41.938
[SPEAKER_01]: Is that a good thing?

22:42.899 –> 22:44.120
[SPEAKER_00]: It’s a very good question.

22:45.306 –> 22:48.667
[SPEAKER_00]: But whether it’s a right question, it’s my question.

22:48.707 –> 22:49.708
[SPEAKER_00]: So let’s look at this one.

22:49.728 –> 22:51.149
[SPEAKER_00]: There’s a human perspective to that.

22:52.169 –> 22:53.730
[SPEAKER_00]: And we had a conversation internally.

22:53.890 –> 22:57.311
[SPEAKER_00]: And I really keep thinking about it more or more as every week passes by.

22:58.192 –> 22:59.652
[SPEAKER_00]: You used to come up with idea.

23:00.192 –> 23:01.153
[SPEAKER_00]: You come to the meeting.

23:01.293 –> 23:02.013
[SPEAKER_00]: You have a meeting.

23:02.073 –> 23:02.714
[SPEAKER_00]: You come down.

23:02.734 –> 23:03.434
[SPEAKER_00]: You have your coffee.

23:03.454 –> 23:04.434
[SPEAKER_00]: You sit down for two hours.

23:04.454 –> 23:05.495
[SPEAKER_00]: You execute that idea.

23:05.955 –> 23:07.316
[SPEAKER_00]: Then you go and try that idea.

23:07.396 –> 23:10.377
[SPEAKER_00]: So you have different activities during the day that you do.

23:11.097 –> 23:16.620
[SPEAKER_00]: Now, you have an idea, you go to the meeting, Cloud executes the idea, you come from the meeting, you start thinking about that output.

23:17.141 –> 23:30.348
[SPEAKER_00]: You have a new idea, you put it in, you go to the meeting, and there is no break, you just have seven, eight hours of exhaustive mental work that you do, because you constantly thinking, because execution is not part of your workflow anymore.

23:30.868 –> 23:33.790
[SPEAKER_00]: It’s like if you would be able to teleport to office,

23:35.251 –> 23:37.815
[SPEAKER_00]: And not going to the gym, there would be zero physical activity.

23:37.855 –> 23:47.829
[SPEAKER_00]: The same thing here, you brain just never switch off the thinking, because there’s always a new chat in your cloud that just stopped thinking, you came with the outcome that you now need to process.

23:48.550 –> 23:50.393
[SPEAKER_00]: So humans really start getting

23:51.134 –> 23:55.536
[SPEAKER_00]: tired more and more with AI because there’s just more things that you need to think about.

23:55.956 –> 24:04.079
[SPEAKER_00]: And from that perspective and perspective, it’s a sustainability because after all, we want humans to be happy, we want to Earth to be healthy.

24:04.099 –> 24:09.580
[SPEAKER_00]: Like we want, when we’re all doing it to live our life happily, that’s probably a negative outcome.

24:09.941 –> 24:15.963
[SPEAKER_00]: And this is something we would need to live with, the same way that social media is not necessarily the healthiest thing for us.

24:17.375 –> 24:30.979
[SPEAKER_00]: The company perspective, that actually can be, like, not this particular aspect of it, but being able to try things faster is good because then the amount of investments you need to put into trying something is now 10 times smaller.

24:31.659 –> 24:34.680
[SPEAKER_00]: You’ll need to hire an agency to make a lending page.

24:34.700 –> 24:37.800
[SPEAKER_00]: You can do lending page, sloppy, not very good.

24:38.261 –> 24:43.662
[SPEAKER_00]: Okay, it gives some traction, some traffic, then you give it to agency and then it becomes so much cheaper to try and test.

24:46.812 –> 24:51.576
[SPEAKER_02]: E-commerce master plan is supported by some of the greatest companies in the E-commerce sector.

24:51.716 –> 24:53.157
[SPEAKER_02]: Here’s a reminder of who they are.

25:04.497 –> 25:12.404
[SPEAKER_01]: Okay, I love this section because it gives me an honest answer so I’m really quick ideas for taking our businesses to the next level and a chance for all our brains to relax after.

25:12.424 –> 25:17.209
[SPEAKER_01]: I’ve quite an impressive chat about AI that certainly sent my head to go off in their multiple directions.

25:17.249 –> 25:19.331
[SPEAKER_01]: So Roman, are you ready for the top tips?

25:19.911 –> 25:20.371
[SPEAKER_00]: I’m certain.

25:20.572 –> 25:21.092
[SPEAKER_00]: Always ready.

25:21.472 –> 25:23.034
[SPEAKER_01]: Okay, booktop tip.

25:23.114 –> 25:29.600
[SPEAKER_01]: If everyone listening to this podcast agreed to take Friday off and read a book to make their business better, which book would you recommend?

25:30.427 –> 25:31.989
[SPEAKER_00]: I would recommend don’t make me sing.

25:32.330 –> 25:39.659
[SPEAKER_00]: It’s a 20 year old book, hasn’t changed anything, but then in the era where I became, I think a hundred times more relevant.

25:40.280 –> 25:47.209
[SPEAKER_00]: Make it simple to order and to work with your business and it’s just the only thing that literally the customers need.

25:47.869 –> 25:49.290
[SPEAKER_01]: an utter classic.

25:49.750 –> 26:00.458
[SPEAKER_01]: I have no problem with that being recommended many many times on this show, so I love that thank you Roman, traffic top tip, which marketing method do you either price above all others or think doesn’t get the press it deserves.

26:01.163 –> 26:08.707
[SPEAKER_00]: Last time I said influencers, this time I’m going to say, I’m going to again, very, very predictable and say, my latest laugh is better advertising.

26:09.027 –> 26:11.689
[SPEAKER_00]: I love going into all the details and information.

26:11.709 –> 26:15.611
[SPEAKER_00]: You can send that back to meta and get a really good outcome of meta.

26:16.391 –> 26:20.933
[SPEAKER_00]: But I’m going to have the honest amount of effort it takes to get the maximum out of it.

26:21.454 –> 26:23.755
[SPEAKER_00]: It’s just becoming more and more over the years.

26:24.095 –> 26:25.636
[SPEAKER_00]: The thoughts from become more complex.

26:27.100 –> 26:43.549
[SPEAKER_01]: It’s a brilliant thing that you can feed all this information back to Meta, but yeah, Roman, if you with all the sales volume you’ve got and the whole team dedicated to analytics and attribution to say it gets a bit much, it’s definitely one which smaller businesses don’t get lost down that rabbit hole, isn’t it?

26:43.569 –> 26:44.509
[SPEAKER_00]: I’m sorry.

26:44.749 –> 26:54.694
[SPEAKER_00]: I’m not even talking about how much time it takes to open ads page in Meta, it takes five seconds to load, and then you have all of this EI stuff going on and then every button is five seconds.

26:57.225 –> 26:58.506
[SPEAKER_01]: Yeah, say that one for next time.

26:59.146 –> 27:09.352
[SPEAKER_01]: The tool to maybe a collaboration tool, a social media plug in, a phone up or just a way of working, is the recall little tool you use that makes you and your team at more efficient from day to day.

27:10.138 –> 27:34.093
[SPEAKER_00]: I started using Claude design a lot, because I, it’s the thing that I like as an analyst, as a technical person, is to be able to organize my thought visually, and I don’t use it just to generate like banners or HTML files, which is absolutely amazing, but also it can generate anything, any mind map or thought or thinking process, just visualize my thoughts really quickly, and it’s my favorite tool over the last few weeks.

27:35.036 –> 27:36.780
[SPEAKER_01]: was not expecting that.

27:36.880 –> 27:38.002
[SPEAKER_01]: Love it, thank you.

27:38.824 –> 27:43.534
[SPEAKER_01]: Carbon top tip, what’s your favorite way to reduce the carbon footprint of an e-commerce store?

27:44.402 –> 27:50.383
[SPEAKER_00]: I always struggle with this one because my carbon thinking always like big things, let’s go to renewable energy.

27:51.243 –> 27:59.065
[SPEAKER_00]: I would probably say having less meetings, meaning that you will be more organized, means that you need to have less commutes, be doing things less.

27:59.705 –> 28:03.106
[SPEAKER_00]: Type things in notion or in docs share the period before meeting.

28:03.186 –> 28:06.066
[SPEAKER_00]: Let’s try to do it like 10 years ago when you were all prepared.

28:06.586 –> 28:08.747
[SPEAKER_00]: To the meeting, that would be amazing experience.

28:09.087 –> 28:11.727
[SPEAKER_00]: But as an outcome, people would spend a little way less time

28:15.129 –> 28:17.111
[SPEAKER_01]: Yeah, people unprepared to meetings.

28:17.331 –> 28:18.652
[SPEAKER_01]: Hey, what are we talking about today?

28:18.933 –> 28:21.595
[SPEAKER_01]: Not heart sinks when that starts.

28:21.615 –> 28:24.198
[SPEAKER_01]: So yeah, get just use your time better.

28:24.478 –> 28:24.939
[SPEAKER_01]: Love that.

28:25.199 –> 28:25.820
[SPEAKER_01]: Thank you, Roman.

28:25.840 –> 28:30.965
[SPEAKER_01]: But before we say goodbye, could you please let the listeners know where they can find you on the web and social media?

28:31.497 –> 28:33.378
[SPEAKER_00]: The best way to find me is on my LinkedIn.

28:33.438 –> 28:34.739
[SPEAKER_00]: It’s a Roman battery chunk of.

28:35.540 –> 28:38.622
[SPEAKER_00]: You can reach out to me always if you have any questions or suggestions.

28:39.042 –> 28:44.145
[SPEAKER_00]: I post multiple times a week, sometimes three times a week when I’m very consistent with my diary.

28:44.986 –> 28:48.468
[SPEAKER_00]: And besides that, if you have any project, you want to work with partner on.

28:48.548 –> 28:50.549
[SPEAKER_00]: I’m always open for any consultancy ideas.

28:51.030 –> 28:54.452
[SPEAKER_01]: OK, because you’ve been doing some consultancy recently with smaller businesses, haven’t you?

28:55.040 –> 28:55.820
[SPEAKER_00]: Yes, 100%.

28:56.000 –> 29:03.262
[SPEAKER_00]: It’s something that keeps me really excited because when you work with something that car wow, it’s very, very interesting, but some ideas takes a lot of time to execute.

29:03.682 –> 29:06.063
[SPEAKER_00]: And when you work with small businesses it’s the other way around.

29:06.083 –> 29:09.644
[SPEAKER_00]: You might not have resources, but you can try and experiment way faster.

29:10.444 –> 29:10.844
[SPEAKER_01]: Awesome.

29:10.884 –> 29:13.745
[SPEAKER_01]: Anyone who wants to reach out to Ramin’ on that, just find him on LinkedIn.

29:14.365 –> 29:17.346
[SPEAKER_01]: Ramin’, thank you so much for coming on the Commerce Masterland podcast.

29:17.366 –> 29:18.986
[SPEAKER_01]: It’s been great catching up with you again.

29:19.046 –> 29:22.287
[SPEAKER_01]: And thanks for giving us the ins and outs of the AI journey.

29:22.327 –> 29:23.187
[SPEAKER_01]: So thanks for being here.

29:23.877 –> 29:24.838
[SPEAKER_00]: Thank you very much, Clay.

29:30.901 –> 29:33.602
[SPEAKER_01]: So much to think about there from Raymond.

29:33.622 –> 29:41.066
[SPEAKER_01]: And I thought that obviously lots of super practical clever advice on the big picture of getting AI to work within the business.

29:41.466 –> 29:50.892
[SPEAKER_01]: But I thought that final thing he was saying about how shifting from 10 to 50 things, because you’ve got those chats running all the time, is not giving you any downtime,

29:53.893 –> 30:00.500
[SPEAKER_01]: leading to the team getting more exhausted more quickly, which of course exhaustion is the enemy of creativity.

30:01.060 –> 30:02.101
[SPEAKER_01]: It’s quite fascinating.

30:02.141 –> 30:03.142
[SPEAKER_01]: So, one to watch out for.

30:03.162 –> 30:04.343
[SPEAKER_01]: I’m not quite sure how we fix that.

30:04.363 –> 30:06.085
[SPEAKER_01]: Do we force everyone to take a lunch break?

30:07.186 –> 30:08.167
[SPEAKER_01]: Maybe that would help.

30:08.667 –> 30:16.134
[SPEAKER_01]: You can get your hands on all our notes from this episode, including those top tips and links to what we mentioned by heading over to ecommercemasplan.com.

30:16.795 –> 30:22.784
[SPEAKER_01]: You can also use ECMP.info forward slash this episode number to get straight to the right page on the site.

30:22.964 –> 30:31.196
[SPEAKER_01]: When you get to the website, scroll down to the bottom to the big banner across the footer and add yourself to our email list so we can help you with other things too.

30:31.997 –> 30:47.605
[SPEAKER_01]: If you like this episode, then make sure you check out Episode 579, which is our first chat with Roman, where we got, we did a little bit of AI, but we mainly talked about consent, attribution, tracking, and all those kind of more analytics things.

30:48.786 –> 30:53.489
[SPEAKER_01]: Thank you very much for tuning into this and every episode that you do of the e-commerce master plan podcast.

30:53.569 –> 31:03.975
[SPEAKER_01]: I bring you a new interview every week to help you inspire and build your business, including encouraging businesses and consumers to make more sustainable buying decisions.

31:04.095 –> 31:09.018
[SPEAKER_01]: So if you know someone this show can help please tell them to listen to the e-commerce master plan podcast.

31:09.538 –> 31:12.300
[SPEAKER_01]: I hope you have a great week and don’t forget to keep optimizing.

31:15.413 –> 31:18.736
[SPEAKER_02]: Thank you for listening to the e-commerce Master Plan podcast.

31:19.177 –> 31:24.062
[SPEAKER_02]: Find out more at e-commercemasterplanned.com slash podcast.