Latest / Inside Commerce: Ecommerce Strategy, CX and Technology Podcast / EP342: The 70% Rule - How AI Workflow Automation Drives Massive Productivity & Quality Gains, With Tom&Co
Transcript
- 0:00One of the really interesting side effects of of AI is it
- 0:03really thinks hard about everything.
- 0:05People aren't so good at doing that.
- 0:06If you're doing a simple task, people tend to switch into
- 0:09simple thinking mode. And we do it quickly, don't
- 0:12apply much thought. But it's not like that.
- 0:13It's never lazy about the task it's doing.
- 0:16In fact, it's quite the opposite.
- 0:18It overthinks most things. But actually this has a.
- 0:21This has a really nice side effect, which is while it's
- 0:24doing a simple task, it often finds something or flags
- 0:27something to you that doesn't look right.
- 0:29And this helps to improve quality.
- 0:31It means that less mistakes actually happen.
- 0:35You listen to Insidecommerce, a weekly ecom strategy, technology
- 0:38and CX focus podcast hosted by experienced digital consultants
- 0:41and re platform specialist James Get and Paul Rogers.
- 0:44Get insights into ecom strategy, leading brands, hear from
- 0:46industry thought leaders and leading tech vendors with
- 0:49practical advice to help you improve your online CX.
- 0:54This episode is sponsored by World pay.
- 0:56Coming back to a. Another interesting topic around
- 1:00AI, but the aim of today is to steer clear of what is some of
- 1:04the ongoing hype in the market and focus very succinctly on the
- 1:08practical ways that senior engineering teams are factoring
- 1:12AI workflows into their own and client processes to drive better
- 1:15efficiency, smarter outputs, and also help people focus on doing
- 1:18the fun stuff and not have to do laborious repeat tasks that no
- 1:21one wants to do in life or sorry, a very small group of
- 1:25people might want to do, but most of us don't want to do.
- 1:27So I've got two really, really experienced and wonderful people
- 1:30I've known for many years worked on several projects.
- 1:32They come out of the Adobe ecosystem, run one of the
- 1:34leading agencies in that space, Tom and Co.
- 1:37So first of all, I've got Tom, who's the founder and director.
- 1:39Hey Tom, how you doing? I'm very well, thanks.
- 1:41How are you? I'm good.
- 1:43It's been too long actually. We've said it's been way too
- 1:45long since we've had you on the podcast.
- 1:46I'm really looking forward to it.
- 1:48And also Tom has the second most played episode of all time on
- 1:53Inside Commerce. I think it's like over 8800 or
- 1:56900 plays over over the time so far.
- 2:00So you're in the Hall of Fame. I'm honoured, I'm very honoured.
- 2:04Yes, and thanks for coming back. And then also we have Jakob
- 2:06Halva, who's the Head of OPS and Lead Business Analyst at Tom.
- 2:08We go. Hey, Jakob.
- 2:10Hey James, glad to be here. Yeah, I was saying before we
- 2:13start recording that it's criminal that I haven't had you
- 2:15on before because every time I've worked with you, I've
- 2:17learnt insane loads of things. And my interest was piqued by a
- 2:21post that Yak had put on LinkedIn recently about AI
- 2:24workflows that they're building using Clawed.
- 2:26And I just thought it's a really cool topic to get into because
- 2:29this isn't about platform, this is about engineering skills,
- 2:33automation and driving efficiency to make people's
- 2:36lives easier, which obviously is centric to the platform as you
- 2:40work on. But it can, it's agnostic.
- 2:41It's not limited to. So that's my kind of preamble
- 2:45and ramble about what we're going to talk about.
- 2:47Does that sound like a like a good enough introduction gents?
- 2:50Absolutely, yeah. Well, before we get into the
- 2:53questions, just for those of you who've not come across Tom can't
- 2:58speak Tom and Co before Tom, do you want to give a bit of a
- 3:02flavour for for what you guys do?
- 3:03And then, yeah, I could maybe you can add on to that your role
- 3:06and how that's been evolving. Yeah, of course.
- 3:09Yeah. I'd say Tumko has been around
- 3:12for what feels like forever now. So we set up in 2011.
- 3:18So we're sort of 15 years into this journey.
- 3:20And our focus at the very beginning was on the world of of
- 3:24Magento and e-commerce and has obviously changed over the years
- 3:28into Magento tier and Adobe Adobe Commerce.
- 3:31And so we do a lot of Adobe Commerce implementations these
- 3:34days across about 2525 clients. And yeah, as a business, we sort
- 3:40of focus on how we can help clients of course bring their
- 3:46e-commerce vision to life, but also how they can grow as well.
- 3:49So we have a, a strategy team as well as a development team to,
- 3:53to hope. Sorry to help on both sides of
- 3:56of the of the fence. And yeah, so just a a quick bit
- 4:01of background for me. So I originally started life as
- 4:05a developer. So I'm a very technical minded
- 4:07person. And since I've been running the
- 4:11business for the last 15 years, we were we were focused on
- 4:15operation, client relationships and making sure that we are
- 4:22there to support our clients as partners.
- 4:27So I've been at Tominco for about 10 years.
- 4:29Before that I was kind side e-commerce and over those 10
- 4:33years I think I've scoped out and re re platformed about 50
- 4:37websites. So I've spoken to a lot of
- 4:40e-commerce teams who had challenges and businesses who
- 4:43had problems with their processes.
- 4:46And I think 3 years ago when AI started, what was really
- 4:49exciting for me and, and why I'm here to talk about this topic is
- 4:52that I realise that AI is a novel solution which will
- 4:56finally allow me to help those teams to solve problems in a
- 4:59slightly different way. So I've always been very, I'm
- 5:02techy like Tom, never been a developer, but I've always been
- 5:06interested in how to make businesses more efficient.
- 5:09And it's really exciting that over the last 2-3 years, like
- 5:12the the means we have are massively bigger than before.
- 5:17So really keen to share some of the stuff we've been doing, some
- 5:20of the stuff we've been doing for our clients and hopefully
- 5:22might be helpful to some people. Yeah, definitely will be.
- 5:25I know from having worked with you guys on a few projects, I
- 5:28mean it's, I know it's been a couple of years now, but what I
- 5:32love is that you enjoy a challenge that you properly
- 5:36enjoy. If if a client or an annoying
- 5:38consultant like me push you back and goes, that's not going to
- 5:40work. And This is why.
- 5:41And actually we need something that's going to do AB and C.
- 5:44You guys love that because you love to be challenged to come up
- 5:47and then then come back and go there you go.
- 5:49There's a solution. And I think this for those of
- 5:51this, and that's why I wanted Jacob and Tom on take is where A
- 5:54is a is heading is rapid, but it's about really understanding
- 5:58how you can properly use it to deliver tangible business
- 6:01outcomes, not just put something in place for the sake of saying
- 6:04we've got an AI strategy. So let's Tom, let's do a bit of
- 6:08foundation of where you guys are seeing the market 1st and we'll
- 6:10come to Jacob about that post he put on LinkedIn recently.
- 6:12But so how are you seeing Tomico?
- 6:15How are you seeing the tech ecosystem evolve?
- 6:18How is AI changing the approach to things like re platform or
- 6:22building out tech stacks and getting value from existing tech
- 6:25stacks? Yeah, so, well, I mean, it's
- 6:29changing it a huge amount. It's a significant shift, the
- 6:32biggest one I've ever seen since I've been working in the
- 6:36industry. I think, you know, from an
- 6:39internal perspective as an agency and as a company that is
- 6:43responsible for writing code, of course we are seeing the the
- 6:48change there significantly. So our development teams are
- 6:53writing less code and AI is writing more code.
- 6:58And so we've had a really strong focus on trying to utilise that
- 7:02as much as we can and move our developers from from coders to
- 7:08to architects, people who understand the system, the
- 7:11system design, the interactions between the various systems to a
- 7:15really deep level because they need to spend a lot less time
- 7:20writing code these days. AI has, it's close to solve
- 7:24that. My view is that it's pretty much
- 7:27solved, and it certainly will be solved before the end of the
- 7:29year, just the speed of which AI is is changing.
- 7:32I have a very, very keen interest in AII think anyone
- 7:35that knows me is very much aware of that because it's the one
- 7:38topic I always want to talk about.
- 7:40And I've watched the the evolution of AIO for the past
- 7:46three years or so go from a fairly unhelpful system that
- 7:53makes a huge number of mistakes to one that now is capable of
- 7:58getting things correct in the first shot very often.
- 8:02So from our internal view, it's absolutely imperative that we
- 8:07ensure that our teams know how to use AI throughout their
- 8:11throughout their jobs. And that is not just code that
- 8:16is that is all levels. And I think we'll talk about
- 8:18that more as through the podcast, but you know, big,
- 8:22obviously a big part of it for us is, is code and the
- 8:25efficiency of writing code. But if you look at it from the
- 8:28brand's perspective, obviously, you know, then they're not,
- 8:32they're necessarily writing much code at the moment.
- 8:35I think they'll probably start doing more of that in terms of
- 8:38scripting and, and automations and stuff.
- 8:41But easier areas for them to enter into using AI in, in, in
- 8:47re platforms or just in their day-to-day is things like data.
- 8:51So AI has been incredible at understanding large data sets
- 8:58and manipulating large data sets.
- 9:00So a practical example of that is product data.
- 9:02I mean, one of the biggest challenges we always see on the
- 9:05client side with a RE platform is getting product data into
- 9:09order. So while we're slaving over code
- 9:11and trying to get functionality to work, they're slaving over
- 9:15product data and trying to get it into a good into a good
- 9:18shape. And so if you are, you know,
- 9:21typical challenges of making sure that your attributes are
- 9:24set correctly, we know they're incredibly important for search
- 9:27and faceted navigation. Well, actually, I think if you,
- 9:31if you use AI in the right way, you can get it to do a lot of
- 9:34that legwork in terms of getting product attributes set correctly
- 9:37across the entire catalog, whether it be 10 products or
- 9:4120,000 products or more. So that's one area where I would
- 9:46you know, say that re platforming from a brand's
- 9:48perspective is changing. It's making it easier.
- 9:52Another area, and we had an interesting example of this is
- 9:56clients doing prototyping. So this was always traditionally
- 10:00done agency side because you need a team of designers,
- 10:05specialist tools, understanding of of what's possible.
- 10:10And actually, I think now brands can lean into AIA bit more and
- 10:14do some of their own prototyping because they are, you know, the
- 10:18the agency's job is to extract the idea from the client, right?
- 10:21Is to understand what their requirements are, what they want
- 10:24to achieve. And if they can come with
- 10:28prototypes, that makes it a hell of a lot easier for the agency
- 10:31to understand exactly what they're trying to, to implement.
- 10:35And then of course, to to, you know, bring our own knowledge
- 10:37and skills and, and add to that. But yeah, I mentioned the
- 10:41example. So recently we have been looking
- 10:45at improving the sort of click and collect flow on top styles.
- 10:49And they came to us with a prototype built by them in
- 10:54Claude code. Obviously not, you know,
- 10:56production ready or anything like that, but it was a
- 10:58fascinating example of how they were able to very quickly
- 11:04explain to us their thinking and idea of how they wanted it to
- 11:07function through what was a very impressive prototype.
- 11:12So I think brands need to be thinking about or e-commerce
- 11:15teams need to be thinking about those two areas at least, you
- 11:19know, how could they bring AI into their data re re platforms
- 11:23and how can they help their agencies better understand what
- 11:25they want to achieve from, from the new website through, through
- 11:29prototyping. And, you know, just to ram that
- 11:33all off. I think ultimately re
- 11:37platforming is, is, is just getting easier and it's getting
- 11:41faster and a lot less risky. So where people used to fear it
- 11:45before, I actually think you know that is going to change
- 11:49over the next year or two as as these large scale projects
- 11:53become much more manageable with the help of AI.
- 11:56Ready to grow your business online?
- 11:58With World Pay e-commerce, you can deliver fast, secure
- 12:02payments your customers love, backed by global reach and real
- 12:06people powered support. Scale with confidence and
- 12:10enhance your customers experience with World Pay
- 12:13E-commerce. I think I think it's really
- 12:15interesting you make the the mention about moving from
- 12:18engineers to architects with more code being able to be
- 12:22written, but people needing to have an oversight of
- 12:24understanding how that code fits into a wider ecosystem, what the
- 12:27other dependencies on that are. And that I think there's real
- 12:30parallels with where client teams are evolving because
- 12:34you're, you're going to have in time fewer junior level like E
- 12:38com admins, execs, because actually you can automate so
- 12:41much the the, the, the groundwork that those people do,
- 12:44which is foundational and critical to storefront
- 12:46management, whether that's image cleaning, alt text generation,
- 12:50product data cleansing, upload and you name it.
- 12:53And then more people able to execute on like thinking based
- 12:57tasks and more strategic areas. So yeah, very interesting
- 13:00indeed. Yeah, people really need to
- 13:04think about elevating themselves now.
- 13:06That's the the key because AI is, is capable of doing the
- 13:10grunt work. So that would be my focus.
- 13:14You know, that would be my advice to anyone is how do you
- 13:17elevate yourself? How can you do more of that more
- 13:19challenging high value work and less of the low value grunt
- 13:24work? Yes.
- 13:26And on. On what Nice no to segue into.
- 13:30So Yak of the the the LinkedIn post you put about and you put a
- 13:35very bold statement and and knowing you that you don't put
- 13:37bold statements out flippantly. You put them because of having
- 13:41put the time and effort to really think it through.
- 13:44But Claude, Claude, code Co worker and design is the most
- 13:46powerful customizer customized ecosystem we've worked with.
- 13:49And you've worked with lots of different ecosystems.
- 13:51So can you if people listen to some context, like how, how is
- 13:54that? How is that like validate?
- 13:56How's that not hyper? Hyper hyperbole, basically.
- 13:59Yeah. So I'll try to explain what what
- 14:02we do. And I want to preface it by one
- 14:03thing. I'll be talking a lot about
- 14:05Claude and Anthropic. And I think it's for two
- 14:07reasons. Because we ultimately come from
- 14:09a tech world and the Anthropic and Claude called has been the
- 14:13leading solution in that world for quite some time.
- 14:17But also, I think Anthropic are having a bit of a moment, like
- 14:21their ARR basically quadrupled recently because they are
- 14:26implementing tools which are very much pragmatically focused
- 14:30on empowering office workers. They have a very narrow focus as
- 14:34a company, whereas some of their competitors like Open AI have a
- 14:37lot of blue sky thinking. But I want to pre phase that
- 14:40whilst I'll be talking about the cloud stack, conceptually the
- 14:43same things can be applied on, you know, Microsoft stack,
- 14:47Google Gemini or with Open AI. It's just, I think right now
- 14:51cloud has the nicest, easiest to use package.
- 14:54So I'll be talking mostly about that.
- 14:56And the one thing I must start with is just agentic versus
- 15:01transactional AI. And I know agentic AI is the
- 15:04buzzword of the year and everybody is talking about it,
- 15:06but they just want to make very, very clear that that is the
- 15:10differentiating factor and the thing which makes the AI use
- 15:15very, very useful for people. The I'll, I'll just use a very
- 15:20simple example the industry is using to differentiate between
- 15:23transactional and agentics. So transactional AI is like a
- 15:26vending machine. You put money into it, you
- 15:28select your drink, it tries to fulfil the order, tries to give
- 15:31you what you want. If not, it throws an error and
- 15:33then you have to deal with it yourself.
- 15:35Agent AI is more like a concierge.
- 15:38So you tell it I want to book at this restaurant for today, it
- 15:42tries to book it. If it cannot book it, it goes to
- 15:46other restaurants, it calls them up, it tries to find solutions.
- 15:49It will work and it'll spend an hour on the phone for you to try
- 15:53to find a solution and only after that it will come back to
- 15:55you. So agent take AI, the real power
- 15:58is that it is task oriented. It's try to deliver a task and
- 16:01it will spend 10s of minutes or even hours working on that task
- 16:04without needing your input in code.
- 16:08When we are developing complex things, the AI can be working
- 16:10for 468 hours in one go without the human interruption.
- 16:15It makes it really, really useful because the more the
- 16:18longer it works and it does things like self validation, the
- 16:22higher the quality of output because you're not just seeing
- 16:24something in generated in a minute, you are seeing something
- 16:27which went through a very big process.
- 16:30And cloud Co work is basically cloud code for office workers.
- 16:34So it uses the same principle of agent of something which will
- 16:38work for you for a long period of time, but surfaces it in very
- 16:42simple interface, which looks exactly like the normal chat bot
- 16:45we're all used to. So the power of it is that it
- 16:49comes with a lot of connectors. So in our world we can connect
- 16:52it to Google Drive, we can connect it to click up our
- 16:56project management tool, Jira, our project management tool.
- 16:59There are many connectors out there.
- 17:00There is a good level of permissions, so you can make it
- 17:04very clear what you want it it to access or you don't want it
- 17:07to access, write, read permissions, etcetera.
- 17:10But ultimately it means that if 05 minutes of set up, you can as
- 17:14a project manager, for example, or as an e-commerce manager, you
- 17:17can get yourself into a situation when this tool has
- 17:20access to do a lot of to a lot of context, which is really
- 17:24important. So it understands you can go to
- 17:27your files, to Google Drive files and see statements of
- 17:30work. You can go to the click up
- 17:31management tool and can see what the interactions are between
- 17:37between different teams. And you can use it
- 17:39automatically. So I'll give you an example, for
- 17:41example, one thing I'm using cohort for.
- 17:43So as a head of operations, one of my jobs is to make sure our
- 17:46team spends our time efficiently.
- 17:49It's quite. Hard because there are over 100
- 17:51of tickets every day touched in JIRA.
- 17:54So I work with Co work and I created a little routine which
- 17:57likes which runs every day at 8:00 AM where it goes through
- 18:01Jira click up and it basically looks at what individual
- 18:06developers and UA members and product managers were working
- 18:09on, which task appear to be stuck, which task appear to be
- 18:12going over their time allocation and then produces a report.
- 18:18But it does more than that because when it's when it sees
- 18:21that there's something is looking strange, something needs
- 18:24for the investigation, it does the investigation automatically.
- 18:27It might check and it might already provide a solution.
- 18:30So it might be, it might basically do the triage of, oh,
- 18:33I can see this task is stuck. I've actually then looked
- 18:36through your drive and I can see there's a statement of work
- 18:38here. And I can see that the
- 18:39conversation which is being had in the quick up tool doesn't
- 18:43really marry up with the statement of work.
- 18:45So maybe this is something where you should intervene.
- 18:48So it's saving me so much time and giving me a lot of
- 18:50visibility. And I can easily imagine how an
- 18:53e-commerce manager could do something similar with is you
- 18:57can connect cowork to things like Google Analytics.
- 18:59So go through my Google Analytics search the web about
- 19:02the what was happening. Are there any new branch
- 19:05mentions about my brand? Was there any news article which
- 19:07came out about my brand yesterday?
- 19:09Go through that, read them, correlate with GA traffic,
- 19:12correlate with my spent or or performance of ads and produce
- 19:18the report And it just does it once you set it up, it just does
- 19:22it every day, kind of runs in the cloud and you basically
- 19:27start your day with this report and that's just so powerful and
- 19:32so easy to access. I want to say one thing about
- 19:34prompting. So prompt engineering I think
- 19:36was a big thing two to three years ago.
- 19:39I would say right now all you need to do to get these amazing
- 19:43results is to provide it with context and then basically just
- 19:47know what you want to do as a human.
- 19:50So I don't think like all of the engineering tactics which
- 19:52weren't necessary three years ago, like you have to talk in
- 19:54special language, etcetera. I don't think that applies
- 19:57anymore. I think now it's just like if
- 19:59you know and you can describe in a human language what you want,
- 20:03the AI usually gets it and you can iterate a bit so that you
- 20:05can you can clarify things. So it's really the simplest city
- 20:09of it and the easiness of normal, you know, standard
- 20:13non-technical people to access this agentic AII think that was
- 20:17the that was the big moment for me and for us.
- 20:20Yeah, and I, you know, interestingly I've been looking
- 20:22at on on what role it will play on the kind of projects I get
- 20:25involved in like a big free platform in tech consultancy.
- 20:29Because one of the one of the big pains in the Arsenal
- 20:31projects is no matter documentation you have.
- 20:35Yeah. And cross referencing and
- 20:37instead of having to write, I've got to open functional spec and
- 20:40the scope and I've got a cross referencing.
- 20:42Well, actually the ability to just go, right.
- 20:45I have a query on this. Can you find me the relevant
- 20:47info line details, the cost of that and you can find the scope
- 20:51and can you summarise it for me, please?
- 20:53There's just so many ways that that every it doesn't matter,
- 20:57does it? It doesn't matter what you do.
- 21:00There are applications you can find and I like the example you
- 21:03gave about running, you know, running code like 468 hours in
- 21:06the background. I had AI had a project where I
- 21:09was using it was Chachi PTI. Haven't tried Claude yet.
- 21:13My Chachi PT to clean up product data and it was running loads of
- 21:16pipe and they were they were like iterative.
- 21:19So it was, I had about 16 different functions I needed to
- 21:23run that. If I went and tried to do that
- 21:26manually without any without developer who would then have to
- 21:29build that out and write the scripts and then run it would
- 21:32take me a long time. But I, I get, I get the the
- 21:36prompt and check that it's understood it correctly.
- 21:38Do you understand, can you run this?
- 21:39It goes yes. Can you just answer ABS you go
- 21:41through this discussive process to the point they go, well, I'm
- 21:44clear. Do you want me to run that?
- 21:45And you go, yeah, off I go. And that probably took 45
- 21:49minutes or so to run through. And it kept iterating and
- 21:52checking itself. Well, actually, I now need to do
- 21:53this. Meanwhile, I'm working on
- 21:55something else and I come back to it and then I might, I might
- 21:58give it the oversight, which is where we're coming from, like
- 22:01the junior level people to more senior who've got the ability to
- 22:04critique it and go, OK, that's good.
- 22:05But you haven't quite got this right.
- 22:07Can you now do that? Now runs for another 40 minutes.
- 22:09Off I go and do more work. And overall, it's probably saved
- 22:12a day or Two's worth of work. Yeah, it's, it's, it's
- 22:17incredible. And as you said, these cases are
- 22:19insane. Like one thing I did on Friday
- 22:21actually, so I had a statement of work and also you can connect
- 22:24quote to your emails. It's actually a it's really good
- 22:27too, to go through emails. So I had a conversation about
- 22:29the statement of work with the client on my e-mail and I was
- 22:32like, well, now I have to amend the statement of work based on
- 22:35that and was like, no, actually Claude can do it for me.
- 22:38And it can access the Google doc, it access the e-mail, it
- 22:41can create comments in the Google doc basically being like,
- 22:44OK, so this is, I think based on the e-mail conversation that
- 22:47needs to be amended. And I am quite proud of how good
- 22:50I am these things, but I couldn't find any mistakes like
- 22:53it was, it was really, really spot on.
- 22:55And once again, there is a you have to put some work up for
- 22:59into it. So the reason Claude works so
- 23:01well for me is that I've at the beginning I've I taught it a bit
- 23:05more myself. So I have a system prom
- 23:06basically telling I like things, you know, briefly, pragmatically
- 23:10in assassin way. Don't write wrong articles
- 23:14because there's also true AI. Some things people don't know
- 23:18about AI is like it's too expressive.
- 23:20It writes too much text. It creates a lot of, you know,
- 23:23AI slop text, all those convergevented.
- 23:27Yes, yeah. All that can be prevented if
- 23:30just like spending some time at the beginning to make sure it it
- 23:33works the way you want it to. And it's really good at
- 23:35following all those rules these days.
- 23:38You know, just to, to add to that in terms of examples, one
- 23:42very simple example that I use it for now on a very regular
- 23:46basis. And, and this comes back to
- 23:48Jacob's point about context. And I, I, you know, in the
- 23:52e-commerce space, we're all, we're all used to the same
- 23:55content is king. That was, that was the main
- 23:59saying in SEO for a very long time.
- 24:01Context is king in AI. That's absolutely crucial.
- 24:05And if you connect these, the, the sort of this data up to your
- 24:10AI, it does open the door to very simple, but very, very
- 24:13useful workflow. So for example, I think we all
- 24:16live in a world now where we've got too many channels of
- 24:19communication. Often it, you know, it's Slack,
- 24:22it's Google Chat, it's Microsoft Teams, it's emails.
- 24:26And you know, I'm sure, I'm sure people could list even more that
- 24:30exist in their life, like Whatsapps become a bit of a, a
- 24:33business channel as well. You can actually connect all of
- 24:36them up to your, to your AI system.
- 24:39And, and one of the bigger challenges is remembering where
- 24:43you've had a conversation, which channel it was on.
- 24:46And, and often you kind of, you know, you know roughly what the
- 24:49answer is, but you need to read it again.
- 24:52And you spend most of your time trying to work out where that
- 24:54conversation happened, which channel it was on.
- 24:57If you connect them all up, actually, you can get to a point
- 25:00where you can just ask the AI, you know, we just, I discussed
- 25:03this at some point. Could you just summarise it for
- 25:05me and it's able to look through your various different channels
- 25:08and bring that summary back saving, you know, saving you
- 25:12time in it in a, in a fairly, in an exercise that you shouldn't
- 25:18really have to have to do anyway.
- 25:20But it exists in this life. So it's finding those little
- 25:23hacks that can buy you back quite a lot of time.
- 25:26Yeah, I love it. Anything that that cuts out
- 25:28tedium is to be thoroughly, thoroughly embraced.
- 25:34Jakob, you also mentioned Claude design, something that I've not
- 25:38looked at, but I was my interest was pique because when we were
- 25:41chatting about doing this episode, you'd mentioned that
- 25:44that Claude design can benefit developers and E Comm teams.
- 25:46And you gave me an example. You said, I think you said
- 25:48Stripe, we're looking for somebody who are going to train
- 25:50their marketing team on using it to get like greater automation
- 25:53efficiencies. Can you talk people through what
- 25:55it's called design? Where does this fit into the
- 25:57ecosystem? So Godesign is about 3 weeks old
- 26:00now and it's basically an iteration of the engine is
- 26:04called code, but it's interfaced as a design tool.
- 26:07So it's it's it's an online tool where you can go and you can
- 26:11upload. And this is really important.
- 26:13You start by setting up your design system as a brand.
- 26:16And you can do it in multiple ways.
- 26:17If you have, for example, a Figma file with your design
- 26:19system, you just upload it and it reads it and it's really good
- 26:22at understanding it. You can give it your code base
- 26:24so it sees the CSS styling. You can give it your website and
- 26:27it'll try to get the styling from the website.
- 26:29Obviously the first, the first one gets the best results, but
- 26:32there are there are ways. And once it's set up your design
- 26:35system, you can ask it to generate, have presentations,
- 26:39generate prototypes, generate full websites for you, which can
- 26:43then be handed over to quote, quote to develop.
- 26:45But where I think it's immensely powerful is like, we've seen
- 26:49these tools before, but two differentiating factors is 1.
- 26:54It's really good at following those brand guidelines.
- 26:57We've been trying these tools all the time and always it's
- 26:59like kind of looks like the brand, but actually not really
- 27:03the tier. If you spend half an hour
- 27:07setting up the design system properly, like everything which
- 27:09comes out of it adheres to your standards really, really well to
- 27:13a degree that it basically produces designs which can be
- 27:17sent to developers in our quite complicated design system, which
- 27:21is very, very impressive. And it's interesting what this
- 27:23will mean to Figma because when it's launched Figma stock dipped
- 27:27a bit because this is definitely a where Anthropic are kind of
- 27:33going into that territory. And have you seen the
- 27:35controversy about like the board members separate from the
- 27:37companies etcetera that's all driven by this because Sigma
- 27:40sees this as a quite a big threat.
- 27:42It's not exactly the solution like Figma, but similar.
- 27:45What it means to us though is that now e-commerce managers
- 27:49account strategies, in our case Atomico are able to generate
- 27:53marketing materials which are high quality and actually on par
- 27:59with what the brand does. I can now create presentations
- 28:03which are looking amazing and are on brand and have the right
- 28:07context. So to finish my flow, I started
- 28:09last time. So once I amended on Friday my
- 28:12statement of work and the statement of work is A5 page
- 28:15technical document and I know sending it to a group of people
- 28:19including ACEO CE OS never going to open a statement of work.
- 28:23So I uploaded it to core design and asked it create a
- 28:27presentation which summarizes this statement of work on three
- 28:30slides you are doing you are creating this for ACEO.
- 28:35And it, it did really good job and had to spend maybe 5 minutes
- 28:38after that correcting a few things didn't get right.
- 28:41But I automatically, I ended up with a really good looking
- 28:44presentation, which with, with with very little effort and was
- 28:49interesting. Last week we saw, I think, and
- 28:51it's, I'll talk about this Stripe example, but I think
- 28:54there are more brands. So this role called forward
- 28:57deployed AI accelerator marketing basically means Stripe
- 29:01are taking one of their technical team members, putting
- 29:05them into a team of 20 marketeers.
- 29:08And the job of this person is to take the marketeers on a journey
- 29:13to get familiar with AI integrations, AI work flows to
- 29:19begin with, help them set up, show them the way, but make them
- 29:23self-sufficient, which I think is very powerful.
- 29:25And we've done something like that on a small scale.
- 29:27Domenko, when we took our account strategist and Tom and I
- 29:31being the techies, kind of show them how these tools operate.
- 29:34And now they absolutely love it because they can create a case
- 29:37study in half a day with no need from for a design team and no
- 29:42need for any tech team. They can turn it into video all
- 29:45by themselves in a really high quality and they just feel so
- 29:48empowered. So now they are like, we've
- 29:50definitely done that change management, right?
- 29:52Because they are now super AI and they are looking for other
- 29:55ways and other things they can be doing.
- 29:57And I think this mind shift and what Tom is saying quite often,
- 30:02it's like with AI, everybody has this aha moment when you're
- 30:05like, this is a tool which is actually really good.
- 30:08This is the tool which will change my life.
- 30:10So for our account strategies was this moment when he had this
- 30:14basically Wardog with the case study and he was able to use AI
- 30:19code design to turn it into a really good looking PDF case
- 30:23study and then make a marketing video on top of it in about half
- 30:27a day. Yeah, amazing.
- 30:29Let's get let's get into a bit more deeper into the AI
- 30:31workflow. So Tom, I'll come to you about
- 30:33how how you're seeing this fit within Tom and Co internally,
- 30:38Yakovsky, one example. And then Yakov, I'll come back
- 30:40to you about the the fun you're having building workflows for
- 30:43clients. So yeah.
- 30:44So Tom where as a as an agency like Co founder and director,
- 30:48where do you see, what do you see a is role in improving how
- 30:52your business operates? And what are the other examples
- 30:55of of where it's having tangible benefits?
- 30:59Yeah, I mean, there's, there are many places that it can help.
- 31:03I think one of the things that we've been doing in Tom and Co
- 31:06for, for many years and the irony is, you know, this has
- 31:10been around forever is, is automations and, and people are
- 31:13familiar with platforms like make.com and Zapier and so on.
- 31:20We started implementing those kind of workflows into, into the
- 31:24business quite a few years ago. I think it's probably about five
- 31:28years ago that we started working with, with workflows to
- 31:31try and help aid efficiency and also remove some of that grunt
- 31:34work that people were were having to do like task
- 31:37management, status management and so on.
- 31:42But about two, two to three years ago, we started injecting
- 31:47AI into that. And I think this is where all
- 31:50businesses can can benefit. And it's part of the reason why,
- 31:54you know, we as a business are starting to provide this as a
- 31:57service. So by injecting AI into
- 32:02workflows you get to to, you get the capability of doing a hell
- 32:07of a lot more than you could before.
- 32:09Workflows were often very programmatic.
- 32:12You know, it's got 10 steps. It must move through those
- 32:15steps. If one of those steps doesn't
- 32:17work, then it will grind, grind to a halt.
- 32:21But whereas if you put AI into it, you really turn it into an
- 32:26intelligent workflow, which sounds a bit buzzy.
- 32:29So what what I mean by that is you can make a workflow non
- 32:34deterministic in an intelligent way, which is very much like a
- 32:38human. Let's face it, you don't, you
- 32:40don't know exactly how they're going to do the task, but
- 32:42because they're smart, they'll generally get it done.
- 32:45And that's really the, the shift that we're seeing now with, with
- 32:49AI is certainly in the kind of the workflow space.
- 32:54So I do think that's important that businesses look at
- 32:58workflows more seriously than they have in the past.
- 33:02Some businesses, certainly businesses that we've worked
- 33:04with have done that. I'm sure I don't know James you
- 33:07worked on. Projects where they've done.
- 33:09Sort of data layer integrations and service layers where sort of
- 33:13systems talk into a single layer and there's some automations
- 33:16around moving data around. It's sort of that kind of thing,
- 33:19but with intelligence inject injected into it.
- 33:23But vitally, it's making sure that those workflows and AIS or,
- 33:28or specifically the AI has tools available to it.
- 33:32And people might be aware of of the term M CP/M CP is the the
- 33:40acronym for a essentially a way of connecting tools to to AI.
- 33:45It's protocol and that allows AI to do a lot more than just look
- 33:51up data and, and, and respond with information.
- 33:54It can take action. And if you allow it to take
- 33:57action, this is where we move into this agentic position where
- 34:00it can, it can do a hell of a lot more and it can work on its
- 34:03own over many, many hours and just kind of give, put that, put
- 34:09some colour into that with, with examples.
- 34:12So just a few of the things that that we're doing internally at
- 34:17Tomaco is things like if we have a client brief submitted to us
- 34:23that immediately kicks off a an AI workflow that first of all
- 34:27searches are various different databases.
- 34:30So JIRA Confluence for any similar brief that we've had
- 34:35before, any similar work that we've implemented before.
- 34:37And that might be from last week, it might be 3 years ago,
- 34:41probably on a, on a different client.
- 34:44And then when it looks up those similar tasks, it can, it can
- 34:48assess what we did in those tasks, what was the information
- 34:52that we captured for the statement work and the
- 34:55requirements and so on. And then it can start to flag
- 34:58potential missing information in the brief that we've received,
- 35:01but also flag questions that we should be asking the client or
- 35:06perhaps want to ask the client to help clarify that brief.
- 35:10And then to top it all off, it's actually able to look at the,
- 35:13the the previous code that's been written.
- 35:16So it can delve into the code. If it's found the similar task,
- 35:19it can delve into that code to give it more additional context
- 35:22around how it works and some of the things that we need to, we
- 35:26need to think about. And then off the back of that
- 35:28workflow, we've got another workflow that will will trigger
- 35:32and that one's to write an initial statement of work.
- 35:35So it will look at what the first workflow is found and it
- 35:39will, you know, Jacob was talking earlier about example
- 35:42of, of generating a statement of work.
- 35:44So off the back of that workflow, we can, we can get an
- 35:47initial statement of work ready for our business analyst team.
- 35:50Of course, it will be imperfect, it won't be right, but it's a
- 35:55starting point. And I think this is also another
- 35:58really important thing to think about with AI.
- 36:01It, it, you know, if you, if you want AI to, to, to solve
- 36:04everything and do it correct first time, you're going to be
- 36:08very disappointed. However, if you accept that
- 36:12maybe 50 to 80% of the time it can, it can solve it.
- 36:16And the rest you need to tidy up after it.
- 36:20You'll be, you'll be, you'll be very happy with the results.
- 36:23And you're obviously getting all of those massive efficiency
- 36:26improvements because it is getting some of the work done
- 36:29for you without your intervention.
- 36:33Yeah, it's, it's some other examples.
- 36:35So we have developer handover notes written by AI now.
- 36:39It was always one of the biggest challenges for us, which is
- 36:42developers hand it over to QA and they get 3 lines of
- 36:45information from the developer about how to how to test it or
- 36:49what the functionality is or where there's a new feature or
- 36:52setting. So now when it's handed over, AI
- 36:56reads the code, it looks for any changes in the, in where new
- 36:59settings might be, sorry, excuse me, where any new settings might
- 37:04be and helps write a detailed handover note to the to the QA
- 37:09member who's about to test it. So you know that it's less
- 37:13likely that they'll miss something in their testing.
- 37:17And just finally, as another example of where we use it,
- 37:21sorry, excuse me, is service desk.
- 37:25So yeah, when a, when a new issue comes into service desk
- 37:30with perhaps something isn't working as it should or actually
- 37:34just looking for help how to manage the website, AI does the
- 37:38initial investigation. So AI will look at the request,
- 37:42look at the code, it has access to our error tracking tool and
- 37:48our log tracking tool and it can do an initial triage with that
- 37:53ticket and help steer people in the right direction.
- 37:55So again, not so, not necessarily solving it, but
- 37:59reducing a huge amount of time that someone might need to spend
- 38:02digging through various lock files and and and errors and so
- 38:06on to try and identify what what might be the problem within the
- 38:09service desk ticket. This is really interesting
- 38:12because work working between clients on ecom teams and
- 38:15agencies like yourself, where where the clients I want to get
- 38:19is they want to spend their money on value add, not on on
- 38:24fixing things that that are not fun to do or on raising task
- 38:31where actually it's not a bug or an issue.
- 38:33It's more just a information that can be passed back on how
- 38:37to do something properly within the storefront.
- 38:40So the the the more that AI can can reduce the time of
- 38:44developers being distracted from the like the proper road map
- 38:48items, the better. Yeah, absolutely.
- 38:52And it's it's about winning back time.
- 38:54So it's allowing teams to focus on on the higher value staff and
- 38:58having more time in their day to get worked up.
- 39:01Yeah, and make developers happy rather than them to go, oh,
- 39:04they've raised that again. Well, absolutely, yeah.
- 39:08The more time you can get them focusing on feature development
- 39:11rather than service desk, it certainly makes them happy.
- 39:14Exactly. But let let's flip to the
- 39:15client. Sorry, Akim, you're going to say
- 39:17something I. Just want to say one thing
- 39:18because there's one thing which don't mention which some people
- 39:20get spooked about, which is it's almost like, OK, you have to
- 39:23accept that the AI gets it only 70 time 70% right.
- 39:27And some people are worried about it for right reasons
- 39:30because sometimes AI might be doing things in your production
- 39:33system which are very important. And if we just say get a 70%
- 39:37right, people, what happens in the last in the next 30%?
- 39:40So you're going to read my database.
- 39:41So I think there's 2T things I want to mention, which is a
- 39:46obviously there is proper permission system which needs to
- 39:49be in place. But the second thing is I think
- 39:51what people maybe don't think about AI is really good is
- 39:54specialist agentic AI. It is able to estimate how well
- 40:00it thinks it did reasonably well.
- 40:03So if it's a task where it needs to do something really
- 40:05complicated and it needs to write into a production system,
- 40:08into ERP system or something like that, you can easily
- 40:11basically intersect. And and have an agent reviewer
- 40:14in the middle being, does this look safe?
- 40:17Does this look like this is actually doing what it should
- 40:20do? And does it look safe?
- 40:21If yes, proceed. If no, there's this human in the
- 40:25loop inter interdiction and basically it can just an e-mail
- 40:28and say, oh, I've prepared this, but I wasn't really sure if I
- 40:33should do it. So here you human, please
- 40:35decide. And then massively reduces the
- 40:37risk because and it's, it's once again, it's not 100% automation,
- 40:41but it's even now have an automation which works 70% of
- 40:44that automatically. And it's fully safe.
- 40:46And I think there's an important concept of people.
- 40:48We, we, we normally just use the rack red, amber, green system
- 40:53green, everything just goes automatically amber, red human
- 40:57needs to be involved in the loop.
- 40:58And this is a way how to massively de risk all this risk
- 41:02around the stochastic nature of AI.
- 41:05Yeah, and it's and, and actually there's a really nice, I mean,
- 41:07you can send an e-mail, but Jacob and I have implemented
- 41:10workflows that that interact on chat.
- 41:14So you it, it an automation that kicks off in the background.
- 41:18It starts a particular task, it gets to a point where it needs a
- 41:21human in the loop and it sends a Slack message and it says, I'm
- 41:25just about to do this thing. Are you happy for me to do it or
- 41:28I need your clarification on this.
- 41:29Can you give me some more information?
- 41:31You send it back and it goes back into the workflow that then
- 41:34continues on working through the original task.
- 41:38But I think this is this is the critical thing.
- 41:40The two bits you guys touch on. One is 1 is governance, but one
- 41:43is open mindedness. And I think there's always there
- 41:47is a tendency because the we've all been in technology long
- 41:50enough to know that hype cycles come.
- 41:51And then there are self interested parties who want you
- 41:54to believe that everything should now be with this, this
- 41:56text going to fix everything in the world.
- 41:58And that then creates expectations that are the
- 42:01warped. And then people use an AI and it
- 42:03doesn't do things perfectly and they go AI is crap.
- 42:06And actually AI is not about being perfect.
- 42:08Humans aren't perfect. Developers aren't perfect.
- 42:10Like we all make mistakes. We can't be perfect 100% of the
- 42:13time. But if if something that's 60 to
- 42:1770% perfect can save me 3 hours a week and that helps me make
- 42:23better decisions because I can Polish it quicker.
- 42:26I use it for content generation, like spinning up summaries of
- 42:32the blogs, the podcast we're recording, because that's far
- 42:35more efficient. You should try than me writing
- 42:37from scratch. But take take the core notes
- 42:39from it and then me add the human editor on top to put it in
- 42:43the right tone of voice and to write it how I want it to be.
- 42:45That's saving me an hour or so every single time is is you've
- 42:49got to come into it with an open mind that says how do I use it?
- 42:53Not expecting it to be absolutely perfect.
- 42:55But then the other bit you've touched on, which is critical is
- 42:57the human part, which can never go away as governance because
- 43:00throwing tech and automation workflow without oversight and
- 43:03knowing how you're going to scrutinize it.
- 43:06It's no different to, you know, senior engineers overseeing mid
- 43:10end engineers who oversee junior engineers and the quality
- 43:14control process you go through to centre check things are
- 43:16production ready. So I think, I think that's an
- 43:18area that ECOM teams who maybe also don't have that engineering
- 43:23discipline when they're putting AI the governance piece for how
- 43:26it's working is going to be critical.
- 43:28It's really interesting that you mention that actually, because
- 43:30internally we talk a lot about how you bring the code review
- 43:35process into the rest of the business, because it is very
- 43:39much true that, you know, within, within technical teams,
- 43:42nothing goes live without code being reviewed.
- 43:45So a junior amid a senior write code and it will be reviewed by
- 43:50someone else. And, and you need to apply that
- 43:53same principle to AI in every place.
- 43:57But if you apply that principle, then you can move away from this
- 44:00worry. Or is it going to do something
- 44:02that it shouldn't do to a position of it's going to do the
- 44:06work for me And then I'm going to elevate myself to just
- 44:09reviewing it, make sureing it's, it's fine.
- 44:11It's very much kind of a managerial type of approach to
- 44:15things. And then either approve it or
- 44:18feedback and, and and and let it go back to work on that
- 44:21particular task to because it didn't get it right or take it
- 44:23off its hands because it's just, it's, it's beyond what it's
- 44:27capable of doing and you need to finish it off.
- 44:29For that learning pieces, this is an area, another area I think
- 44:32people need to to realise that is the the learning capacity is
- 44:36it's going to have hallucinations.
- 44:37They're getting better to to be fair, there's fewer it's going
- 44:40to make errors, it's going to have false assumptions.
- 44:43But every time you correct it and give it the evidence, it's
- 44:46learned and it's stored that now human brains can get things
- 44:50wrong and then we can go right. Oh, actually, I shouldn't do
- 44:51that. But there is a risk of it
- 44:53repeating because we are, we don't have a, a like perfect cue
- 44:59retrieval system in our brains to that exact piece of
- 45:01information at the time that we need it versus the way that an
- 45:04AI model will look up to its learning base.
- 45:07So I think that's the thing. The, the where it's going to get
- 45:09more and more powerful is the context that humans, senior and
- 45:13experienced humans give it. And the control means that the
- 45:16accuracy levels when you're doing repeat tasks are going to
- 45:18go up and up and up. Well, this is where you're,
- 45:22you're really getting into the kind of the, the, some of the
- 45:24technical challenges that exist in, in AI models at the moment
- 45:28that people are trying to solve. So although it's recall of stuff
- 45:33that you're working on is near perfect.
- 45:35And they talk about needle in the haystack test where you
- 45:38know, for a long time now, for over a year, AI has been pretty
- 45:41much acing that particular test, which is remembering a small bit
- 45:45of information within a large piece of information.
- 45:48It's exceptionally good at that. But we operate in AI within a
- 45:52context window. And it's, it's like a
- 45:54limitation. So most models now are pushing
- 45:57the 1,000,000 token context, which is a vast amount of
- 46:02information. But in a lifetime of working,
- 46:05it's a it's an absolute drop in the ocean.
- 46:09You know, you can fit the whole set of Harry Potter novels into
- 46:12it, but the amount of information that you deal with
- 46:15on a day-to-day basis means that that context window is filled up
- 46:19within within hours. So once it moves out of that
- 46:24context window, it's forgotten. And this is where you see this,
- 46:27this problem with AI making the same mistakes over and over and
- 46:31over. So.
- 46:33You know the there's. Incredibly intelligent people
- 46:36working at places like Open AI and Anthropic and and and Google
- 46:40are trying to solve this problem within the model, but you can
- 46:44already solve it quite well outside of the model with sort
- 46:48of long term memory, memory strategies.
- 46:51So there's something called RAG, which is essentially a system
- 46:56where you can retrieve information that's outside of
- 46:59the context window. And that might be things like
- 47:04the things that you learnt a week ago about how to write the
- 47:07statement of work, the statement of work that I like and what are
- 47:10my preferences about it. Or you can use markdown files
- 47:15for, for memory system as well, which are really just text files
- 47:18that get indexed and pulled back into the context when they're
- 47:21needed. So there are techniques that can
- 47:24give AI models long term memory and but you need to be selective
- 47:29about, you know, what it is you're actually remembering over
- 47:32a long period of time. One of the workflows that Jaco
- 47:36worked on recently, he implemented A beautifully simple
- 47:40solution to this with a with a spreadsheet.
- 47:43And the client has access to the spreadsheet, they can insert
- 47:47rules, essentially memories about how it should do the
- 47:50particular task. And every time the workflow
- 47:52runs, it pulls those those rules from the spreadsheet and they
- 47:57act as long term memories around how they want it to complete a
- 48:01particular task. That's, that might actually be
- 48:04an interesting workflow to talk about because James, I know we
- 48:07wanted to mention some we did for our clients.
- 48:09So this one was a for B to B client and he was cut two things
- 48:13in one. So firstly, they, they get
- 48:17enquiries through their website. People are asking for quotes, So
- 48:21and those enquiries contain some basic if I want to enquire for a
- 48:25quote, I need to type in my basic contact information and
- 48:27some some other details. And basically they, there was an
- 48:32inbox, they were going to a member of status monitoring that
- 48:34inbox. New inquiry came, did like a
- 48:37little triage, like is this spam?
- 48:39Does this look valid? If it's OK, we go to their ERP
- 48:42system and raise it as a sales opportunity in the ERP system.
- 48:45And by raise it, I mean copy paste the name, copy paste the
- 48:48phone number, copy paste the e-mail into relevant fields and
- 48:52the ERP. And then we'd reply to the to
- 48:54the lead, basically. Yeah, we got your inquiry,
- 48:56something like that. So the first step we did was
- 49:00obviously automated this flow, because their challenge was that
- 49:03it wasn't coming from single phone.
- 49:04That would be too easy. The the e-mail inbox was getting
- 49:07stuff from multiple different places.
- 49:10So the data was unstructured. Conceptually, it was always the
- 49:13same data but in an unstructured form.
- 49:14So you can write the simple automation because the e-mail
- 49:16looked slightly different every single time.
- 49:19So we build the flow monitors the inbox.
- 49:21Every single time e-mail comes, it checks, is this an inquiry?
- 49:25If yes, copies to the ERP and basically replies back to the
- 49:29operator saying all done. Once again, if it if it fails or
- 49:33something, it triggers a different flaw.
- 49:35Would be like, OK, I couldn't do this one.
- 49:38I please could you help me? And then when the ERP, what
- 49:41happens is they call the customer and get some more
- 49:43details. So they were already using like
- 49:46kind of transcript technologies. So we have a transcript of the
- 49:49call. They, they ask more details
- 49:50about the job, etcetera. Now they need to produce a
- 49:52fault. So they had a database of like
- 49:57prefabricated. So if if somebody wanted a
- 50:00service a, the code would would contain this text and would
- 50:03contain this like item and it would cost this much and they
- 50:06had it in like different spreadsheets, etcetera.
- 50:08So we simplified, we put it in place and now the AI gets the
- 50:12details about the sales opportunity, the transcript from
- 50:14the call, the business rules about, oh, if they talk about
- 50:17this, this is the service we want to we want to adjust which
- 50:20is in a Google sheet. So if they are the new service,
- 50:23just add it to the sheet and runs a workflow and pre creates
- 50:26the quote in the ERP system. So once again, the the the the
- 50:30sales, the BDR previously had to go to the spreadsheet, copy from
- 50:34the spreadsheet to the to the ERP has a prepared quote, scans
- 50:38through it, applies a discount, sends to the client and it's
- 50:42flexible. So it's not limiting.
- 50:44They need new types. You just add it to the
- 50:46spreadsheet and it works. And that was really, really
- 50:48impactful to them. And also it was quite a good buy
- 50:52in because the people, the BDRS thought they had control because
- 50:55when I was demoing it to them, I would create a call and it'd be
- 50:58like, well, that's wrong. It did this wrong.
- 50:59I was like, OK, let me adjust in the spreadsheet, run it again.
- 51:03And they saw, oh, you got it right this time.
- 51:05I was like, yes, this is what you do in the future.
- 51:09Yeah, I, I think it's really interesting you guys have gone
- 51:12down this route as an area of focus in addition to the, the
- 51:16like the core services that you provide as an agency, like
- 51:18migrations, maintenance of existing sites, you know, growth
- 51:22acceleration stuff. So it makes a lot of sense
- 51:24because I can see more demand coming out of E Comm teams for
- 51:27this skill set, not just traditional development.
- 51:30And it is agnostic to platform because at the end of the day,
- 51:34it's about connecting data sources and running functions
- 51:38and giving people outputs they need to be better at their jobs.
- 51:41So yeah, I'm, I'm, I'm pretty hopeful that this episode will
- 51:45have opened a few people's eyes to what the potential is within,
- 51:49within Clanside, but also with how the agencies can support
- 51:53them with, with smarter workflows.
- 51:57Yeah, I think, you know, we, we, we kind of discovered the, the
- 52:01efficiency with automations and AI internally to a point where
- 52:06we thought, I mean, this is, this is clearly beneficial to
- 52:10all businesses. I mean, our, our focus and
- 52:13understanding is obviously within the world of e-commerce,
- 52:16but it, it doesn't stop there. It's, it's anyone in, in a
- 52:19working job really can benefit from from these things.
- 52:22So as an agency, it felt very natural for us to create a
- 52:26service to provide this to not just our clients, but anyone
- 52:31who's interested, regardless of whether we manage your, your,
- 52:35your website or not. And I think it's also important
- 52:39to highlight the sort of, you know, the boring but important
- 52:42operational side of businesses where we spend a lot of time
- 52:45looking at how they operate and how we can gain efficiency.
- 52:49So we do do a lot of work within marketing teams around
- 52:53e-commerce efficiency, but also just pure automation, sorry,
- 52:57pure operational side of of businesses too, where a lot of
- 53:02efficiency gains can can be found.
- 53:04So that's our aim really is to provide a service to to help
- 53:07businesses unlock internal efficiency, removing grunt work,
- 53:12which no one wants to do and allowing people to focus on high
- 53:15value thinking tasks. And one of the things I think is
- 53:20sometimes overlooked because everyone focuses on efficiency
- 53:24and not having to to do certain tasks that AI can do is actually
- 53:29quality. So one of the really interesting
- 53:33side effects of of AI is it really thinks hard about
- 53:37everything. People aren't so good at doing
- 53:39that. People, you know, if you're
- 53:40doing a simple task, people tend to switch into simple thinking
- 53:43mode and we do it quickly, don't apply much thought.
- 53:48All of that kind of AI is not like that.
- 53:49It's never lazy about the task it's doing.
- 53:51In fact, it's quite the opposite.
- 53:53It overthinks most things. But actually this has a, this
- 53:58has a really nice side effect, which is while it's doing a
- 54:01simple task, it often finds something or flags something to
- 54:04you that doesn't look right. And this helps to improve
- 54:07quantity. It means that less mistakes
- 54:09actually happen. And so yeah, that packaged up as
- 54:14a service is I think, a really interesting offer.
- 54:17And certainly since we've been talking about it and providing
- 54:21it as a service, we've had a lot of interest from people about
- 54:24how we can how we can help. But it's it's sorry.
- 54:27You go. Yeah.
- 54:28No, I just, I just want to say one thing, James, like I think
- 54:30when we might be approaching it or where we if we add value is
- 54:33that we are very pragmatic. So we would normally just go to
- 54:36a business and after a day there just do a bit of discovery to be
- 54:40able to answer to the senior management.
- 54:41What can AI do for my business on like specific examples.
- 54:45So we're then using the big words and the big presentations.
- 54:49I think what we do quite well is to be specifics.
- 54:53So talk to an finance department an hour later go back to CFO
- 54:57being like these are the specific kind of things we could
- 55:00be doing for you. And I think people appreciate it
- 55:01because there is still a lot of from the senior management,
- 55:04there is still a lot of I'm going to do AI, what does it
- 55:07mean for my business, what can I do?
- 55:09And I think we're trying to answer that question relatively
- 55:11quickly. So because once they see these
- 55:13tangible examples, it just all starts making a lot more sense,
- 55:17whereas these big concepts and buzzwords etcetera can base
- 55:21sometimes overwhelming. Yeah.
- 55:23And I think I'm back to your start, your starting point
- 55:27around agentic and transactional use of AI is, is that the best
- 55:31way to see see this is it's a collaboration and it feels weird
- 55:34tonight because you're collaborating with basically a,
- 55:37a bot and but then you start you, I've realised this.
- 55:42I start actually saying please could you and thank you because
- 55:46because it's helping me. But you've got to see this
- 55:48collaboratively and it's to get through outcomes.
- 55:51It's not perfect, but it's iterative.
- 55:53But you've got to build, you've actually got to build a
- 55:56relationship with understanding the limitations of the AO models
- 55:59because they do make mistakes. And to think that they're not
- 56:02going to make mistakes is utterly crazy because they're
- 56:05going to, they're going to send you wrong information.
- 56:07And you've got to have the human expertise to detect when it is
- 56:11wrong and guide it and correct it so that it's learnt for next
- 56:14time. And I see this in, in platform,
- 56:16native AI smart assistants, they, they are brilliant in how
- 56:22they've evolved, but they still get fundamental things wrong.
- 56:24But this is it, this, this, you know, whether whether people
- 56:27like it or not, the, the use of AI is there to stay.
- 56:29And I think, Tom, you said it that to differentiate yourself,
- 56:32you've got to demonstrate how you can use it to get value and
- 56:34add value. Yeah, Yeah, exactly.
- 56:39And, and, and do you know you mentioned there like sort of
- 56:42marketing help, chat bots and stuff.
- 56:44It's it's actually an area at the moment we tend to steer away
- 56:48from in terms of kind of our recommendations because when you
- 56:52are dealing with clients, it's absolutely imperative that
- 56:56you're giving the best possible service.
- 56:58So internally at Tomaco, we want our team to spend their time
- 57:02dealing with clients, an AI to spend time doing the graph work
- 57:08behind the scenes. And I think it's true for
- 57:10e-commerce brands as well. You want your internal teams or
- 57:14your customer service team to have time to speak to your
- 57:17clients, not to move a ticket around or to try and solve a,
- 57:21you know, a problem with a return or something like that,
- 57:24where it could be handed off to an AI that can investigate and
- 57:28find the answer for you. And the, the for me, this is
- 57:32key. This is where you, where you
- 57:34still see a huge value in, in humans rather than AI, which is
- 57:41that really human, human LED nature.
- 57:44You know, you want to speak to a human on a phone or, or a chat,
- 57:47not an AI bot. And so if you focus on how do
- 57:50you remove time constraints from other areas that are not
- 57:54customer facing, then they can do more of that.
- 57:58Yeah, I think that's that's a nice way to to round this off
- 58:00actually. And it is about it's about
- 58:02saving time on tasks that are are not adding mass amount of
- 58:06value, but they have to be done because they're just
- 58:08foundational and freeing up time to think to help to add value.
- 58:12Whether that's added value to customers, to your colleagues,
- 58:16to the to agencies, your whatever it might be.
- 58:19But look, thanks so much for joining.
- 58:21I knew this episode I'd be really interested.
- 58:22I hope that people listen have got a lot from it as well.
- 58:24I could, I could sit and ask you loads more questions on this,
- 58:27but anyone's listen, if you are interested, do reach out Tom and
- 58:30Jacob's contact details on the landing page.
- 58:32They're great people to connect with in the industry.
- 58:34Generally always wanted to have a chat.
- 58:36But also, if you're looking at how you might exploit AI
- 58:40automations and workflows to benefit your business or
- 58:43business you work with, why don't you reach out and there
- 58:46there might be an opportunity for them to, to help you solve
- 58:48the Riddle. So thanks gents for for joining
- 58:51the podcast. Oh, thank you for inviting us
- 58:54our pleasure. Yeah, take care you're listening
- 58:58to the Insidecommerce podcast. Hosted by experienced
- 59:00consultants James Gert and Paul Rogers, Insidecommerce is your
- 59:04fast tracked additional knowledge feature in interviews
- 59:06with leading e-commerce retailers, industry thought
- 59:08leads and the C-Suite of key tech companies to explore the
- 59:11latest digital strategy, customer experience and tech.
- 59:14Discover what the latest trends really mean and get practical
- 59:17advice based on the experience to help fine tune your E Comm
- 59:20strategy and tech. Follow us on LinkedIn and reach
- 59:23out to James and Paul if you'd like to discuss an ecommerce
- 59:26project. We hope you enjoy the latest
- 59:28episode.