Ben Love [00:00:00]:
All and welcome to today’s webinar. Today we’ve got the second in our two part series on AI with our guest expert Andrew Herbert. Andrew Herbert is a leading expert in the field of AI integration for business processes, bringing over 15 years of experience in managing intricate projects across a range of industries, including defense, aerospace, proptech and of course, AI technologies. Currently at Remap AI, Andrew has been instrumental in leveraging AI to streamline business processes, positioning the company as a leader in practical AI enhanced solutions. With a master’s in Business Administration and ongoing studies in data science, Andrew has earned recognition as an australian top 50 small business leader and his latest venture, Remap AI, is highlighted as one of the australian top ten tech startups to watch. His unique blend of expertise makes him a go to subject matter expert for business owners and leaders, which of course is why we have him along here today for you. So without further ado, Andrew, good morning.

Andrew Herbert [00:01:10]:
Good morning Ben. Thank you very much for the lovely introduction. Now, I just have to say that if at any point in time anybody’s got any queries, questions or wants me to recover anything, please enter it into the chat. This is about practical impact, so if anyone wants any clarity at all, please reach out. I’m more than happy to go over anything. I’m going to rapid fire through a fair bit of this. If there’s any queries or questions that you want to take offline with some sharing of personal or commercial circumstances, please reach out via the chat or directly via email. So rolling straight into this, I will be gazing off briefly into the distance.

Andrew Herbert [00:01:51]:
That’s not because I’m not paying attention, but I’ve got a second screen running to be able to share and we’ll be doing some live demos. So rolling straight into it. This is going to be very succinct in comparison to the last one. The last presentation was around the foundations and terms of what is AI? And some of the concepts you’ve probably seen kicking around recently. This is all about actual use cases that you can take home today. So first things first, integrating technology and bringing AI into your organization. So there are three categories of AI as far as we’re concerned, and artificial intelligence is a catch all umbrella term for a variety of technologies. And deploying a solution or integrating an AI solution into business is all around one of three steps.

Andrew Herbert [00:02:45]:
So the first step being an off the shelf product, you may be able to secure a software solution that is ideal for your organization or for your use case that is ready to go. Now, this is great because it’ll be serviced by a commercial provider, be in a SaaS or a tech organization, and they are cost effective and they are very rapid to deploy. But they may restrict you in customization options and what you can and can’t do with that platform. Now to amplify that one, you roll into the second stage and you get automation. So automation through a platform like Zapier or make or Microsoft automator enables you to take an off the shelf solution and just make it that much more. Now you can in this second tier actually connect it into an API or into an AI model like chat GBT and augment what you’re doing in that. And I’ll go into a use case of that that we built in Zapier shortly. But automation is where I feel as a business owner or a business leader, you’re going to get the most benefit out of AI.

Andrew Herbert [00:03:56]:
The third level is a full custom AI application. Now in our experience, these are extremely powerful. However, they take more money, more time, and they are more complex than initially first thought. We sunk a lot of money into this category and we are quite good at rolling out new solutions very rapidly, but they are always more complex than they initially first think. So, AI automation, it’s really where business owners, leaders can be able to get that benefit and value very rapidly. Now, four points around integrating technologies and AI into your business. Four recommendations. So the first recommendation is create a playground.

Andrew Herbert [00:04:45]:
So give your team, your staff, your partners, whoever it is you’re working with. Give them a series of files or an environment that they’re able to play, that they can go and test things and that there won’t be any ramifications if documentation gets distributed. So use open or non classified or non commercial documentation to be able to test these AI, these models and these platforms. The second one is release or enable testing funds. So monthly stipend or an annual amount, whatever that looks like. Just enable staff to be able to have the confidence that they can go and spend $20 to subscribe to GBT four and have an understanding of how you can use that in your organization. Going out and testing and playing is crucial to be able to bring a new platform or a new solution into your business. The third one is create a business case.

Andrew Herbert [00:05:34]:
So the person that’s had that playground experience has secured some funding from the organization to test and subscribe and play around with that platform now needs to put forward a business case to leadership and to management around how exactly this AI or this emerging tech solution will apply into their role or into the organization and what benefits that would look like, and ideally look at it, a return on investment calculation. So whether that is a calculation based on the technology, subscriptions that this new technology will replace, which is something we’ve done recently for a client, we identified a shift from one provider to another. And as a result of that shift and a reshuffle of some platforms, we’re able to identify over $50,000 per annum in savings just by a rejig in the way that we were approaching something. And then that’s on top of that, you look at the efficiency creation. Now, a lot of people can have some challenges around calculating the impact of AI, or an automation, or a workflow improvement. For myself, we use a pretty simple measure that we look at the hourly rate of the individual that we are supporting, and the time that that is smoothing out and the time saved, and then we just calculate that over the year, and we see whether or not there’s a return on investment being the AI investment of subscription service or a full application against the actual impact that it’s going to generate. So we just do that fairly straightforward calculation. So that’ll be a part of the business case.

Andrew Herbert [00:07:08]:
And then third, rolling it out. Make a stated rollout, do a test rollout, a careful adjustment, put in front of two or three staff members before you roll it out organizational wide. Let them run it for a month, see if there are any issues or any concerns that you’ve missed that you haven’t been able to capture as a part of that playground environment, and then start rolling that out organization wide. And a point to note as well, amongst all of this integrated technology, please be cautious or conscious of your commercial and confident information. So, giving information over to an AI model, just be wary about the ramifications of what they do with that data. Now, somebody like OpenAI, they have an API, which is a connection point from an app, an external app, and if you pass data in through that, they won’t actually hold that data, they will actually delete it after you’ve passed it to them and come back. Whereas if you use their GPT environment, their prompt environment, any data you give them, they actually use to train their models and improve what they do. So just be conscious of those two different methodologies of getting access to OpenAI’s GBT, but also any other technologies you’re looking at.

Andrew Herbert [00:08:20]:
Just be wary around that as well. Okay, so rolling into the recent updates as well, these are relevant for a couple of different reasons. So the first one there, GBT, four web access what this means is that all of a sudden chat GBT from OpenAI, their fourth variant, GBT four, the most powerful, the one they have at the moment, now has the ability to browse the Internet. I’ll show you a use case of that. Actually, you know what, I’ll just show it to you now. So this is a pretty straightforward use case of how you can use GBT’s four’s browsing function, and you can get that this is only available in the paid plan and you do have to enable it in your settings. But once you go into there, you can just ask it something as simple as this. And this is a social media content generation strategy that we do in house.

Andrew Herbert [00:09:09]:
Pretty straightforward. Tell me the AI news for today and you can see that it’s browsed the AI. You can browse the Internet and it’s given us five options. You can ask it for more if there isn’t anything there that interests you. But in this case, we’re interested in the small house AI models and seeing exactly what Zenml is doing there. So that’s gone out and given me a little bit more context and essentially summarized the article to key points for me. So made a nice and fast reading then I’ve gone, well, actually, this is relevant to our audience on LinkedIn. So I’ve just dropped on a LinkedIn prompt that we use as the default prompt.

Andrew Herbert [00:09:43]:
You can see there it’s about remap AI as a finishing maximum word limit target audience. I wanted to actually analyze a so what question like why do we care? And then it generates this article for us. And obviously there’s a little bit of tweaking there that’s required. Wouldn’t necessarily post something on LinkedIn with a hello, business leaders. But you can grab about that much of the post and just drop that straight into a LinkedIn post right now. And that has probably taken me about a minute to generate from the start to about where we are now. But then there’s a recent update that’s come out and that’s on this slide here, which is the Dali three, which is now the new image generation model from OpenAI. And that allows you to generate custom images.

Andrew Herbert [00:10:30]:
So there is a use, and there’s a bunch of different uses. But in this case, what I’ve done is I’ve asked it to write me a prompt for Dali three to generate a unique AI image for this article. So nobody else in the world will have this, and this will be unique and tailored to this article that I can post online. And I’ve actually got that generation here. So you can see here, I just dropped that prompt straight into a Dali three thread, and I’ll show you how to bring up Dali three and a couple of other things in a second. And it’s generated four images there, and it’s illustrations and they’re all tailored. You can see there Zenml. It’s all tailored to that article and rolling down, I’ve actually said, you know what? I wanted a bit more photorealistic and it’s dropped out those four images.

Andrew Herbert [00:11:25]:
So within about two and a half minutes, I’ve secured, or I’ve generated a time relevant LinkedIn article using GBT four, open browsing with Bing. And then I’ve gone out and I’ve been able to generate an image tailored to that article to be able to post on social media. So this is just one of the many ways you can go out and access or take advantage, sorry, of the new functionality from OpenAI. And the third one that GBT four has released recently is the document side of things. Now, I haven’t brought up a document as a demonstration on this because it can take a little bit of time. I didn’t prepare one for this presentation, but you can see here, when I start a new chat on GBT four, if I just roll my mouse over this, you’ll be able to see the options that are available. So that’s the live web access browse with Bing. There’s also the image generation Dali three.

Andrew Herbert [00:12:19]:
And then there’s also the advanced data analysis. So the advanced data analysis is the new function that is still in beta stage. So it’s not as tidy and as clean as it could be, but what it allows you to do is it allows you to upload pdfs or images into chat GBT and allows you to ask key questions. So if you have a contract, for instance, you can load a legal contract into this and say, provide me with a summary of the terms and conditions and identify any points of concern that I need to be aware of, and it will read that document for you and it will then provide you with an output. So, pretty powerful stuff. You can also ask it to.

Ben Love [00:13:04]:
Go.

Andrew Herbert [00:13:04]:
Into documents and provide analysis. So if you have multiple pages document of maybe financial reporting or a business strategy or something similar, you can provide that over into the GBT four, and it’ll provide you with the summary feedback or any other instructions or any detailed queries that you need around that documentation. Now, a point to note as well is that with GBT four, in all of these cases, it will take your data. So from a commercial and confidence perspective, just be conscious and be cautious about what you’re feeding across. There are ways around that which I’ll go into in a second, but just something to be aware of rolling with the other releases that have occurred or updating recently. So you’ve got MS copilot that is due to launch on the 1 November. So that is a Microsoft version of AI OpenAI, and it’s a copilot option that sits inside of your office suite and works with you and with your files as reference points. So very powerful impact in the Microsoft environment.

Andrew Herbert [00:14:05]:
Google has actually released duet a couple of weeks ago, and that is their variation of copilot. Both of these instances are about $30 a license per user per month, and we’ve had a play with DeWitt and we’re looking forward to having a play with copilot. Llama too long is bit of an interesting one at the moment. So it’s been released by Meta as llama two model. Llama too long has been announced, but it’s not formally been released. It looks like that they’re going to be keeping that in house to power their own AI side of things, but there are llama models available with 128k context that have been released on hugging face. Now this may sound technical, but from your perspective, the more context or the longer context you can have, the more information that a commercial AI model can consider in one time. So imagine it like a desktop.

Andrew Herbert [00:15:03]:
You place down papers on the desktop, and the larger desktop you have, the more papers you can place down and you can analyze that information at once. So if you have a very small desktop, you can only maybe place four pages down, and that’s all you can look at at one stage. Whereas obviously the larger context, the more pieces of paper down, the more you can see and the more the AI model can understand and provide you feedback with. So the longer context is really important. That’s really going to allow longer analysis and allow more higher quality responses. Zapier is a workflow automation tool. They’ve just released their beta of their AI workflow generator. So you don’t really need to be able to have an understanding of how things work as such.

Andrew Herbert [00:15:49]:
And you can just type in, I want all of my emails to be categorized and then tasks created in our task management platform like Trello or Monday, and if you just write that in as text and go, it will generate a framework of that workflow. So really powerful stuff in that sense. And then canva tools. So from an image generation perspective and a presentation PDF outputs this is a really big takeaway, which is they’ve been implementing AI since Dali two came along for a while now. So they’ve had it inside their organization, their ecosystem for a while, but their most recently is grab edit arrays and expand elements when Dali three came out. So what this means is that you can have an image and you can grab an element of the image and pull it out of the image as a selected item. So if you had a landscape with some people in the front, you can grab those people out of the image and pull them out and just place them into a new environment. So pretty powerful stuff.

Andrew Herbert [00:16:52]:
You can also background arrays, which is similar kind of thing. You can also expand. So if you have a small image that doesn’t quite fit your layout, you can actually ask it to expand that image out into the larger area and fill that space. Now I know it’s very graphic design side of things, but from our perspective of content creation, of document creation, and of working in multiple environments and multiple operations inside a business, really useful to look at that. They’ve also got this magic switch function at the moment, which allows you to very quickly reformat your presentation into various sizes and functions. But it’s also got a translate function. So the translate function is pretty cool. It enables you to take a presentation and at the click of a button translate it into any language that they have on file.

Andrew Herbert [00:17:47]:
So for instance, this page, one click translated it over into Spanish. So really powerful function. If you’ve got a multinational audience, if you’re say presenting webinars on AI, it’s very handy to be able to provide a presentation handout in the native language of the people there, instead of having to try in English. So rolling into some use cases. So use cases we’ve got here, you can see on the page there, top left, email management using Zapier. So this is a pretty straightforward use case that will filter, sort of analyze the sentiment of an email that’s coming in. It will then summarize that email down into something really quite succinct. Then it’ll then go and push that into a slack environment, which is our chosen chat framework for our internal team.

Andrew Herbert [00:18:38]:
And then it will go out and create a filter to see if any of the email then needs to be cast out into Trello. Now you can do this inside of Microsoft Automate as well using a GBT call or an OpenAI call. And in Microsoft using automate and GBT, you can actually keep it data sovereign so you don’t have to be concerned around your data a going anywhere because it’s running through an API and then b also being processed offshore. So from a compliance perspective, using Microsoft Automate to create an email management workflow like I’ve displayed there in Zapier, it’s probably the best case solution. Rolling across these are some more customized outcomes. So a marketing database. This is a machine learning pipe that targets 7000 websites to monitor the marketplace and also capture contact information. And then from there it enables bulk sending of SMS and emails in line with opt in policies and so on.

Andrew Herbert [00:19:37]:
The competitor monitoring similar kind of concept, providing market intelligence and product alerts. So looking at these use cases, but looking at market intelligence around pricing, product pricing. So looking at your own product pricing against your competitors product pricing and providing your difference and seeing exactly who the leader is in relation to that. And also product alerts around whether or not a competitor has released a new product into their online portal. So that’ll give you a heads up as to whether or not you need to take actions around that. Chat bots. This is our own busy chat chatbot that is deployed on our website. It is an AI powered chatbot, takes about three minutes to deploy.

Andrew Herbert [00:20:24]:
There are a variety of options out there like this now, and Microsoft and Google have also rolled out their own variations of it. You can then also deploy a chatbot internally so you can put it out from an operational perspective. You can give it to your operations team and load all of your operations manuals and workflows into the chatbot so that people can query it about how to do their job or if they’ve got any questions. This is a massive time saver potentially on the operations manager and the disruption that these just quick queries about. Where do I find this? What do I do here that’ll save her time during or him time during the course of the day legislation wise, like maybe mentioned, I spent some time in defense and working with Department of Veterans affairs. It’s quite a cumbersome or a complex process because there’s so much information out there from a variety of sources and which legislation applies. So this is a chatbot that we built in order to support that and support veterans being able to get correct information. And so this compiles all of the DVA website and also holds all of the sops, which are statements of principles which are based around injuries or defense members may have had during their time in service.

Andrew Herbert [00:21:44]:
So a couple of really functional use cases that will make a sizable difference or just create a better quality output for team members rolling forward into some other use cases. So we’ve got here some content creation example, which is what I just went through before reviewing a sentiment. So getting feedback from either product reviews, whether you’ve got something on Amazon, or whether you’ve getting Google reviews, wherever that sits. I’m trying to think of all the other platforms that take reviews at the moment. There are hundreds out there, but you can get that review and you can automate it through automate or another platform, zapier make and push it through. And you can see there that the review has been generated and sentiment analysis on this as to whether or not this is a positive or a negative. So if you’re looking at an NPS score, you can try and translate that review into an NPS score as well. So you can get an understanding of where you sit and kind of feed that into the metrics of your organization.

Andrew Herbert [00:22:46]:
B two B automation. So these are a couple of publicly available SaaS solutions at the moment. So Apollo IO is a massive communications database. This got 247,000,000 contacts in it that you can go through and search quite powerfully and refine that down to some, by company, by industry, by role, by size of business, expected revenue. Like there’s a variety of filters you can apply on that to create yourself a data set. The other option, as well as dripify, which is our preferred choice at the moment, is that we’re running through this is an automation program that allows you to expand your network. Very powerful, very powerful functionality. And these are both examples of emerging technologies or maybe smart uses of AI.

Andrew Herbert [00:23:40]:
It’s not really as blatant as something as a GPT where it’s content generation, but inside of these have also got their own tools that are applying limitations and restrictions to ensure that your account doesn’t breach any form of user agreements and so on on platforms. So they’re using the AI in a compliance mechanism rather than actually a content generation. And you can see here some custom uis as well, that you can easily go out and secure and generate. So, social content. And the reason why we look at UIS and these are relatively straightforward for businesses to be able to go out and get made for themselves. The reason we do this is because it creates a framework for users of anyone in the organization to standardize and systemize the quality of output. So whilst top P and temperature may not, or token length may not be something that the average admin, ops, sales agent, whoever it is in your organization, can understand and comprehend, word length, variety and creativity is something that’s a little bit more easy to understand and apply ranges to. So by creating a user interface that translates the technical language into more common language like keywords and title, it standardizes the prompt that is given over into a platform like OpenAI and then to be able to get that back.

Andrew Herbert [00:24:58]:
So it means that you can then not have to sit there and go through the quality control that you would see in the OpenAI open prompt that you saw. And as well as that using a custom UI being pushed through the OpenAI API means that all the back and forth that you provide or you have there actually isn’t held by OpenAI for further training. There’s a bit. So that’s our use cases. I know it’s a bit of a fire hose there. So rolling into a discussion, is there any queries, questions, thoughts, challenges, anything that anyone who’s like to raise, put in chat.

Ben Love [00:25:39]:
Thanks Andrew. I’ve got so many of my own. I mean, we’re already using AI a fair bit within grassroots it. We’ve played around with some fun things. It’s been very helpful on occasion. I remember one particular afternoon, we needed some fairly quick understanding of a particular client’s workload patterns on our help desk. So what sort of workload they were putting onto our help desk? Honestly, within about 20 minutes I was able to pull some information out of our ticketing system, upload it into Chat GPT, and suitably anonymized, but uploaded into Chat GPT and it came back with data visualization showing me graphs and plots and trends of how this client was interacting with our support team. Now that would have taken me hours to do manually Chat GPT did it in a matter of seconds.

Ben Love [00:26:33]:
So hugely useful for grassroots it. We have had some great questions come through from a couple of people, which I’ll raise now, but to everybody else, if you have any questions or comments, please drop them in the chat there and we will get to you. So the first question we’ve got is from Colin. Colin is asking a question. What platforms or tools do you recommend to help with a monthly newsletter, content generation? Now, I’m happy to share what we do in grassroots it, but Andrew, I’d love to hear your.

Andrew Herbert [00:27:08]:
Mean the thanks for catching me off on the back foot here. So just creating on the spot look from a content generation around monthly newsletters, you could definitely go down the pathway that you saw before with using ChatGBT’s open the browse with bing function to be able to get those regular news generations so that you can create. And I’m thinking education around the industry specific of what’s going on, to be able to provide that out to your target audience because education is key in order to gain that traction and create that trusted advisor role. So getting that education and that news and going and using Chatwt to be able to grab those daily articles and create those LinkedIn posts, you could then push that post over into an excel file and the Excel file grabing that and giving it back into chatgbt to then draft up and generate that monthly newsletter content for you pretty quickly and easily, unless there’s something more commercial in confidence around it. But using that, and you can build that inside Microsoft, automate fairly straightforward or zapier something along those lines. The only other thing that I’m thinking is that in copilot you can go out and it will generate Microsoft slides, sorry, PowerPoint presentation from a word document. So you could have that layout pushed across and generated, and then you might be able to actually have that output into HTML, which you’d be able to load into Mailchimp or something similar from a format and a layout perspective. But you’d also be able to have that done in GPT four as a separate line of code on the addition of that content creation.

Andrew Herbert [00:28:52]:
So I’d have it take all my articles that I’ve generated over the course of the month from my Excel, drop it into GPT, give it an instruction to create a monthly summary of this, give it words, maximum target audience and so on. It’d generate the actual content and then an additional line item. There it’d be. Now put this in code as a layout. Our colors are this and define your colors associated with it. That would generate you that output that you’d then be able to just copy and paste directly into Mailchimp or another emailing platform to be able to get it out.

Ben Love [00:29:25]:
No, that’s useful. Thanks Andrew. I know at grassroots it we do use AI absolutely to support our marketing efforts. So probably if I think about the use case that you put forward there, Colin, the first bit would be, I guess for me would be getting topics to write about. Now you are going to be the subject matter expert in your field and your audience wants to hear from you. So ideally, if you’re staying in touch with your industry and you know what the current affairs are or what your audience likes to be educated on or to read about. But sometimes we’re so deep in our own industry that it’s hard to actually step back and get a bit of inspiration for what to talk about in this content I do actually find that Chat GPT can be useful for that. Now, Andrew’s given us some great demonstrations on how to use Chat GPT for that purpose, to maybe pull down today’s top headlines in a particular industry or what have you.

Ben Love [00:30:31]:
I know I also subscribe to various email newsletters or RSS feeds and other people’s blogs, various ways of just staying in touch with the industry news. So that’s the first place that I go to to actually get my ideas for the content. I’ll often then turn to Chat GPT to help me structure the piece that I’m writing. So for example, if I’m writing a blog post, I might give Chat GPT a prompt that says something like help me structure the outline for a blog post on blah blah blah. And then Chat GPT might come back and just help build that framework, maybe of headings and subheadings and a couple of ideas there that I haven’t thought of before that might go into that blog post. And the same approach could apply to your email newsletter content. I will then. If I’m then going out and actually building out and writing the longer form content, the sentences, the paragraphs, et cetera, the big lesson I’ve learned using AI here is don’t actually just let AI do it all for you, right? You use AI to support your writing, but if you let it simply do the writing on its own, if you say Chat GPT, write me a 500 word email newsletter, you will end up with a 500 word email newsletter.

Ben Love [00:31:47]:
But it’s not necessarily going to be the best, right? So you can use it as a guide and a prompt and maybe if you get a little bit of writer’s block at various times, it can be useful. I find that another AI tool, which is called Jasper, Jasper AI or something, is the website I find is quite good for that. I’m quite used to that. I know there is a competitor, that, which is called copy AI. I can’t comment on that tool, I haven’t used it myself. But I’ve found that those things can help the other way that we’ve also used Chat GPT in a similar scenario there, Colin, is to help it, sorry, is to use it to help build out a calendar of content over the next 30, 60, 90 days, whatever you want. So the prompt that you might feed Chat GPT in that scenario is please help me build out a content calendar for my email newsletter for the next 30 days. My target audience is blah, blah blah.

Ben Love [00:32:48]:
And the things that I generally write about are blah, blah blah. And you could even then say to Chat GPT, please format the output in a table and it will actually build you a nice little table on the screen there with ideas for the content and topics and stuff that you could touch on in each of your email newsletters over that month or so. So I hope that helps. Sorry, big rambling one there. But of course, happy to go into more detail in another forum if that’s of any use.

Andrew Herbert [00:33:18]:
Definitely looking at Chadwt to go and generate the strategy and I was just eyes off the side there because I was just trying to pull up there’s a platform that goes out and looks at layout side of things and the actual generation of the monthly layout that you could use. There are a couple of variations on actually what the framework is you’re going to put it into. So canva’s got an AI generation option there as well that’ll assist you in selecting the right template or actually creating it for itself. And I will make sure that I pull up the name of the platform and include that in the email that goes out post this, because there is a slide deck email framework generator that you can just basically provide the description along, give it to you as well. So not only from a content generation and a strategy generation, you can go out and create that layout and look. And I want to echo what Ben’s saying there, which is view AI as the knowledgeable, enthusiastic junior. You will always have to overlook the content, you will always have to do quality control, you will always have to do minor tweaking, but it takes you from having to do 100% of the work and a creation role into a reviewer role, which is only potentially 10% of the workload. So yeah, definitely reiterate what Ben’s saying there around that overview that you’ve got to apply.

Andrew Herbert [00:34:38]:
And the reason of that is because AI takes the instruction that it’s given and it does what it’s told. There’s a whole bunch of unspoken instructions that you have as a user that you haven’t necessarily articulated and applied and given the instructions over the user. So it just does what it’s told. So you’ve just got to be conscious that those little unspoken elements that you haven’t prompted it around you now come into the second round and you can do that with an adjustment tweak or a minor prompt to reiterate or to adjust like I did with that demonstration around creating something more photorealistic. I hadn’t said it and I wanted it more photorealistic. So it’s really important for you to be able to shape that in the second and third round of content generation.

Ben Love [00:35:26]:
Colin, just quickly, before we move on to the next question, Colin, you mentioned just in the chat there that you are an AI virgin, so still getting your head around the whole thing. We actually put up a post recently just with getting some real getting started tips on AI, including some example Chat GPT prompts. Just to give you a bit of a kickstart there. I’ve just put the link into the chat there just in case that is of any use for you. All right now, Ken, thank you for your question. Ken. Ken had an interesting question here that I have no idea about. So, Andrew, I’m hoping you can help using Dali three, does it alter the original image? I.

Ben Love [00:36:09]:
E. If you remove people from a landscape image to copy to another document, does it change the original image and remove the people permanently or keep the original?

Andrew Herbert [00:36:19]:
The same depends on the framework that you’re actually working in. So you can use Dali three inside of chatwt, and when it doesn’t have an input option into Dali three, I believe it’s only got an output option. So the initial creation that you have inside of Dali three in Chatwt, it’ll always be there in a part of the history. So if you then work down through the iterations and you make those adjustments to remove people out of the landscape, you can always wind back. If you go into something like canva. When you import an image into canva, which is where you can grab those people in their grab mode and then drag it across, it will actually separate them completely and you will theoretically lose the original image. However, when you imported it, it will hold the original import in history. And so you can always just drop that straight back into the editing page again at any point in time.

Ben Love [00:37:17]:
Fantastic. Ken, I hope that answers your question. Like I said, I know almost nothing about image editing and AI imagery is not my superpower, so I hope that made sense to you. Now, we had a comment from Cassandra. Cassandra, I can’t tell whether that’s a question you’re looking for help in generating more spam or not, or I’ll assume not. I’ll assume that that’s not. And move on to Jessica. Jessica’s written here, chatbots for business.

Ben Love [00:37:48]:
Too many options. What do we do? I think that’s a fantastic. I mean, I’ve seen chatbots deployed internally within a business. I’ve seen them externally facing to clients or customers. I’ve seen them in so many different contexts and there are so many different tools, and some of the tools are almost free, and some of the tools are extremely pricey. So I mean, if we are interested in the chat bots, how do we get started? Andrew?

Andrew Herbert [00:38:15]:
Yeah, look, it’s kind of like a car. There’s a range of options out there for you, and you can choose the $10,000 entry level model with the cloth seats, or you can spend the $100,000 on the luxury vehicle. So it’s about fit for purpose. I say that because looking under the hood of how these things operate, they use what’s called a vector database, which basically means it stores all the content we give it in a database, and then when a user queries or writes a question in, it then goes into that database and uses a search function to find a chunk of data or material that is most suitable for it or most fitting for that actual query. And then it takes it over to the AI model, and then it asks the AI model, here’s the question, here’s the chunk of data. Can you restructure this response in natural language? And then it feeds that back to the users. There’s a bit of a triangle there. The thing that you’ve got to be wary of is the size or the structure of the vector database.

Andrew Herbert [00:39:23]:
So if the vector database, they can do a couple of things to create a cheaper version. What they can do is they can reduce the chunk size. So think of chunk, a data chunk, as to whether or not they’re saying that each word is a chunk, or each sentence is a chunk, or each paragraph is a chunk or a page is a chunk. So you can imagine that if you ask something about a business and it’s only got the words set as a chunk, then the response is going to be really poor quality. The sentence would give you a better quality output, and the paragraph would probably the best balance of being able to provide a quality response to the user’s query. So that’s first thing. So that’s the biggest difference amongst them all is the size of the chunks. And the second one as well is whether or not it carries the conversation.

Andrew Herbert [00:40:14]:
So what I mean by that is that if somebody just happens to walk into a conversation and just say, who likes oranges? There’s a bit of a taken aback. And people potentially go, I don’t know what you’re talking about. And so the carrying of the conversation is about providing that background context of what’s already been said and leading up to that query, you’re going to ask at the moment. So other cheaper entry level chat bots will go out and we’ll say, we don’t carry the conversation. So it’s called one shot. So if you ask a question about a business or a question about a company and say, for instance, what are the products on offer? And it goes in, queries the chunks, restructures it and gives it back, and then the user goes, what’s the price of that? It doesn’t know what the user is referring to. And so there is no context about what the price is. So it won’t be like a robust conversation.

Andrew Herbert [00:41:12]:
So they’re the two main things you can look at from a quality perspective. The other options you have around is about ease of deployment, flexibility and customization. Look, of course, I can talk all day on this, and I’ll lean towards what we’ve built, because we considered all of that when we were building it. And we believe in quality, not quantity. So you can look at something like that. You can also go into Microsoft. They have their own product there, and as do a lot of the others, it’s about ease of deployment, flexibility and quality of output. They would be the main three things I’d look for in a chatbot.

Ben Love [00:41:49]:
So Andrew, if somebody was looking to dip their toe into the waters of a business chatbot for themselves, where would they start? Where could they go this afternoon without fear of expense and commitment and all that sort of stuff, and just start to dip their toes in the water?

Andrew Herbert [00:42:11]:
Look, busy chat is our chat bot that we build. So busi chat, I’ll just put it in the meeting chat. Now, if you go there, there’s a seven day free trial on any of the chat bots, so you can deploy it into your business. Takes about three minutes. Signing up is pretty much instantaneous. And then from there, you just enter your website and hit go, and then take the short code and install it onto your website. And that’s it, it’s done. And then from there, you can update the back end and give it more information and more knowledge, and you can tweak it and play with it.

Andrew Herbert [00:42:47]:
So busy chat with the platform that I’d recommend.

Ben Love [00:42:51]:
Fantastic. Thank you. All right, we’re getting some good questions here, so excuse me if I keep us moving through these, Andrew, I don’t want to keep everybody past the hour that we’ve committed, and we’ve got a string of questions that keep coming in. Jamie has asked about Microsoft Copilot. How do you suggest we learn about the functionality of copilot?

Andrew Herbert [00:43:11]:
Look, it’s only in beta or early access program at the moment. So the first thing I’d do is jump online and start reading up on the feature articles and the releases coming out of Microsoft. Reach out to Ben. I understand that you guys will have some product in relation to that. It is definitely coming out from an enterprise level in about four or five days time. So there is a lot of articles out about news about what it can do. And the second thing I do is go and look at some of the YouTube user reviews and they’ve got some people that are test driving it and seeing what it can do and you can just go and watch that online when it comes out in under a week. I’d say go and buy a seat, spend the $30 for the month and have a test, have a play, create that playground environment.

Andrew Herbert [00:44:01]:
Go and get those files that you’re happy theoretically to not be too concerned about commercial and confidence and just even create a new user account just to test with it and see whether or not it works for.

Ben Love [00:44:14]:
Yeah, Jamie, I think Microsoft Copilot is going to be quite a game changer for a lot of us who are on the Microsoft platform. I know I have to create PowerPoint decks on a semi regular basis and I’m not great at it. I’m really looking forward to copilot doing that. For me, that’s just one of those little use cases there all the way through to power bi data analytics and visualization and I mean, you name it. Copilot is just going to be amazing because it has access to all of your knowledge that’s already within your Microsoft three six five tenant. So it has access to that knowledge, which is the exciting bit. Chach Ept doesn’t know anything about you or your business. Microsoft Copilot does.

Ben Love [00:44:58]:
So how to get started with that? Look, I love what Andrew said there. Get a license and just dive on in. I think it’s $30 a month or something for a license, even if you get a license for one month and have a play. But as soon as we can get our hands on it at grassroots it, we will 100% be getting into it. Hopefully that’s pretty soon. Cassandra, I love your comment. AI isn’t going to replace your job, but somebody using AI will. This is a message that I think should be coming through loud and clear.

Ben Love [00:45:29]:
If you are not using AI in your business at the moment, your competitors are okay and it is changing business operations significantly. So there’s still time, but don’t think that you won’t need to integrate AI into your business. All right, scrolling down, I’d echo that.

Andrew Herbert [00:45:47]:
And I would say even to the point that it’s going to be trying to use a fax machine against email or email marketing, it just won’t be in a different league and borderline irresponsible not to look at how to adapt it into a business.

Ben Love [00:46:00]:
Yeah, 100%. Jose, is there any AI tool that can assist our marketing manager and LinkedIn outreach specialist?

Andrew Herbert [00:46:10]:
Okay, so I covered off on dripify. That is definitely something that I’d go and look at. Just be conscious that from a LinkedIn perspective, automation is a potential breach of their terms. So anything you do comes and carries a small amount of risk with it. But if you stay within constraints and don’t spam everyone constantly, then usually there’s absolutely no concerns. But drippify is something that I use. I like it enables automated sequencing and messaging, and I’ve had a solid amount of success from so I can link into there. One of the key parts of drippify that I like is their support.

Andrew Herbert [00:46:55]:
When it’s queries, I’ve had responses within two or three minutes any time of the day. So that’s been fantastic. Support has been always able to, that’s enabled me to get up and running very quickly.

Ben Love [00:47:09]:
Fantastic. Thanks mate. Steve, good morning, how are you? Steve, workflows. Any recommendation of workflow software that can manage our sales order processing that goes through every step of our process, from inputting the order and like 20 steps to completion and invoicing. I love this question, by the way. I’m looking for workflow software. However, it would be great to have one that has some AI to help analysis KPIs, reminders, data queries, et cetera. Might be too big of a question.

Ben Love [00:47:39]:
No, I think it’s an absolute awesome question and I’m excited to talk about it.

Andrew Herbert [00:47:48]:
It may sound funny, but this is what we do at remap. We look at workflows and we build solutions for it. So if there was a software package out there that could do our job as succinctly as that, I would be all over it. Having said that, what we look at is we look at, first off, mapping the process. So we use a platform called Scribe to be able to do that. And for more detailed mapping, we use loom. Both of those are screen capture platforms. Scribe does any click and any interaction, and does an automated description generated with screen snims and loom is a video recording of the user going through the flow.

Andrew Herbert [00:48:26]:
So first off I would do scribe to capture the workflow and then from there look at analyzing the number of clicks and the number of interactions the user or that the staff member is having, and then you’ve got to look at creating a workflow itself. So a step by step breakdown of actually what they’re meant to be doing and then analyzing that workflow to then create the efficiencies out of it. Unfortunately, I’m not quite sure if something that’s out there right now that does what we do like that because we use a variety of softwares and we have an on team business analyst that actually looks at the workflow and looks at the emerging technologies available and looks at efficiency solutions and smashes that all together to create an output. So yeah, it’s a complex piece, but I would definitely look at scribe as a first start.

Ben Love [00:49:14]:
Fantastic. Look Steve, what I can probably add to that conversation in there is that Microsoft Power automate is the first thing that leaps to mind as being useful for that. Power automate has functionality in there which they call goodness process mining I think it is, which actually helps to review and analyze a process that you have and look at opportunities for automation power automate will drive a lot of the automation. If you’re familiar with Zapier, power automate does the same. So we’ll drive a lot of the automation between one or more tools, whether they be Microsoft native tools or non Microsoft native tools. You can integrate AI into power automate very easily, so it can do things like document recognition. So you can simply provide it with an invoice, for example, which has been sent to you from a supplier. Power automate with the Microsoft Azure AI document recognition functionality can look at that document, pull all the relevant information out of it, and then put that wherever that information needs to go in your billing system, into Xero, into your ERP or your purchase orders or what have you.

Ben Love [00:50:36]:
Power automate also has functionality which is referred to as robotic process automation. So RPA is useful for when you want to automate other software products that don’t have nice APIs and nice modern integration points. So robotic process automation will kind of literally reach out and click the key on the keyboard for you. If that’s how you need to interact with some legacy software or tools or platforms that you might need, there is a body of work there to actually make it happen. As Andrew said, there’s not a tool that is magically going to do that for you. But in terms of the actual automation engine, to let all that happen, I’d be putting my hand up for Microsoft Power automate, to be honest.

Andrew Herbert [00:51:28]:
And to add to that, Microsoft’s OCR. So optical character recognition that’s inside their document scanning is top of class. We’ve tested a variety of platforms out there from all the major organizations and Microsoft just keeps coming up as number one. So sitting that inside your automate workflow combined with what I touched on briefly before about legislative compliance. So if you’re looking at any of the protected sectors at the moment that are being covered by any of the legislation around patient or client confidentiality, around material or data handling, Microsoft and the use of automate inside of Microsoft is the go to solution. It’s the only one that we’ve been able to find that would be compliant. Looking forward as well, because remember, building your solution isn’t just about now, it’s about trying to build something or looking to build something that’s robust for the next 18 months at least.

Ben Love [00:52:20]:
Justin, business chat bots sound like a great solution to overlay on an internal intranet to assist employees. I 100% agree. That’s actually one of the use cases that excites me the most is bringing some sort of an intelligent AI chat bot to bear on our internal knowledge to really surface that information for our staff and our teams internally, I think a lot of us have got a huge amount of knowledge of documentation, what have you within our business, but a lot of it just sits away in a dark hole and when someone needs to go and find it, there’s a process there for them to actually find it and search all search engines and what have you will only get us so far. I love the idea of having that internal looking chat bot there. Cassandra. Sorry Andrew flying through it here, mate, we’re running out of time.

Andrew Herbert [00:53:12]:
I’ve just realized you’ve been scrolling down in front of me and I was like, he’s given these really great, really good answers. How’s he? Oh, he’s going down the chat. Oh, got it, yeah.

Ben Love [00:53:23]:
Cassandra, will we be running sessions on Microsoft Copilot? 100%. I think that’s going to be a really exciting area. So as soon as we get our hands on it and can actually get in there and have something to show you, we will be doing. So scrolling through. Cassandra. Yes, Cassandra, in your space there with bookkeeping and accounting, I think there are some amazing off the shelf products there.

Andrew Herbert [00:53:53]:
And touching back on Steve Clark’s message about loving to keep it in three, six, five. Look, you know, we’ve reiterated that and automate also accesses externally, so that’s the best in class solution at the moment.

Ben Love [00:54:05]:
Fantastic. Justin, does document recognition replace OCR technology?

Andrew Herbert [00:54:11]:
Andrew, so OCR is optical character recognition. It is essentially document recognition. If you look at chat, GBT’s four advanced document analytics that we didn’t cover off on today, but I did touch on, that’s using a form of OCR to be able to read the actual flat document. So in a sense, they are one and the same in that broader concept before digging to deeper technology. But for the business use, it’s all the same stuff.

Ben Love [00:54:40]:
And here is the perfect segue into our exit into wrapping up for the day. Thank you. Cassandra. Where do you see AI in 18 months to two years? Seems to be at its infancy at the moment. However, with the speed of adaptation, its potential is overwhelming. Andrew, closing remarks.

Andrew Herbert [00:55:00]:
Wow. Okay, so I want to look back to January of this year and then fast forward to about June. July. So in January of this year, generative AI wasn’t even on anyone’s radar. And then fast forward to June, and it is everything that everyone’s talking about. In August, statistics were released that 80% of research was in generative AI at that point in time. So there is this massive adoption curve and a massive emerging technology evolution, and all of a sudden that hits us at once. So what we’re not seeing is the momentum building up in the background that’s quietly done in labs and quietly done by other organizations until it suddenly breaks on us.

Andrew Herbert [00:55:44]:
If you rewind twelve months, even 18 months, the whole concept of AI within business wasn’t even a thing, except for those of us that were working in the IT space and for those that in corporate that had large buckets of money to be able to throw at it. Now it’s available to small to medium enterprise for as little as $30 a month. Like, it’s massively powerful. So the speed of adaption and its potential is absolutely huge, and you do need to keep across it. And it’s one of the reasons why I love what I do, because I’m across the emerging technologies and constantly learning about it and constantly looking at those uses. Is it overwhelming? If you don’t take a concept of 1% incremental change and adaptability, it will overwhelm you. There is a huge wave that’s coming that in 18 months to two years time, AI will be as functional, as I said before, as email inside of an organization, it will be integral into the way you operate. And if you are not operating with it, you will be living in the stone ages, rolling out so many different examples of imagine being a business without a website, imagine being a business without an email, imagine being business without a phone number, like, how exactly would you operate? These are massive, massive changes and leaps forward.

Andrew Herbert [00:57:02]:
And I think that at the moment we’re seeing exactly the same. Again, where exactly do we see it? Generative AI is huge at the moment. That is going to see content generation and so many impacts in business. I think that there are amazing technology advancements coming from a vision perspective, and we’ll see leaps and bounds around that. And then the biggest impact, I think, is going to be deployable models. So own models, Zenml, which is the news article that I was reviewing as part of the demonstration, they’re deploying short iterations to be able to provide your own data onto so self training models that you can control and you can deploy. And I think that’s really where the value is going to sit, is about being able to control your information and then fuse it with an AI to be able to get the most value out of it. And that’s the only real way that you’re going to be able to get that longer term, broader context window or analysis.

Andrew Herbert [00:58:00]:
I think that’s the direction that’s going in, and that is going to be absolutely amazing for business that want to take that on.

Ben Love [00:58:07]:
Thank you, Andrew. That was fantastic. Michael has just snuck in. The final question, is it likely to increase data breaches? I’m only letting this through because it talks about cybersecurity and this is one of our big things. Michael, is it likely to increase data breaches? I’m going to go out on a limb and say no. And the reason is because there is an arms race between the good guys and the bad guys. The good guys have been using AI to protect our businesses actually for quite a number of years now. The bad guys are only just getting started with AI now, right? They’re going to learn quickly, very quickly, because there’s a lot of money in it for them.

Ben Love [00:58:42]:
But the good guys and the bad guys are just going to continue the arms race. They’re both going to use AI. I don’t think it’s actually going to have a significant impact on where we.

Andrew Herbert [00:58:52]:
See data breaches, those that don’t adapt AI based cybersecurity measures. Yes, they will be. Absolutely.

Ben Love [00:59:00]:
To be fair.

Andrew Herbert [00:59:00]:
Yes, absolutely.

Ben Love [00:59:01]:
If you don’t keep up with that. Yes, you will be left.

Andrew Herbert [00:59:05]:
Yeah, it’s going to hurt.

Ben Love [00:59:06]:
Fantastic. Okay, the time is 11:00. Thank you for the 60 minutes that you have very kindly spent with us today. I do hope that we added some value. Please follow both Andrew and myself on LinkedIn and of course, Remap AI and grassrootsit on LinkedIn. We will be sharing more about AI, of course, as it all unfolds. Any other questions, please follow us up afterwards. Thank you.

Andrew Herbert [00:59:30]:
Thanks very much. All.