Talking Data and AI: Unlocking AI: The Curious Story of CAIT

TALKING DATA AND AI EP 1: with City Facilities Management

This webinar originally took place in 2025, all information was correct at the time of recording. 

The story of City Facilities Management's adoption of CAIT, where IT championed innovation and helped the business discover the potential of AI

CAIT, or City Artificial Intelligence Technology, is an AI platform designed to enhance various aspects of City's operations. CAIT is available in multiple variations to address specific use cases, applications, and customer or colleague needs. It includes CAIT Assistants, which provide a natural language chatbot experience for colleagues and customers; CAIT Agents, which automate elements of process workflows; and CAIT Analyser, an AI tool that analyses data, identifies exceptions, and understands trends.

Speakers: 

Andy Prêle, Business Systems Director.

Vidmantas Blazevicius, Lead Technical Architect.

Transcript

SPEAKER_00 0:00

Good afternoon, everyone. Welcome to today's webinar, the first of our talking data and AI webinar series. I'm your host today, Lyle Ritchie, Head of Talent Solutions here at Head Resorting. For those that don't know us, we are a technology, digital, and transformation recruitment company. We are delighted to have Vid, the lead tech architect, and Andy, Business Systems Director from City Facilities Management, who will be delivering today's webinar where they'll be talking about City FM's adoption of Kate, where IT championed innovation and helped the business discover the potential of AI. They will focus on the business side of the story and the technology journey they went on. So before I uh hand over, just a couple of housekeeping points to go through. The session will last approximately four to five minutes where Andy and Vid will deliver their talk and then we'll have a brief QA session. Please use the QA box to ask any questions and the thumbs up icon to upvote those questions that you're keen to hear uh the answers to. Please also feel free to get involved by using the chat function which should appear on the right hand side when selected. Okay, so uh enough of me. Over to Andy and Vid. Thank you.

SPEAKER_02 1:23

Hi, thanks, Cleo. Appreciate that. Um good afternoon, everybody. Um thanks for joining Vid and I on today's webinar, currently hosted by Head Resorting, uh, where we're going to share our story of unlocking AI within City. Uh first let me tell you a wee bit about City just to give you some context about the story we're going to take you through. Um we at City Facilities Management offer comprehensive integrated facilities management services across three key areas facilities management itself, technical bureau and energy management and construction services. Within facilities management, we offer both hard and soft FM services. Examples of hard FM encumbrance, maintenance of HVAC systems, refrigeration, electrical, building fabric, and water hygiene. And the soft services include cleaning, supply chain management, and risk access. On our technical and bureau and energy management, we offer remote alarm monitoring and event handling services, network infrastructure management, specification management, system access control and data maintenance. And then on our construction services, we provide special construction services across various disciplines like refrigeration, mechanical, electrical engineering, fire security, food services, and total resilience. Vid and I both work within the CETA's IT department where we develop and support our own in-house resilients that are designed to underpin how we deliver our services to our clients. And these include our AFRIM system, which is called Metri, and our AI solution, which is called Kate. Today Vid and I will share our AI journey with you with a focus on the technical, commercial, and business stakeholder experience of implementing AI as a concept and then as a solution. Our journey started 12 to 18 months ago, early in 2024, where we discussed the concept of AI as a potential solution to some of the ways of working improvement challenges that were put towards us. At that time, um AI was not embedded in any of our business solutions. Most of the user experience of AI was based on the consumer experiences where you have phone assistants and things like that. So our initial discussions with the business had led them to being initially skeptical of where AI could bring value. And at times it was seen as a bit of a gimmick from this side as to where that could really bring value. So from our side within IT, with myself totally, we took it upon ourselves to identify possible use cases and develop real-fast PLC to demonstrate the possible to our stakeholders. We started off with an AI assistant called Kate, which was talking to our core metric data source to enable users to ask Kate questions about maintenance activities using natural languages. A key consideration of us implement this from a business point of view was how would the users access tape? And we were keen to avoid application context switches. And to that end, we developed a Teams application, an embedded kit within the Teams application for users to access, both on their laptops, but more importantly for our field users on their mobile devices. Following the Kate assistant POC, we developed further POCs to spread the capability of our AI solutions. We demonstrated how we could use AI to interpret images to support parts identification. A key area of challenge for us to have our technicians take pictures to have Kate then identify that part to improve the turnaround of part delivery. We utilize Kate to read documentation. A lot of the time our partners are uploading documents to our systems of compliance documents, for example. We are using AI to read those documents to validate they are what they say they are, but also validate the information that's contained within those. And then finally, we have numerous SharePoints across our organization with numerous files of policy documents, etc. And we've used Kate to put a friendly, quick way of retrieving information, searching information in front of those SharePoints to improve the usability for our colleagues. So I guess the result of our PLCs were successful and importantly they demonstrated the value of AI to our business. It demonstrated that we could create AI solutions quickly and bring value to our ways of working by improvements in those ways of working. The trial of our Kate assistance showed that users could obtain responses to operational questions about maintenance activities quicker than using the MEFI application itself or reporting dashboards that are in place today. And importantly, they could do it on the go on their mobile devices. Users in the trial were providing positive feedback, such as this is a game changer, Kate has helped me when I'm out and about. I'll always use Kate first before opening my laptop. So it was really positively received by our colleagues. The scanning of documents PLC that we did to read the information on it evidenced and provided that we could automate low-value repetitive tasks with consistent outcomes, and leads and that led to further use cases being identified within the business of that type of technology. And following the PLCs, ultimately the business stakeholders started to see the value of AI as an enabler to improve our ways of working, which led to a desire for the business to explore further opportunities within city. Now that the business stakeholders had understood that of the possible and saw the value in AI, they have now taken ownership of leading workshops to identify and prioritise the value, valuable use cases. That which is a 180 degree shift in ownership from IT to the business. Key business SMEs are now collaborating with SIT on a regular basis to identify to review use cases and review the results of POTs. And with COSFOL, push that the cost flow in terms of the boundaries that we can use this technology. The business is invested in expanding our initial our internal AI capabilities, enabling our development team to expedite the delivery of AI solutions into our ways of working, and also to stay up to date with the ever-changing emerging technologies. Almost now on a daily basis, we have we have colleagues asking could Kate assist with this problem, could Kate assist with that problem? It's now seen by our business as a strategic enabler for positive changes to our ways of working. So to recap, I guess, uh our journey over the last 12, 18 months. Um often designed those and spent six months creating POCs before we put put out to trial. Our Kate chatbot trial and lasting over three months. We had approximately 70 colleagues interact with that, asking six and a half thousand questions. We had feedback loops for the users to identify improvements and feedback on AI hallucinations that we've experienced in. During a 12-week trial, we almost improve Kate on a weekly basis. Um, and we're looking to identify issues and create new features. Our AI journey is ongoing and is a long way to go to address all the use cases that have now been and continue to be identified by our business, but it will empower our colleagues to embrace AI within their ways of working. Thank you. I'll pass over to Vid and let him share with you the technical aspects of our AI journey.

SPEAKER_01 9:13

Thanks, Andy. Um so when we originally started this journey back in February, the first thing that was actually difficult to do is to have some kind of common ground because there were quite a lot of misconceptions flying around of what AI and LLMs in particular actually do. And one that was the main one was that the expectation was that the language models actually learn as they go as opposed to just being pre-trained, which is kind of the definition of GPT. So when we did original scoping sessions and workshops of what we should do, uh I tried to help uh framing ideas that people had in terms of this concept of crossing the street. So when you are in the busy road, you take in some visual cues, you look at left and right, you maybe there is a weather condition or something, how uh how fast a car is coming on, uh, and stuff like that. And so basically, you take in some real-time context and structured inputs and very scattered ones, and you interpret them to ultimately make an assessment whether you should or you shouldn't cross the street. And this was a very good uh way for us to uh take in a business process and see if that analogy applies to that business process. So, is there a human being involved in our business process that basically just takes in a bunch of uh unstructured inputs from potentially various places and makes an assessment? So, this usually applies somewhere where there are approvals or checks in your business process being done by people because LLMs are great at taking you know 10 different unstructured input sources and saying summarizing them effectively or and assessing the sentiment and what those sources actually uh mean. So the typical architecture of our AI solution is quite simple. Uh, we have business application uh and some UI in it. That UI is underpinned by an API or a set of them, and depending on how monolithic your architecture is, you might have one business application or more. Uh so we then take those APIs and we put an MCP server on top of them, and that stands for Model Context Protocol. Uh, you can actually take uh an existing API uh and use a tool, there are quite a few available now that can take those and automatically generate you an MCP server. That is not something I would recommend you to do. Uh but you can use it to start and then amend what the output is of that automated solution. But generally, when we took an API and automated it into an MCP server, at least with what we have now to do that, it didn't actually work to then be useful to within an agent context. So once you have an MCP server, it is quite simple to then just take an agent, mount it onto that MCP server, add it to its basically toolkit. And and the last arrow is quite optional. You can go back and embed that agent within your UI, making some parts of the subsequent problems you might face easier, or you can choose to embed that agent somewhere else, like we did, like Andy mentioned, our Kate chatbot part at least is embedded in Microsoft Teams rather than our business application. So let's discuss a few common dilemmas and challenges that pretty much everyone will likely undergo within their AI journeys. And uh every one of these probably could deserve a webinar of their own. So if you want to discuss any of these in more detail, feel free to reach out to me in LinkedIn or an email or something like that. And to be fair, I don't have an exact blueprint on how to solve them, and it is very contextually dependent on where you're putting the solution. But I will discuss some aspects of how we approached uh these challenges. So the first one is authentication and authorization. So, like I mentioned earlier, we decided to uh put Kate in our uh Microsoft Teams. Now, Microsoft Teams, when you log in into this application, you log in via their Azure AD. Uh, and our business application actually had a decade-year-old custom authentication method, means that as far as our business application is concerned, uh your user logged in in Microsoft Team means nothing at all for our business application. So that poses a couple of challenges, primary, primarily the ones being on we want the users to only be able to do actions and access data that they are already configured and authorized to do so in our business application, but the two authentication methods don't play together at all. So to solve that, and this was actually almost majority of the effort of implementing that whole chatbot was to basically take uh and implement uh an exchange layer in the middle of the MCP server so that a token from Microsoft Teams can be exchanged to a token for uh our business application uh so that the authorization of the user can basically happen. Uh, the next one is how to rag properly. And RAG is the retrieval augmented generation where you uh enhance the response from an LLM by using uh data from your database or some other kind of knowledge base like SharePoint or blob storage or whatever. And I wouldn't be surprised if a rag term will soon be replaced with a rag where an A stands for agentic rather than augmented, uh because of the way uh that the rag needs to ultimately be solved. And uh the rag sounds very simple in a nutshell, and the problem of actually taking some keywords and searching an index database and getting a top K results uh based on your search that match your keyword uh most uh that have a closest match to your keyword has already been solved. However, with the AI in the mix, the additional problems have emerged, particularly on the queries that cannot be solved in a single shot. So if you have two documents in your knowledge base, and in order to answer a query, both of those documents need to be used, then you cannot just use a single shot to basically query both documents and uh produce a response via an LLM. So what we have done was we have built a custom orchestration layer uh between the agent and our index databases, which basically first lets the LLM to build itself a plan on how it's gonna address and uh segment your the user's query to then put uh keyword searches or multiple ones one after another into our index database. So this is what we call a multi-hop problem where you have to do multiple hops effectively to actually reach uh uh uh deserve uh the answer you want. So uh the next one is around the low code, and we actually have next to no, next to zero low code in production, but I want to be quite clear that low code does have its place, at least in my opinion, in terms in terms of AI journey, and being able to quickly uh go into something like Copilot Studio and then create a co-pilot for the RHR uh SharePoint, and then go into a workshop with the RHR, with that copilot that you just created in a couple hours and show them how it can work potentially is quite powerful. And we use that quite a lot in almost all our workshops because it lets people to actually see and kind of get an idea of how that would work. And from there on, if you downgrade it from low code to your code or however one you want to call it, it will most likely only get better. Um also we found that we try to get some of the solutions via low code journey only, and we found that it doesn't get you all the way. Something somewhere, especially if you have some kind of legacy application in the mix, will kind of prevent low-code to work. Um lastly, is which models should you use? So there is actually no exact answer to this question, and it ultimately depends on uh what's the context of your application. So sometimes we actually use four different models in uh our current applications in production at least. Uh we use Gemini 2.0 flash. Uh we don't use any thinking models yet. Uh and we use GPT for all and for all mini, and we use uh OpenAI 01 as well. Uh but ultimately you have to basically understand is thinking model fit for use and is the cost uh uh justifiable? Uh and thinking models, for example, are quite hard to justify and not just because of the cost, but because of the user experience we give you. So if a user query needs to be answered quite promptly, then you know waiting for a response for 45 minutes is is is not something that will be a good user experience. But at the same time, if a user query can be answered asynchronously and the results delivered via an email, or you are using a completely autonomous agent, then in cases like that, a thinking model can make more sense. Um so this is some of the toolkit we use. Uh, it's not all of it, but it's some of the ones that I thought might not be quite well known. So we have we started recently dabbling a bit in voice calls, and Bapi is the tool that I found to be by far the best in terms of providing a voice call experience via AI. So if you have a help desk or something like that, or anywhere where users answer calls or proactively make calls to maybe your customers to inform them of something, you can use that platform to automate that very easily. And the technology has gone quite far in order in it in terms of being able to do that. We use any 10 uh to basically visualize our workflows, uh, and I'll show a screenshot a little bit later of that. Uh, we use prompt flow uh and lang chain, uh, which are effectively interchangeable. And from flow is basically uh just an Azure version of Langchain. Um, and these are the ones that we use to build our orchestrations. So it's basically what happens between you submitting some kind of query and the agent actually calling the The underlying LLM model. And lastly, if you do dabble in voice calls and you need to customize or synthesize a voice, then we use 11 Labs for it. So I thought I'd mention this. So this is an example of the NA10 workflow. And I chose the screenshot just purely to showcase of the ease to basically communicate logic with your stakeholders, potentially our product owners, even between developers. So this is where using a low-code tool can actually help speed up your journey quite a lot, as opposed to needing right, uh needing to write like a custom code orchestration in cyber landchain or promo. And this is something that can be put together, you know, in hours basically. And maybe if you have something complex, it can take a day or so, but definitely something that can help you with uh fail-fast POCs at least. So a couple of things that we learned along the way. Uh so uh we want always try to engage with our stakeholders early. Uh we basically, if we envision a particular department or something to be the benefactors of our AI solution, then they'll be invited into our initial workshops events because uh ultimately they are the users of our product. So if we don't make it fit purpose for them, then what's the point of doing it in the first place? Uh to be able to start small and think big is quite a difficult task, actually. Uh we wanted to do something that uh basically builds some confidence. So we basically, whilst we have pretty much the entirety of our organization in mind, we just built a couple of POCs to build the confidence quickly. And those POCs were very low effort ones. Now, in in terms of adapting and evaluating your AI solutions, that is something that you probably have to do constantly. And to be honest with you, uh some of the things we have wrote last year are already obsolete, and we have rewritten them a couple of times since then. And I would not be surprised if some of the things that has been discussed today will be obsolete next year at this time as well, and either has to be rewritten or adapted somehow else because of the rate that technology evolves. We try to communicate our success stories quite often just to make sure that people are aware of what's working well for us and what isn't. Uh, we integrate the feedback loop everywhere we can. So this is quite important because if you don't know how well your AI solution is doing, then you're not going to be able to evaluate how much value it can potentially bring if you apply it somewhere else. So we integrate some kind of feedback loop everywhere, even in the autonomous agents. We try to re-evaluate them automatically as to what the user would do and how closely it matches to what an autonomous agent does. So I'll show you a couple of screenshots here of our uh solutions. And we chose not to do an actual demo because when we thought about what the demo will show to you, is some things that you have already seen somewhere else. And in itself, the AI was the least challenging part of the uh AI journey. The most challenging part was everything else around it. So all the dilemmas and challenges that I had discussed were the biggest issues for us. So this is our internal chatbot, and as you can see, it's embedded inside of our teams. And the biggest challenge here was to actually get the authentication and authorization working. So make that the user only can execute the actions that we are authorized in some other business application that we have done was the biggest challenge, and to make sure that we can always see the data that we are supposed to as well. The other one is something that shouldn't even have an AI, and we shouldn't even be able to show you what it does because it is a completely autonomous thing. Uh, we actually have built with have built this AI to be able to show other people what it's meant to do. Because if you just talk about it, it's quite difficult to visualize. But ultimately, this is something that takes in a compliance document and is able to validate that there is some data inside that document, and it's not just filled with picture of dogs or something like that. Uh so this is something that has been previously done by ultimately manual process by a human, and now it's completely automated and just tickle backs in in the background. The last one that I'll show you is around the procurement uh uh for our parts. So engineers go to jobs uh in city, and as part of the jobs, we sometimes need to procure parts. That procurement process was quite cumbersome before, and effectively what we have done here is we have implemented a Google lens, but within the context of city. Uh, so city partners with a bunch of suppliers to procure the parts from our from our engineers. So to make sure that the results of the solution were scoped to only the parts that our suppliers provide uh was the hardest part. And the Google Lens, it's the Google Lens part where you take an image and then you scroll uh down or you basically crop it to the part where the part is and then the AI identifies it was the easiest one. Uh so yeah, thank you. And uh we'll move on to some QA, I believe, right?

SPEAKER_00 28:30

Thank you so much, Andy and Vid. Really appreciate you sharing the story of Kate so far. I think there's a lot to take in there, lots of challenges, dilemmas that you've gone through, and I'm sure people on the call today can um take um some food for thought moving forward into their own businesses. So thanks again for that. Um a few questions in there, so please keep them coming. Um I want to kick off with a question for myself. So where with regards to the use cases, was there any discarded after POC?

SPEAKER_01 29:03

Oh, yeah. So actually, one of the use cases that we wanted to do was around uh uh applying uh AI to be able to generate uh powerful like Power BI reports. Uh and we actually have gotten that idea from the sales demos from Microsoft because we've shown we've seen where a user types in some query, and uh basically uh within Microsoft Fabric uh it just completely generates you an entire Power BI report. Now, when we tried it ourselves, we quickly realized that uh the reality was quite different for us, and uh we couldn't get even remotely close uh to what the sales demo was effectively. So ultimately, I think we spent a weekend and quickly discovered that idea. And even now we are still waiting for someone to come up with something that actually works in that area to be able to generate like analytical reports.

SPEAKER_00 30:09

Okay, yeah, so nothing's ever straightforward and changes. So being able to adapt to that is is important. Okay. Um so I'm just looking at some of the questions here. I think this will be uh uh a common question and one we can all relate to. How easy was it to get exec buy-in? Um or did you have the opposite problem where execs wanted AI but not the problem to solve?

SPEAKER_01 30:33

So I'll give that to Andrea.

SPEAKER_02 30:36

Thanks, Anthony. Um I think it was uh we we did have to get buy-in um when we started the journey and we put AI on the table as a potential solution. Um I mentioned earlier, it was it was very the the execs were very sceptical of where it could bring value because the only experience they had was consumer type applications where AI was summarizing reviews and things like that. Um so it did take a while, but the the POT approach is how we did it. Ultimately, that fail-fast POT delivering proof that we can use AI to do constructive uh improvements to our ways of working to bring value is ultimately what what got that over the line. And now the execs are button and at that level that Kate has been discussed as a part of our solution, our strategic enabler to how we deliver our solutions ultimately.

SPEAKER_00 31:26

Great stuff. Thanks for that, Andy. Um we've got another question from Adrian. So it's probably more for you, Andy. From a business perspective, how do you measure ROI and increased efficiency following implementation of K any kind of best practice practices for it at all?

SPEAKER_02 31:43

Well that I I guess at the moment our uh our production uh product that's in is is the AI chatbot. And and I guess when you when you look at the the ROI on that, when we when we ran the trial, it was more around how do we change people's behaviours, how do we get their engagement, and that is what we measured to look at are people engaging with the product, are they using the product? And ultimately every week we were measuring the number of questions being asked by every individual user, um comparing that against different areas in the business that were using it and different contracts, and and that's what proved to give the evidence to the business that actually the users are engaging with it. It's not just IT have come up with this concept and it's just nobody's really buying into it. And some of the feedback we've had has been pretty exceptional in terms of where our field teams are seeing it really bring value. And I think when your operators are uh are bringing, are seeing that value, I think for the exec team that's enough. Because that, like they said, the the cost track for implementing that is pretty low from my point of view. It's more around will the business use it and will they and if they use it, that's where we'll get the value in terms of their now getting time back. Because if you're a field manager and you get a call at the weekend saying there's a problem at a given store, what's happening? Normally they would have to go, I think you get back up the road to my laptop, open up my laptop, whereas now they can look at a mobile phone, open up Teams, and let's go, what's happening at the store A? And it's going, all right, wait a minute, I know what's going on. So it that's where we bring value. And it's hard to measure that, that, because it's those incremental. You save some time here, a minute here, a minute there. But on that example, that's that's where we bring value. On other examples, it is more tangible. Um on stuff that is in progress just now that's not yet implemented. We know that on the parts identification pieces for technicians, there's a percentage of the order requests that come in where the technicians have failed to identify the part correctly. And we know that AI will be able to enter the majority of that exception, if you like. And that exception has an incremental uh administration burden, if you like, that when a tech can't identify an incident process where our procurement teams have to work with our suppliers to pass information between them and the technicians to identify a part, and we'll effectively take that administration away and therefore improve the time it takes to identify parts, the time it takes to order time takes to other. So that becomes more quantifiable when we get to implementing that through the production side of that. So yeah, so it is it's it's not always there's a true ROI advocate why. Some of it is particularly with the eye piece and it's more on that incremental game where somebody's now getting X minutes back a day, and that X minutes over a week makes a difference over a month, does not wear it Yeah, absolutely.

SPEAKER_00 34:21

So that kind of snowball effect um over time is is definitely increased. Um good. Okay. Um and I know there was a lot in the presentation regarding kind of learnings and challenges. Uh if you could pick kind of one, what was the kind of most transformational AI learning that's made the biggest impact to City, would you say?

SPEAKER_01 34:43

So I I'll start on this one at least. Uh but absolutely the one that we have learned the hardest at least was that nothing the technology moves literally so fast that something that is not possible now will might be possible next month, right? It's that fast. So there were when we originally started thinking about being able to verify documents and their content with LLM, uh, I think it was GPT 3.5 Turbo at that point, and you couldn't do it. It was not, it it had maybe a 45% success rate. And within uh a couple months from us originally discussing it, GPT 4.0 came out, and then it went to maybe like 89% uh rate, and now it's almost at 100%, right? I the AI just very rarely gets it slightly incorrect, to which we just you know uh deal with it basically and go back and fall back to the original uh process. So ultimately, the the lesson we learned is you should always keep checking on what's going on in the market because there is so much money being poured into this space right now and across the world that it's just evolving way too fast. And you need to put serious effort to keep up with that.

SPEAKER_00 36:17

Yeah, no, absolutely. As you said, the money being spent um and the the speed of change is astronomical, so um keeping ahead of that. Um okay, uh quick question from David Richardson. What size was your dev team when you created the first POC bid?

SPEAKER_01 36:33

It's actually was zero developers, okay. Right. Yeah, so uh a fair question, well, uh zero developers. So I mean, I consider myself like a half-developer now. I used to be a developer some years ago, but we were able to create some of the POCs without any burden on top of our existing development teams because the tools are just so simple, right? Uh, and the POCs themselves are so simple to just showcase that you can submit. You actually can just buy uh a chat GPT subscription or anything like that, and just go and paste an image of your document into and say, you know, evaluate this for me, and then the rest of your solution will just be getting someone to automate that, right? But to actually get going, it's not uh doesn't cost any developers as far as I'm aware. Um it only will cost you once you want to actually implement something, and for that we used uh uh two to three developers mostly uh over the last year.

SPEAKER_00 37:45

Okay, I think that'll be music to execs uh around the world, so maybe not much of an upfront cost um spending on lots of developers to get things up and running. So um good stuff. Okay, and what's next then? Like what's the kind of the most exciting project that you have in your AI journey? Like what does the next kind of six months look like for you?

SPEAKER_01 38:09

Okay, I'll say one and then let Andy to come in as well. So for me personally, is uh the voice call uh opportunities. And I personally absolutely hate when you call someone and and when you get if you want this, please press one, if you want this, please press two. So being able to completely eradicate that experience is something that is like a personal goal for me within our business at least, right? Uh and uh I just think that there is so much potential in in terms of like transforming the voice call experience for your customers now that uh it's it's definitely something that is most exciting for me. Uh Andy, do you want to chip in?

SPEAKER_02 38:59

Yeah, thanks, Fed. Um I I guess the most transformational um piece that I see within our AI solution is in the parts space. And we've talked earlier about parts identification, but we've got a further improvement to that called parts availability. So between parts ID and parts availability, we're effectively a technicians scanning and taking a picture of a part they're looking for as identifying that part. And then the availability piece comes in when we start looking at what stockholders do we have across city? Do we have is it in your van? Do you have one? Is there one in a colleague's van that's five miles away? Is there one in a site van? We've got clients that let us hold stock at the locations. Have we got one there before I have to go to purchase? So again, using the I to be able to control part spend as much as part of education becomes really powerful for us. But I think that is a really exciting change that's coming um for us in the next six months.

SPEAKER_00 39:52

Good stuff. Thanks for that, guys, and and getting it from both perspectives there. Okay. Um I think that's it with regards to questions. If anyone's got any others, please put them in the chat. Um, but I just would like to, yeah, I think we're just gonna finish up there. That's all the questions. So big thank you um to Vid and Andy for presenting um uh the story of Kate today. I think a lot is learned from um the journey you guys have on, both from a business and technical perspective. And we do really appreciate you guys sharing um that information. So thanks a lot for that. Um so this is the first of our data webinar series um in Data and AI. Um we will be hosting our second one towards the start of July with Gary Crawford from uh Avondale Advisory. And this will be um based on when intelligence offer understands the user. So looking forward to that session. Um, and thanks to you all for joining. Thank you.