As telecom operators face increasing cost pressures and demand for faster service delivery.

and digitalization. Hello everyone and thank you for joining us for today's Fiber Broadband Association webinar. My name is David Norris. I am the webinar producer here at FBA. We appreciate you taking time to join us. And today's webinar is Build Right First Time and Ensure Fiber Data Consistency. And this is being brought to us today by IQGeo with a guest from Deepomatic. And so we're excited to hear from them. But before we do, I just have a couple of quick housekeeping items to go through. First of all, as attendees, you are in listen-only mode today. However, we do want to hear from you. We want to keep this as interactive as possible. So please do ask your questions. And you can do that by typing them into the questions tool, which is located in the lower left-hand side of your screen. We also have several resources available for you. You can find those in the resource tool, which is in the upper right. And we'll be referencing those throughout the presentation. So this will also be available on demand from our website, the FiberBroadband.org. And that will be up within 24 to 48 hours from the live broadcast today. So watch for that and feel free to share that with others. And finally, we do have a post-webinar survey. It's pretty brief and your input is very valuable for us. It helps us to shape these webinars to make sure that they're useful and valuable for you. So if you do have a moment to complete that survey, we'd definitely appreciate your input. And with that, let's go ahead and get started. So as I mentioned, we've got some folks from IQgeo and Deepomatic today. First, we have Stephan Schneider, who is the product manager at IQgeo. And he is being joined by Thomas Lullier, who is the general manager for Deepomatic North America. And with that, gentlemen, if you want to learn more about them and their backgrounds, you can find that information in the speaker tool on your screen there. But I will now step out of the way and hand it over to our presenters. So Stephan, go ahead and take over from here. Thanks so much, David. And thanks, everybody, for joining this webinar today. We wanted to talk about how IQgeo and Deepomatic can help you build right first time. It's a really interesting journey. So, you know, Deepomatic are not just esteemed guests. They're now part of the IQgeo family. And, you know, we can't wait to share with you all the interesting stuff that we can do with Deepomatic today. And some of the things that we're thinking about the future. So as a matter of introduction, we know where everybody is in terms of the industry. Right? We need to deliver services faster. We need to keep the cost under control. And we need to make sure that when you're doing work, your teams are being very efficient with the work that they do. Right? What we're trying to propose and what we're, you know, being building, it's, you know, methods to use computer vision to make sure that when work happens in the field, not only gets captured quickly, but also can be used to ensure quality and make sure that you're not doing repeats and rework. And with that, I'm going to hand it over to my colleague. So, you know, he can start talking a little bit about Deepomatic. Thanks. Thank you, Stefan. As Stefan just mentioned, very happy to have joined IQgeo family. It's been six days. But I wanted to give you a couple of numbers about Deepomatic because Deepomatic, it's been 11 years that we've been building on our key technology, which is a branch of artificial intelligence called computer vision, which basically allows us to train what we call AI models, specific systems to recognize visual patterns in photos and videos. And for 11 years, we've built our capacity to train those models very efficiently, quickly with little data and deploy them in the real world, meaning we process images and videos that are not just from digitized assets, but from photos you take outside. We've been working with a lot of different industries in the past, but for the four or five last years, we've been only focusing on the deployment of networks, fiber networks, as well as utility networks. And we're proud to have achieved today over 30,000 technicians now using Deepomatic technology every day when building and operating networks and in particular fiber networks. In terms of volume, 30,000 technicians, that represented over 20 million different work orders that were processed through our platform. Work order can be installing a new network asset, doing repair, connecting new subscribers to your network. And for each of those work orders, Deepomatic would process a set of photos in order to extract the information from those photos, where the job was done, how well it was done, what's the data that we can see there, and validate to the technician that was just performing this job that the job was done properly. Sure. So, you know, you should see that pull up on your screen now. And a pretty simple question first, which is, do you take photos in the field today? So, yes or no? And we'll give people a few minutes to respond there. Or not, a few minutes rather, a few seconds, I should say. So, get your responses in. And so far, it looks like… for the most part there's about three quarters of the people so far do take pictures in the field to help just another second or so here and let's have a look at the results all right so pretty clear result most people do take photos on the field a couple people do not so you know if you have some comments about that I'd love to hear about it we have a Q&A station at the end so for each of those poll questions if you have any extra thought you want to share just put it in the chat and we'll try to address them but thank you for sharing that first set of response and I move on to the rest of my presentation the first thing I want to talk about is where did we come from in terms of starting to work with the well telecom industry and building an operating network when we started to work with our first customers like I mentioned four or five years ago the first area that we really paid attention to was quality control and we realized fairly quickly that the quality control process had a couple of weak points that I'll talk about here so the first one is that it's a fairly manual process today we're building networks in a very geographically distributed area in different scenarios set up equipment etc and so it's difficult to consistently check everything being done and so the fact that you're building large network but even smaller network doesn't have to be millions of passings to represent a really large area to cover physically and also potentially for all of you that are currently taking photos a really large amount of data to look at so even if you're currently doing a quality control process in the back office that involves looking at the digital data capture from the field so forms and photos it's still a time-consuming costly manual process and so the result is we've noticed that most customer that have worked with us prior to having onboarded automation and artificial intelligence would leave a significant amount of blind spots in their operations they're not able to look at every single job every single asset and every single photo being captured what that means is you're leaving potential issues in your network you're not addressing things that might be missing and so all of that creates impacts both in the short term and in the long term in the short term the results might be higher cost for your rollouts because you need to revisit certain sites that could have been done properly the first time but wasn't weren't it also delays your rollout because the more you use resources to repair things that should have been done right the first time the more you use resources to control things instead of building well you just take more time eventually having this technical debt let's say in your network also create the opportunities for more failures longer maintenance time and potentially delayed activation missed activation so that also impacts your top line and so this is what we're talking about when we're mentioned the cost of not doing quality control in the short term and we'll talk about the long term after so what i want to talk about in this next slide is well how do you know where you stand in terms of your ability to monitor and control your build and operation well we outlined a couple of questions you can ask yourself just to see what are your capabilities so the first one is well do you know can you tell if your contractors comply with your quality standards and at what rate that's a good question to evaluate if your current quality control process is is efficient or thorough in terms of piloting your build and operations from that question the other one you can ask yourself is can you accurately and with confidence list the top five causes for reworking your network is that something you're able to track do you know what they're caused by when they are created as well as how to address them another question that relates more to how how you're monitoring and monitoring your building and monitoring your records is are you able to have accurate closeouts and as built from the field in order to update your record and keep a clean picture of what you've been building and what you operate and finally the last question you can look at is well what's the state of your network in what condition are the different assets you're operating what's the health of your infrastructure and are you able to make any sort of predictive maintenance are you able to identify the cause for degradation so all these questions are really important in order to assess if looking in your quality control process is something that you should be thinking about and this is a good way to start thinking about this question and so why should you care about this question well first of all i already started talking about a couple of those ideas but this has costs now and it has costs in the future and so fixing things today has a different couple impacts and so the first one is you need to pay for rework so depending on the way you are building your asset the contractor might be eating the cost in certain scenarios but in a lot of cases if the initial questions i just asked in the previous slides are not answered with a very positive response what happens is by the time you discover problems you'll have to actually issue new work orders that you have to pay for and so not accurately building your network has impact right now and can increase your cost per passing as well as your cost per connection which we all know are some very important metrics right now in the way you evaluate your current operation but also the value of your network and that will have a whole range of different impact in the long term so i mentioned construction delays as well as service activation delay and eventually you just see an increase in your onm costs as your operate your network in the long term and so we'll jump into a quick demonstration of the product uh both diplomatic uh from the field perspective as well as the back office with some uh sneak peek into integration with um iq geo so the first thing before i jump into the video is a quick look at the process um so you can access those systems in mobile devices we don't require any specific hardware so to use diplomatic computer vision uh diplomatic computer vision solutions basically you just have to have a smartphone and then you can capture the photo of your work get the feedback and that's what serves as making sure your operations are done properly so that being said let's just go in a quick video i will try to comment over it it's fairly quick um so i'm sorry if that's a little too fast but happy to address any question after it all right so you should see a video all right so you should see a video popping up this is our mobile application where a technician is able to take a photo of their job so what i'm doing right now is uploading a photo i apologize if it's a little choppy i think the recording was a little difficult um and so this is the photo of an asset i just built as a technician on the field and you can see exactly what the eye is checking so i'm making sure this asset is properly secured on a pole and here we can see that there is a quality standard that was not properly met so this select loop is not properly secured i can look at what i'm expected to do what photo i'm expected to take and then correct any mistake that was built in the first instance that was flagged by the ai so i repair my work upload a new photo and now i can see that everything is green so i did my job properly and can go back to the list of requirements that i have and finish my job so now we have a second poll question i'll let david take over now too thank you very much and uh this is do you or your organization use some type of ai in your field support operations yes no or you plan to in the future so we'll leave that up for a few seconds here all right it looks like we have a good set of answers here so we will go ahead and have a look at the results looks like a pretty uh pretty good split yeah and a good chunk of people who are thinking about it in the future um so be curious to know also what type of ai you're looking at are we talking about llms and chatbot such as the um chat gpt type of solution or are we looking at solutions similar to diplomatic looking at um physical world uh data photos videos or maybe different type of imageries or data thank you very much for your response and i'll move on to the next slide and now what i will show you um this is just a still image of the video i will play afterward is how this information that we capture through the photo taking process that i just demonstrated earlier gets tied back into your system of record gis enabled systems such as iqgeo and so what you're seeing right now is the workflow manager from iqgeo you see a couple different tickets um and we'll be jumping into a specific ticket to show you exactly how information can be pulled um from it should be playing all right sorry i thought i was playing the video here we go um so you can see we are in the ticket management system from iqgeo i will be filtering them to look at a specific ticket if i look into it you will see that there's a lot of information on the left hand side that was captured and i can click into the link that directs me straight to the specific job that was done and the specific asset captured through the photo and here you can see the workmanship quality as well as below all the information that was captured into workflow manager was captured by the ai directly in the photo so everything that you see that's pulled into the form oh a dog i think that might be a used to find um so everything that um was captured by the ai is then pulled into your system to ensure that you have the right visibility and that everything is accurately logged so you know the asset was built properly you know your inventory is accurate and you can update your design based on the as built and so that can be linked to a system like iqgeo or your erp or any other system where this would have the right place so now let's take a a minute to talk about the business metrics as well as the opportunities in terms of deploying diplomatic i will share a couple of data points and also what to expect when deploying something like diplomatic i think we have one more uh quick poll here yep uh we do here's our last poll and uh the question here is which of the following use cases are of interest to your organization uh so poll survey splice case audits underground and aerial rollout contractor monitoring and billing customer service activation or the catch-all of all of these above and to clarify we're talking about use cases for automated quality control through photo taking like diplomatic all right we're getting our responses in give just another 10 seconds or so and if there is any other use cases that we haven't mentioned here feel free to also drop them in the question in the chat and we can talk about it in the q a station afterward all right let's go ahead and see how people responded all right even we move a majority looking at that or with a lot of people looking at all the above oh yeah a lot of all of the above a little bit of a focus on the construction piece which is no surprise you want to make sure that things are built properly before you either hook them on a paw or bury them so i'm not surprised there is a big focus on the build right now we might see more customer service activation down the line once we've finished such a big focus on the construction phase um but uh yes so thank you for your responses and we'll talk about it a little bit more after it so here have two slides that just summarizes the different pieces of the solution i won't spend too much time on those because you've just seen a quick demo but basically just to restate exactly what the process is when a technician that is either well rolling out your network building assets or operating them or connecting a subscriber to the network they will have jobs work orders scheduled for them with a set of photo that they need to capture through diplomatic in order to document the job validate it's done properly and get feedback if anything needs to be properly corrected so they take the photo that we saw earlier get the feedback and then they see the progress of the job they're doing and then all of that gets tied back into the back office so i showed how this connects to tools like iqgeo where we populate forms log the photo etc there are also tools that we have at diplomatic to do that review process and storing of information so you saw what's on the left hand side is our review dashboard it's a monitoring tool that's what was part of the video on the right hand side we also have capabilities to give you monitoring dashboards business intelligence dashboards because what we capture in those photos is not only the real world information such as reading labels detecting splices doing inventory but also all the metadata associated to test these photos so that means the time stamp the idea of the technician the geolocation and all of that can give a very accurate picture of the progress of your builds the state of your builds who is doing what where etc so those are tools that are natively integrated into the diplomatic solution and that you get access to once you develop you deploy one of our solutions so let's take a quick look at three examples of realized operational gains by some of our customers so this is just a hand-picked selection of um swisscom city fiber and telephonica that have done a different approach so swisscom has been deploying diplomatic pretty much on every part of their uh fiber operation and rollout so both for construction on aerial underground plant as well as inside plan as well as service activation so they've really put it a little bit everywhere um and where they were coming from was the need to increase their capacity for quality control and so a tool like diplomatic was able to multiply by seven-fold their audit capacity to reach a standardized systematic quality control approach that was uh necessary to address a rollout and rollout and activation delay of about six months which we managed to address and across these different metrics and those are only a selection there is actually a couple extra metrics they're tracking what was realized is an eight-figure return investment across the um the several years that we've worked with them city fiber was focusing more on their central office um we were able to multiply their audit capacity as well and another metric they were tracking really intensely was the number of revisit that were necessary uh in their central office to address problem and actually realize the uh patching that were necessary and thanks to diplomatic and the real -time quality control they were able to eliminate and measure that elimination of tens of visits um every month and finally telephonica is focusing more on the customer service activation piece which seems not to be a big focus but i'll still talk about it and explain why it's an interesting uh use case and why this metric is interesting um the first thing they were tracking is the accuracy of the documentation of the job they required the technician to take photo but because service activation represents such a large volume of operation the number of photo was in the millions and they're not able to look at all of them and so they capture a lot of data that was just not really useful and through the diplomatic validation we ensured that they documented accurately all their operation and a very interesting metric was how quickly within a couple of months we reduced the number of for visits here by 13 percent which has a really big impact on the cost of passings and the overall efficiency of their operation why this metric is interesting to look at is because this 13 is only the the revisits that were directly attributed to diplomatic flagging a problem that would have required a revisit potentially a service call and addressing them before the technician leaves what this does not capture is actually all the jobs that were done properly because of a change of behavior because what happens when you provide real-time feedback on the field is you make your workforce actually more accurate more consistent when diplomatic provides a flag and says this has not been done properly it's usually due to human error forgetting something or someone who doesn't know is fairly new to the process and needs to learn but that goes somewhat quickly when you build something that was that is not done properly and you can see it you're not going to take a photo of that knowing that the ai will flag it so if you see that there is a problem you'll fix it before you take the photo and so all this change of behavior is something that has an even bigger impact than this 13 percent so how do you um track those metrics what's the first couple of questions you can ask yourself in terms of what's going to be the return on investment for me in my operation when i deploy something like the computer vision solution that we propose at diplomatic well you can start by asking yourself what's the size of your quality control operation and how much are you able to go to control right so you can look at the cost of those resources so how many employees are doing field audits usually not full-time so you can look at the fte how many of those are doing back office quality control and then if you have those numbers if you are doing quality control you can start looking at the number of errors that are detected and the cost detected to revisits linked to these errors i put in it like a couple of metrics that are circulating in research papers but 22 of fiber network failures are actually caused by human errors that are happening in the initial construction phase or operation when someone touches a piece of your network the rest of the failure might be exogenous events and weather events etc but that still represent a very large amount of failures that you can address and then in terms of revisit cost it really depends on your architecture and if you're working on urban or rural areas but as a conservative number you can look at between a hundred dollar to five hundred dollars for a truck roll to address those mistakes that could have been caught when someone was already there and already working on your asset um now i will talk quickly about well what does it take to deploy something like diplomatic how much time does that take um so the first thing i will show you is a typical timeline to deploy diplomatic and specific milestones so once we have scoped a use case that we want to focus on we will start collecting photos on your asset in your operation to train those models the very first thing that we're going to train is models that are able to validate that the photos are taken properly this is the document accuracy milestone so within 30 30 days we'll be able to have solutions that will validate accurately that whenever the technician captures the photo we are requiring them that photo is captured properly otherwise it's flagged for the photos to be retaken so that's something that happens really quickly and already provides a lot of value it allows you to start documenting your asset more accurately and all these photos that have been already taken properly can be reused once we have trained the rest of the solution for analysis on the workmanship and then over the next 30 to 60 days we'll train the rest of the models that are actually validating the quality of the job that's picking up the real world information that you're interested in so did i put the gravels in my hand hole is my grounding rod there is my fiber splice case properly sealed if it's in the risk of uh floods in the flooding zone for example so we'll train the models to look at the specific visual check for workmanship quality across this timeline so within 30 to 90 days you're starting to see the first impact of the solution already improving your operation and at the month four mark you have a fully deployed solution that now is part of your process and ensure that every single job is checked properly so how do you get started well the first thing to ask yourself is either you currently have a quality control process in place that involves taking photo properly documented or you don't but in order to have the right process deployed by diplomatic the first thing to identify is what are the major challenges you want to address what's your priority what's the work that you need to validate and what's the process to capture these photos so who captures the photos what photo do we need when are they captured and then we have to define what corresponds to a good situation and what corresponds to a bad situation so basically deep amatic would work with you to help define your quality control process that can be ai enabled and then we'll look at the data set that's necessary to train the ai so i mentioned earlier how our platform is able to deploy those solution quickly so you saw what our timeline looks like but also with a small amount of data so we don't need you to come with a million photo couple million photos to train those models a couple hundred photos maybe a thousand will be enough to start training those model and provide accurate results but we need to define what assets you're working with what a good photo looks like so for example an open splice case might need to be taken properly from the top framing the entire photo and not from far where at an angle so those are the metrics that sorry the scope that we need to define and then where those photos are taken is this an inside environment such as the mdu splitters installation in a basement that we showed in the initial demo earlier or is it a on the pole outside splice case and then we'll be looking at what is a good situation and what's a bad situation those two slides i just showed in terms of defining what's necessary to get started we have extensive experience in doing that and we can help you go through that process very quickly what i'm showing here is we can address your operation at different stage of your network life cycle so from the central office like we showed with city fiber all the way to the service activation and every stage in in between that involves building an asset operating it testing it etc whenever you can think of a photo based process that is visual even if it means taking a photo of an otdr test of a parameter reading those are visual information that we can process so from this you can identify what's your priority and from there we can expand the use cases and go in different areas and now i'd like to finish on our vision for the long-term benefits of using computer vision embedded into system like iq geo and what does it mean for the future of operating fiber networks and i have two examples i want to talk about the first one is the capacity to do accurate splice inventory and so what i'm showing here is a deployment that is being done to use a specific technology in computer vision that we call segmentation which basically allow us to identify very precise shape and color within the image images that we're using to check for the color of the splices of the fiber strands and what they're connected to in order to provide an accurate splice inventory in practice what that looks like is if we're looking at a big fiber splice case we can think of a process where the different areas are labeled in order to indicate from what tube those strands are coming from and thanks to segmentation we're able to look at the different color match it to the strand it's connected to and then thanks to the labels we're able to provide a very accurate automated audit document that allows you to either do the inventory on existing spice cases but also ensure that new assets that you're building are corresponding to your design and if they are not built according to your design update your design based on the reality of the field the second example is kind of the same thing just on different assets this is an example with a terminal with pre-connectorized ports you can use the ai to validate where those ports have been used imagine that this can be the same situation on a large fdh street cabinet for example we can do the entire inventory of the ports that are used or not used in order to update your system of record ensure that when you're connecting new people you are aware of what ports are available and not be in a situation when you think you can connect new addresses but you actually are missing certain ports because your inventory is not accurate and so i'll let stefan take over now to explain why capturing this information is so valuable thank you so much thomas so our story is being written at the moment in three very distinct chapters right so what we're working with today it's making sure that work that's happened in the field that's done correctly right it's this idea of right first time when you go to the work you're doing the right thing you're lowering the number of revisits you're not lowering the amount of rework to be done and making sure that you know what gets captured from the work actually reflects reality the next chapter that we're starting to write at the moment is helping automate your audits and s builds think of it in two directions the first direction is you know as thomas was explaining you know capturing the details from the audit and having basically both from a workflow perspective and from the system of record update perspective having that information correlated validated and verified so your ass builds are actually you know as accurate as they can be but also your audits you know have the ability to modify and affect the system of record data this is towards leading away from the traditional form-based interfaces that we're seeing in the field we've taken a lot of feedback from customers and they've told us repeatedly that the worst thing that their technicians have to do when they go on the job is open a laptop or you know take a piece of paper and input data to basically capture the details of the work done you know a picture is worth you know over a thousand forms you know when it comes to capturing accurate as builds and then what's coming afterwards it's using this information to build more predictive proactive automation into our network models so as we build a much deeper digital twin with very accurate information about what's in the ground what's in service what's operational what's not operational uh we're using this information also to do things like say for example tracking change over time for audits you know do things basically change take a poll for example how much is this poll essentially you know leaning in a in a specific direction over subsequent audits so you can identify where you might actually have to go in and you know get a a new poll rebuilt because this one might actually be you know sipping into the ground um capturing you know environmental details over time for things like splice closures and manholes uh or capturing details about surface and and you know other information around the area of work that you can also use to basically determine the uh or help you guide um you know the performance of your network can use this data for example to identify surfaces through which you know fiber deployments have failed over time this is all super exciting and it's just you know part of our initial journey so what we're doing is we're building a much more automated and identified network life cycle the idea is that everything that you do from design to plan to build to monetize and to operate can be extended into a you know number of assets that can be identified and apologies for my dog um we're basically we're you know adding more complexity to the agents and enriching the data they can do so we can go from digitization you know from a very descriptive digital twin and an informative digital twin you know that helps you digitize these assets and make sure that you end up with clean data at the end of it towards a more proactive and self -optimizing digital twin that helps you you know model revenue identify proactively places where you know data can be used to you know automate work or trigger predictive maintenance moving into close loop analytics and task agents and self-optimizing models that can do that work for you or can do that work with you the idea is that we we're in a in a path that it's about augmentation through agentification so we can do things before humans have to intervene give the right information to humans so they can actually check and validate the work that's being done uh and approve it or you know automatically you know reject it uh preempt some you know basic maintenance work or some advanced maintenance work that can you know come out of you know not having the right information and not keeping your system of record and as builds updated the idea is that the more that we can do to automate the process the better data quality is going to be the less friction there would be with your field uh operations and the more opportunities to actually maximize the revenue of your networks is going to be and with that you know we'll open it to q a uh and i have the q a open so i'll you know i'll ask as the as a moderator so we'll take the first one um so how quickly do we get the response of an okay or not okay from the photo um i can take that uh it's a very quick response um out of the millions of photos hundreds of millions of photos we processed last year the average across everything was under 1.5 seconds that includes some longer image to process like the complete inventory of a large street cabinet i mentioned earlier on average it's less than a second the analysis of the image itself is extremely quick um what can take time is the upload of the photo to the cloud for analysis um and i'll just do a quick comment extra on this that's why we have deployed now native apps on ios and androids that now allow us to deploy those ai models embedded in the phone which means we can process those images now also offline which will increase the speed of the analysis because now we do not need to upload the photo for cloud analysis anymore and allow us to provide real-time feedback also in areas with absolutely no connectivity but uh thanks for the question great question excellent so another question how how do we verify that the photo that's being referenced is the correct one um i'm not sure i understand the question exactly the photo reference is it the correct one well i can understand that a couple different ways i would say the first one is well the photo that we are requiring and processing will correspond to what you actually have as part of your quality control process so it's not like we come with requirements and say these are the photo you need to take even though we can give you advice it's more about what photos you want to have and so then based on those photos and based on the data set we'll collect as part of the timeline i showed earlier we'll train models that will validate that the context the overall photo is accurate it's the asset that we want it's framed in the way we want it's not too dark it's not blurry and so that very first check is the first thing we train the ai to do that makes sure is that when the technician takes a photo it's the photo that we want and it's taken properly that's how i understand the question i hope that's how it was meant but yes we train the models to validate the photo was taken properly based on your requirements excellent a question related to splice case audits and we've spoken a little bit about that but what exactly can we check in uh in that use case so that's a good question because it leads me to give a pretty general answer which is that the rule of thumb is if you can see it in the photo with your own eyes then the ai can as well the ai is only slightly better just because it will look you know at the pixel level it doesn't get tired it doesn't make mistake but it can't see things that are not visible so as long as you can see it we'll be able to see it if the color is visible we can see the color now if you're telling me well i want to make sure i can follow that strand throughout a very big case with a lot of other strand as to where it's coming from it's probably not possible because you can do it either yourself just because visibly it's not possible if you're looking at a very large central office panel with a ton of cable that are very messy and you want to do a complete inventory of that you can't even see the ports behind the cable that's not possible either so it's very powerful we can check anything that's visual in your process but it's not magic but that leads me to another thing which is how important the process is so in the example i gave i mentioned putting labels in this place case in order to track where that strand is coming from that's to address the fact that you can't see everything in the photo but by addressing that by taking a series of photo by adding labels potentially then you can expand the capacity of the ai to track the right information yeah i'll add something you know to that um which is the the role that the system of record at that point plays in the process right so you can capture what is in reality you know through the picture it can tell you hey you know ports one and two were supposed to be red green but it's actually you know blue yellow uh well whoops you know what's connected there the information that you have in your system of record and you know system of record that's dynamic an actual dynamic digital twin can allow you to trace back you know those fibers to find out service connectivity and give you accurate reports of what should be connected there what was designed there um and what might actually be flagged for service there that you can use to then you know as part of your uh audit process determine whether or not you know your customer service might actually be rerouted differently and then trigger a workflow to go through the audit throughout all the you know the entire service path all right moving to other um questions um all right there's a question yes it's a it's a combination of diplomatic ai and iqg a mandatory requirement or it can be integrated in existing apps and work order management systems if the answer at the moment is yes you can integrate diplomatic as a standalone solution um you are going to be missing some of the you know uh capabilities of the integrated solution but by all means you know yeah and to add to this um we have more than 30 different large fiber network as customers i would say we have maybe 30 different integration into their own project management tools um field service management tools gis tools etc a lot of the time uh homegrown sometime uh tools like iq geo workflow manager that are you know used by different customer uh so we have a lot of experience in integrating into a different tool so we don't require you to change your processes if you have something that works for you an area where people are taking photos and managing a project deep medic will integrate into that process and those tools excellent so we have a question for folks that use contractors to perform this work they may not have you know access to the same level of detail that you might have if you have people in-house doing uh what you know could we do if there's only partial pictures available um so that's a it's an interesting question because i would say all of the customer or 95 of the customer that we have are doing all of their work through contractors um some of them have a process that involves giving all of their contractor access to project management tools where they're requiring to take photos in that scenario it's fairly straightforward diplomatic is embedded in that process and they have control over what's being taken we have uh i would say it's a 50 50 split another section of customer that do not necessarily give tools to their contractor to use as part of the processes before diplomatic in that scenario you don't need to have a tool in place diplomatic can be used completely as standalone you saw you saw in the demo how we have mobile application back office application that can just be accessed from any device and so in that scenario you require your contractors to capture the photo through diplomatic we will have visibility over what photo they took who took them so it's very easy for you to over time address contractors that might not be capturing all the photo that you want and make sure that the adoption of the tool is progressively higher and higher and because we will look at every single photo everything single work order it becomes an easier conversation to have when you have visibility over exactly what's being done by your uh contractor and that's something we've worked on with many of our customers and that change management process is something we're used to supporting excellent um so here's an interesting question who owns the visual information data great question it's the customer so we don't own um the photos we don't own the output from the photos you own this data we store it for you we store it as long as you're a customer we can send it to other system if you want to either uh store it there or store a copy and if you want to stop working with diplomatic you are entitled to receive all the photos all the data you might not have stored in a different system and we will not be keeping this data uh yeah another question about splice closures can we check counts and labels such as correct counts used and correct fsa information against the design yes absolutely um so visually uh these are things we can pick up in photos uh again with the caveat as like i said the rule of thumb is can you see it in the photo uh for these yes so we can capture that can we compare that to the design and make sure it corresponds well it depends how we integrate in your system how this data flow flows between the different system and if this information is already available i assume from your question this information is available so we know what we're working on and what's expected on that asset which means that diplomatic can use this information to flag any problem directly to the technician and say you were supposed to build it this way this is how you build it so either they can fix it and that's good or you need to update your system of record because we demonstrated that it was done differently uh so yes to both of your questions excellent so here's another interesting one um how about in building and inside an mdu inside plant networks how efficient is the system in those use cases uh those are use cases we are currently deploying for a couple of our customer um i i don't know how to qualify exactly how efficient it is i would say just as efficient as any other use cases um there is a lot of visual quality control that we can perform to ensure that the job's been done properly maybe you're referencing to making sure the asset is exactly on a precise location in a building obviously that's not necessarily something diplomatic will help you with a ton we provide geolocation based on the photo metadata so you're not going to be able to say just based on that photo it's you know i don't know on second basement uh in the northeast corner um but for everything else diplomatic will be extremely efficient and if it's something installed in a basement where you have no connectivity as i mentioned earlier we can even have the ai directly embedded into a mobile of the technician which means that they still get the real-time feedback on the quality of their job so this is a great use case and if you want to discuss it further i'd be happy to spend some time with you okay here's a good question can the model be used to provide things like attachment heights without having to use a cvt stick uh that's another excellent question uh the answer is you need a reference so like a measuring stick currently um so that's a use case that we uh have in place with a couple of customers so measuring heights of attachment on a pole with a measuring stick or the depth of a trench as well the depth of a um a drop a very drop of a hen hole again with a reference object now we do have r&d projects in place to be able to do those measurements um without um a referenced object and we're also working on some systems to use different data sources uh of poles in particular that would include um metadata on the um hardware that we use to capture the photo as well as well as lidar data which is going to enable us to do extremely precise uh pulse survey uh based on those large data sources excellent so how could tools like this um handle things like say for example satellite imagery uh to recognize uh to recognize i know this is a little bit of a left field question but i i love those uh how can we use ai to recognize you know or or do things with say for example satellite imagery or you know um like you know fly over imagery on specific areas for doing things like say for example you know uh identifying routes for fiber or identifying potential like you know existing assets so i would say pretty much in the same way we're using it right now in the field on mobile devices to check for quality um we need to define a process a scope what photo we expect what elements in the photo we're looking for and with a properly defined a process we will train the different ai models to be able to perform those tasks so we have customer right now that are using other data sources than just from a mobile device it can be drone data footage images it can be a satellite imagery as long as we can have the process and the visual scope we can train ai models that will be doing that manual process for you automatically excellent um there's a lot of questions about offline um and to summarize them uh you know there's mention of an offline mode right um just running on the on the handset um is there any like difference in accuracy between you know the online mode and the offline mode um that's another great and very precise question um so i think the person who or the persons who are asking these questions are aware that well when we upload a photo to the cloud we get access to very powerful devices and your smartphone might not be that powerful um so when it comes to what we deploy offline a lot of the visual checks that our customer currently do are fairly and i put that in quotation simple so which means detecting elements in a photo or reading labels even though they provide a ton of value they don't necessarily require a lot of computing power so all of these we will be deploying offline and the accuracy will be just as good as online we deploy the exact same models and the computing power of the phone is sufficient when we're looking at more complex and by more complex it just means you need more computing power so that could be the segmentation for example of your splice cases that i showed earlier or doing an inventory of a very large street cabinet fiber distribution hub you need more iteration meaning that the models will work a lot more on your photo than just to detect something that what we're doing for those is not deploying them offline so basically if i'm taking a photo of a street cabinet in an area completely offline i will still get the ai that says you took a great photo they will be the ai that might be able to do the before after and check where you plugged the new uh connections you did but the model that will do the entire analysis of the cabinet will just not run on your phone and will wait for your device to be back online to run that analysis um in in the in the cloud so the answer is we will deploy everything that can be done in your phone which means probably 80 85 percent of the visual checks that you need to perform will be available in the phone offline accuracy just as good as online and anything that requires a bigger inventory process in the background will be done online excellent and you know in the interest of time um you know we'll have a well-masked question and then you know let uh you know uh they close out the the webinar again but this is a really interesting one what about legal requirements in specific geographies where there needs to be security around you know um photos taking say for example in a home or in a secure area or in a public area how do we handle that um so that's a another important question uh diplomatic originates from uh france uh so everything we've done in those past 11 years has been and continues to be gdpr compliant which means that when required we'll be able to make sure that no personal identification data is available where it shouldn't be uh so making sure you don't have faces or license plates or stuff like that most of our use cases do not entails capturing any of these information when we do cn we have processes in place to protect this data and ensure it's securely processed and so a lot of our customers are in the european union which means we have to be extremely careful with how we process this type of information um and we have a lot of experience in this excellent all right and with that you know if we didn't get to answer your question uh you know don't be shy you'll get you know access to us and you know back to back to our host thank you thank you very much stephan and thank you thomas for taking us through this today this is uh has been great and we really appreciate you taking time to to share with uh with everyone today and uh yes we received a lot of questions so if uh we didn't if the the team didn't get a chance to get to your question they will receive them afterwards and i'm sure you can expect some follow-up uh if you do want to uh have direct follow-up with um with the team today uh there's a when the brief survey comes up at the end of today's webinar uh one of the uh questions there is whether or not you'd like to be contacted and have a follow-up conversation so um please do uh reach out to us if that's the case and be sure to check out the resources uh that are available for you as well uh with that we're almost at the hour we just have a couple of quick announcements for you of course we are uh continuing our regional fiber connect uh workshops uh uh uh you know throughout the country and beyond in fact uh we'll be in toronto canada uh in august here we uh had a great event in uh in alaska last month so uh toronto is this month and then we'll follow up in september with spokane washington uh will be in scottsdale arizona in october and uh finally wrapping up the year in kansas city uh missouri in uh november uh so if you happen to be near those areas or love to travel uh please join us at one of those regional events they're smaller than our big fiber connect show every every year uh these are uh you know just as you would expect they're regional events so it's uh great networking opportunities more intimate uh great chances to uh interact with uh uh you know peers in the industry and um uh i encourage you to participate there i also just want to highlight once again our optic path program uh this is our program to change uh future uh fiber technicians if you're interested in uh learning more about that program or uh even wish to volunteer uh we'd love to hear from you so follow that qr code or follow the links that you'll find in the resources tool and finally please go out to our uh website check out our upcoming webinars including fiber for breakfast you can check out this morning's fiber for breakfast where we uh spoke with uh andrew butcher from the main connectivity authority and it was a fun one so encourage you to check that out on the replay uh but go out to the website follow this qr code or follow the link that's there in the resources to see what else we have in store coming up as far as webinars go and with that thank you very much for attending and we look forward to seeing everybody on the next one have a great day