Validating Telecom Field Work to Reduce Costly Rework

It's more than speed, it's more than light It's connection From sea to sea, it's changing lives It's progression Forever is our future we share Tomorrow starts here Hello everyone and welcome to today's Fiber Broadband Association webinar which is titled Stop Guessing. Start Knowing. How Visual AI is Transforming Telecoms Field Operations presented by IQGeo My name is Lisa Rosen I'm the webinar producer here at the FBA A few housekeeping items before we begin Please note that all of our webinars are available to you All attendees are in listen only mode We will have a Q&A session following the presentation so please ask your questions throughout using the questions tool located beside the video panel Also in the upper right hand corner you can find reactions if you want to share instant feedback with your presenter This webinar will be available on demand on the FBA website under the events webinar section And at the end of this webinar you will be prompted to complete a brief survey Please take a moment to respond as your feedback is valuable to us Please take a moment to respond as your feedback is valuable to us With that we will get started Today we are pleased to be joined by Thomas Thuillier Head of AI Solutions Americas at IQGeo. And with that I will turn things over to Thomas Thank you very much. Nice to see you everyone. I will be sharing my screen today in order to present because I have a couple videos I'd like to show you So maybe you can confirm that you see my screen just quickly so I know it's all right Thank you I think that's the case all right Thank you We will be keeping an eye on the Q&A tab as well as any reaction so I'd like to make this webinar interactive if there is anything that you want to ask while I'm talking on a specific slide a specific item I'm showing please just click on the Q&A tab and see if there is anything that you want to ask while I'm talking on a specific slide a specific item I'm showing. Please just react share something on the Q&A section and we'll have someone notice me that there is a question I could be answering That being said I will jump straight into it we are going to talk today about visual AI and its application to telecom field operations And here is an agenda we'll try to keep this around 40 minutes to leave time for Q&A So we'll talk about some of the costs that we want to address some very specific KPIs that we have tracked over the years with the many customers that are using our technology in the field today I will present our visualization studio which is a new recent product development that we've made available and give you a few demos We'll also look at a specific customer deployment for more precise feedback from their use of our technology as well as some indication as to how to get started with us and why work with IQGeo So stay through the whole webinar there is a lot of good information throughout from the beginning to the end And that being said I'd like to start with a provocative thought which is nonetheless true that right now as we're speaking today there is somewhere in the US in Europe a field job that's not completed properly This is the reality of the work that we're doing many technicians performing many types of different tasks on different types of assets that are hard to control And this is something that today we have more or less accepted as normal and it's not the case in a lot of other industries we've had some discussion with in the past and that's something that is often being brought up by the customer we work with Which is that if we look at other large scale type of operations such as manufacturing quality control has been implemented throughout as a systematic requirement of how things are being done built and operated And if we compare today modern manufacturing to telecom field operation we're far cry from this type of precision Obviously this is not necessarily to say we can reach the level of a Toyota with three defects per million parts But 20% of failure in average across the entire lifecycle of a telecommunication network is something we think can be addressed And it has a cost so I'll go through those in order the first obvious one is that when something is not performed properly on the field initially It requires in many cases an immediate track roll and sometime a follow-up down the line it could take months but that track roll will still cost something We're using a metric of 200 to 300 dollar this correspond to a cheaper service call at a customer's premises for example But that cost can really balloon if you're talking about buried equipment in rural areas where sometimes a single track roll can reach a thousand dollar twelve hundred dollars As I said in my introduction 12 to 20% of jobs being performed in the field will include a mistake That might trigger a truck roll either immediately because it results in no lights for an end user or down the line when an incident happened and something has not been built properly It will require to be fixed So those are direct costs direct impact to your operations that should be addressed In parallel to this obviously we know that this is happening and the systems we use to address them are today pretty antiquated In most cases it involves manual quality control when the data is available If someone is taking photo you'll have a supervisor a regional manager contractors reviewing this data spending a significant amount of time and usually to only sample some of the job being performed On average we're seeing 15 to 20% of data coming from the field being looked at and this is looked at after the fact So you may catch some of those mistakes, but they are already there and you're spending time actually detecting them The impact is not just on direct cost and on time lost. It's also in terms of achieving your objectives What we've seen with our customer is a direct impact on meeting your rollout targets your connection targets by having defect in the field repeat truck rolls problem being discovered as we move forward We've had customers impacted as up to six months of delay, which really is costly So what exactly is the problem we're addressing in summary with that last point you're building your assets problems are being built into it Lack of visibility leads to then impacts on the connection on the operation side of the network and then on the operation and maintenance side Every time a technician touches an asset or connects a subscriber bear themselves Either putting more defects in the field if not control or themselves being impacted by the facts that are already there assets that are missing or not built properly So how do we address this what are the KPIs I mentioned in my introduction that we have seen being impacted by computer vision in the field So we'll start with getting the work done right the first time So if you're familiar with the way IQgeo is deploying computer vision on the field, we make it so that it gives the technician doing the work a real time feedback So they install an asset to take photos and we immediately tell them if we see something that is wrong If the documentation of the asset is not done properly, if there is a problem that they should be fixing, if something is missing And so that real time feedback is where we dramatically reduce the number of time where someone leaves an asset with a problem So the bottom of my slide you'll see some metrics that come from they're not all from the same deployment Some of them are for the same customer, but they come from real deployment that have been done with customers as large as having several million FTT subscribers To smaller networks that connect businesses And some of the successes have been reaching over 95% of first time ride operations being measured What that means is we can actually prove that over 95% of the time when a technician did something on your network, it was done perfectly We have divided by two the number of connection to an FTT network that had a problem in the first visit We have achieved accuracy metric of over 98% of defect detection And contabilized exactly how many return visits in certain cases have been avoided Those are things we can demonstrate and that have a real impact on our customer for the first time ride piece of the first time ride piece of our business cases On the right hand side is about building faster In North America, we're still very much in a rollout phase Which means that having the first time ride piece of our solution as part of your rollout process You increase the likelihood that you meet your rollout targets and KPIs By reducing the amount of resources tied into repair And reducing the amount of resources tied into repair Approving jobs and just getting things moving faster overall We also increase the trust in your records Because using computer vision is part of your quality control means that you have an automated system that will look at every single piece of vision Data coming from the field So that you can trust the picture that is being built in your system of record So now you're creating a collection of proof points of visual and accurate physical representation of your network That means when you're looking at a specific asset when you're preparing new jobs, you know what's there, you know in what state it is And that has a direct impact on your current planning, future planning, as well as future operation and maintenance We have the capacity to handle very large scale We have customers taking over 10 million photos every year of their network while they're building and operating And the scale of that operation is what warrants automation And at the end when you're working with contractors, which is the case for many of you Having an automated system that systematically looks at what's coming from the field Means that you create a third party For trust in your relationship between You as the operator and the asset owner and your field partners And you're helping them proving the quality of their work proving that it's been done And having a baseline of conversation around what has been done and how it has been done That is grounded in reality in truth Which benefits everybody in that ecosystem So the vendors as well as the operators So now I'd like to introduce a new innovation which we call the Visual Agent Studio So beyond just having computer vision as part of your field operation We also want to accelerate the delivery of that technology to the field So we have prepared a library of AI-powered visual agents That are specifically trained and built for telecom and utility networks And ready to be deployed immediately That means that if you resonate with what I've been describing If you can identify certain areas of your builds and operations That could benefit from our technology We have agents that are ready to be plugged in your system And deliver initial value right away with a first-time right approach That means we have three ways to deploy computer vision in your creation today The first one is the pre-built That's what I've just described Agents that are already prompted with an existing workflow That we've based in our experience with the field standards That we know through our many customers today That you can already put in place in your field operation You can use those agents to configure them This is step two where you can tweak certain prompts Tweak the workflow, add certain checks And that's where the studio part of Visual Agent Studio really takes its name Where we can support you configuring these agents But you can also do that yourself You get access to a studio where writing new prompts Changing workflows, adapting the requirements Is done very easily and automatically updated to your user on the field And finally the advanced part Which is what you know from IQGeo's computer vision platform And previously Deepomatic Is where we train more advanced deep learning models That have some additional specific capabilities In particular doing measurements, counting equipment And more powerful being deployed offline on device So the advanced part will be particularly interesting If you're working in more rural area Where the cell connectivity is really low or non-existent And having that real-time feedback is crucial That's where the deep learning piece comes in All of this is part of a framework That we can use to move from the more simple and easy to deploy system On the field to the more complex system deployed at the edge And our new approach is really to start with the pre-prompted visual agent At the beginning regardless of the situation In order to iterate on the field And identify where are the areas where more focus needs to be done And improve the workflow, improve the standards, the requirement As we're moving towards the edge when necessary This method is the result of over 12 years of working with computer vision system And deploying them on field operation And this is what we trust during our deployment of customers today So to summarize We have a platform that helps you on the field in real-time At the point of capture Where when the technician takes photos in order to document their job We validate that that documentation is done properly The photos are done properly Well-framed, we see the required equipment But it's also built properly We give them the feedback so they can correct anything wrong This gives you a better understanding of what's on the field Because it's documenting everything in your system of record And allows you to act with a new found source of data and visibility Because not only do you get the photos Quality photos With quality control and inspection being done You also get all the metadata associated to it What I mean by that is The geolocation of the photo The timestamp of the photo The ID of the person who took it As well as what part of the company it belongs Meaning contractors, etc So this wealth of information Allows you to take action throughout the lifecycle of your network Quick summary of the different visual agents that are available today And currently being used We have a larger library But those are the ones we're focusing right now on putting at our customers It's the asset inspection piece The drop installation and service activation are specifically telecommunication So we're staying close to the premise of the customer being a person's home or business Meaning focusing on getting their service activated properly And connecting them from the point of connection and the distribution point and their home The reason why we're pushing these agents initially as part of our visual agents to do a rollout Is because those are the scenario where connectivity problems are the less impactful And where the specialized general models that need cloud connectivity are going to be the most appropriate We have agents available for build as well Very aerial But that's where we might want to think more around building towards the steep learning models As we are facing more rural and low connectivity situation That being said, I will move on to a live demo I will show you a specific visual agents So the service activation one with how it works Give you a look quickly on the flow as well as the back office And then we'll look at how to configure this agent So I'll get it started You can see that for those of you familiar with our platform You have the same look The technician is able to take photos of their equipment This is a little demo I recorded recently And these are the different checks that are already existing in an agent That can take photos of many different types of equipment Is fairly generalist So here you see we're inside the home We'll take a photo The processing time is really quick On average, we're under two seconds The process is done inside the home Outside the home And I'm just going to conclude on this one So you can see the whole flow We're checking quality We're checking the quality of the photo And giving that feedback in real time All of that is then centralized Just like you've seen in the past And this one is a job that we've done with a customer In a real scenario where we have a full workflow A little bit more complete than what I showed earlier You can see that we have all the information related to when this was done And all the data that was collected on the photo So this is this you can probably tell is a lot more than what I was showing on the phone And this is because we want to capture as much data as possible That could be useful for your system of records I paused for a second because what I want to show you here is that In the scenario of that build We have a technician where initially we're taking a photo of installation That does not meet standards And you can see that we keep the history of the jobs that are of the photos Sorry that are being taken So that in that flow Oops, sorry, it restarted the video I will move forward You can see here that if I go in the history of that photo We have the first photo that was taken The different problems that were detected Then a second photo And then finally we have the final photo that has everything shown as correct So it shows how the interaction between the technician and the result of the AI Helps get that first time right metric up So as part of the workflow You can have the requirements that if we see certain things not being perfectly done You get a back office technician or your contractor to review them And you can see that then you can mark them as process This is instituted so that as we need human in the loop For the remaining 1-2% of photos that might have edge cases We have something in place to address those And reach the 100% of job validation that we're seeking Then I will show you how we can configure So not just the pre-built agent But taking a pre-built agent and tweaking it So this is the studio that you would have access to To make changing on your own Trial new things And what I'll do here is create a new test So we'll require a new photo on the field In order to validate in this scenario That the TV services is done properly So what I'm doing here is I have a set of prompts prepared And I'm just pasting them in there And this is something that we have training material for So that you can be completely autonomous in creating those new tasks So I'll create a validation of that new photo with two different tasks Two different prompts And then what you can do in that studio is test that your new check works properly So what I'm doing here is uploading a photo test To see the behavior of the agent in that scenario And we can see that I'm getting the result I'm expecting And finally one of the improvements that we've done to the platform Is the ability to set an example up to 10 With what is expected in a scenario where the photo is good As well as what is expected when the photo is bad In that scenario I'm using the same photo So this fits requirements And putting what is supposed to happen And this is one of the things that makes IQgeo's visual AI particularly powerful Because adding context like that is a way for us to improve the accuracy of the AI Without having to retrain it on different equipment So that's what makes our visual agent so powerful from the get go From day one without having to spend time and data to train it Finally, I will give a quick refresher on how the deep learning piece works And looks on the system with a different example Which is more linked to the build the rollout of your network as I was saying So this is a system that lives directly on your phone on device It does not need any connectivity And that scenario I will take a couple photo of a drop in a drop terminal being built And we'll walk through what the interaction for the technician is So here you can see the different photos that are required I can have access to an example photo to make sure that I know what I'm taking a photo of And then I'm getting the result just as earlier of the job I just did In this scenario you can see that the AI will indicate in the photo what the problems are So that the technician can fix it So here we have a broken seal, we're missing clips You can see that the tray is properly built So now the expectation is for the technician to fix it Take a new photo and show that they corrected what was on the field So now we have all four seals done properly Which means that this buried equipment is not likely to flood And the clip is present to secure the drop cable The entry cable in the terminal So that once the next splicer comes to do a new drop They don't accidentally break existing splices Finally I did an extra photo I'll show you why in the rest of the demo in a second To conclude that job You can see the progress of the job You can see that this particular photo was validated So now I can move on to the rest of my task I will move directly to the back office Where you'll see that the job I just completed has been logged Just as previously for the visual AI studio I can sort through job I can sort through the dates it was done The different requirements In that scenario I'm looking for the job that we just performed Which is the number seven And you can see that my photos are there And as previously if I go into the history of the job I will have the different photos that were captured with the problem And the different fixed that were made by the technician So this is the second photo I took And what you can see here is that I have one red area That was not showed to the technician initially And that's the presence of those zip ties In order to secure the strand properly inside The reason why I'm showing this is I want to present how flexible our solution is In order to ensure that On the one hand you have the best possible build being done in your network But on the other hand we don't necessarily want to impose too much burden To the field technician and slow them down So as much as here we're checking that the zip tie is present We don't think with that customer that it's necessarily a defect that needs to be addressed immediately Or blocking the technician They might not have the zip ties for example The reason why we're still capturing it Is it helps you adapt the way your contractors are working in your network Because then the best course of action is to collect this information And maybe at the end of the month You could talk to your vendors and say Look we really want to have zip ties in our job terminals To ensure they are as secure as possible We've noticed 30% of your jobs don't have zip ties Can you make sure you tell your technicians to have zip ties and install them? So that's one of the ways we increase the quality of your network Without slowing down your technicians So that's it for my demos We'll take a look now at specific customer metrics I have a quick summary here of different deployments that we've done In terms of the amount of photos being taken The improvement to their work In terms of savings Both material saved Truck roll saved As well as time saved And a specific example I want to walk you through Is one we've recently presented During a keynote in the UK Is the work we've done with Circe Which is a large contractor for telecommunication and utilities network Employing several thousand technicians And taking tens of thousands of photos every day Which represents an extremely large volume That is very difficult to verify without substantial investments In manual quality control It also means that a lot of issues are found after the fact And would benefit from real-time feedback And at the end coming from a contractor Vendors that are paid by the job Having to redo job Not being able to quickly provide information on the job being completed And the quality of the jobs means it takes into their margin And it delays payment So we started working with Circe about four years ago And we've gradually deployed more and more use cases Consider one use case such as service activation Another use case drop installation Another use case could be pole attachment validation Buried equipment installation Inside plant patching for example So we gradually started with our visual agent studio Some pre-built agents that we were working on at the time And then slowly deploying more and more of those specific agents And so this was possible with Circe and our other customers Across a vast amount of technicians Also because we don't require any specific equipment You can use any of the commercial mobile devices That everybody used today on the field To have access to this technology Some results When we presented this in Q2 this year We had already taken more than a million photos with Circe At the beginning of the year And the measurement had been saving a significant of time On quality control In order to get paid on job being completed And the measured on this million photos Of the actual time it takes to process on the field Was under two seconds Which really confirms that this investment In quality control in the field It does not come at the expense of slowing down your job We know that this is a concern for many of you And not only do we catch problems You don't have drug roles And it also does not take time on the field to use So what are the lessons we learn on that specific deployment That's common across most of our customers The first one is start with one use case We don't recommend starting with many different areas Of your build or operations It's easier to select the one with the most impact Typically that's a type of task that has large volume Meaning many photos that are systematically consistent But is not possible to review manually because of that volume So that's why we're focusing on service activation Drop installation The more downstream in your network The higher the volume of repeat jobs Both for quality control but also to improve Track role reduction Second, keep it simple I hope that the demos I showed today Was demonstration enough that the tool is very straightforward And easy to use We have developed many different ways to ensure adoption By technician and even potential vendors that might be opposed to the idea And one of that is keeping it simple for engineers So the tool is built so that it's easy to use Very intuitive and actually helps the technician on the field You want to think about your data Not just for quality control but also for record keeping If you're doing quality control manually That does not really valorize your data You're not extracting additional KPIs and metrics that are linked To when those jobs were completed What they look like Are they associated to an asset in your GIS? Anything that will improve the visibility that you have Almost creating a digital twin of your network Via the automated analysis of physical data coming from field photos And finally, the client relationship changes This is from the perspective of a vendor But the reality is the same from the perspective of an operator Who decides to choose this tool for their vendors Having a baseline that is a basically a software An automated system that says this photo was done properly or not Along standards that are mutually agreed Really changes the way communication and relationship is built Between operators and their vendors You improve everybody's performance by having the technician on the field Being guided on collecting the right information Doing their job right the first time And then shared across the organization as what has truly been done What is flagged as being a problem But also simply just approving and green lighting jobs That the AI says it's just been done properly immediately You save time It's a faster payment for the vendors When they are able to prove immediately that the AI says All these jobs have been done properly We can move on and be faster I'll conclude on some words about what we've been doing over the years Who we are This is something that might be redundant for many of you Who I've talked to from Diplomatic before the acquisition by IQgeo Today we work with tens of telecommunication providers, operator, as well as vendors Representing over 40,000 field technicians that are using our technology every day We've processed over a billion photo 700 million photos processed annually With hundreds of tasks that are already available in the portfolio of visual AI agents That can be deployed for different part of your network lifecycle I want to give a quick comment about Why would you want to work with the visual agents of IQgeo and not build your own With LLMs that are now widely available? For a few different reasons The first one is that we have a wealth of experience that allows us to have really precisely trained agents Like I said, with over a billion image analysis But mostly it is a way for you to keep a lid on an ever increasing token cost of LLM use We have innovation inside our platform that constrains the way token are being utilized And mostly this goes through preventing re-analysis prompts that are not properly handled And trigger several reprompting and usage of unnecessary token Using the right models We have our own LLMs that are being utilized in a way that makes them appropriately nimble or heavier As the situation demands Which means that you don't need to worry about using the right module tier that can dramatically increase your cost As well as controlling the output and the input Which are things that you have very little control when using one of the widely available platform So as conclusion, not only do you start a journey towards first-time rights through our tools in the field You're also making sure that using visual agents on the field is not something that will compound and cost you a lot of token So how can we get started? I'll conclude the webinar now Basically we can have a chat about the part of your operations where you feel computer vision would have an impact I've suggested a couple areas such as service activation But there are many other Identifying what standards we want to address The metrics we want to address And identifying the right agents to put in production Is the way to go And is something that we can then put in the field in a matter of days Test over a couple of weeks And getting you live very quickly So if that interests you, please reach out You have our emails I'm also putting a little website link at the bottom where you can reach out But I'm also happy to answer your questions right now And make sure that we cover everything you're interested in So I'll stop sharing my screen now And I will take a look Thank you so much Yes, that was great And we do have a few questions that have come in already And we do have some time So the audience should continue to enter their questions as they see fit And we'll get to as many of them as we can So let's start with this one I think you may have touched on the timing already But how long does it take to get up and running? So like I said, it will take a couple of weeks at most If we're starting with one of those visual agents Especially around service activation So if you have an area where you want to get started Please reach out And we'll define a plan of action That will get you something in production within a couple of weeks Great What happens when a photo fails validation? Does the field technician have to redo the work immediately? Or can they continue? So that's a very good question The answer is it depends You've seen how during the demo A problem detected leads to a negative feedback to the technician Ideally, they fix it, take a new photo, and move on We have a scenario where our customers want to block the technicians from moving forward This is possible They cannot close the job It's especially possible when we integrate into their field service management application And that this FSM app waits for us to approve that the job was done properly Before they can close the job We don't necessarily recommend it though It's not necessarily a good way to speed up your operation Most of our customer requires the technician to redo the work and take a photo But do not block them if they don't do it Typically what we do then is we have a way to notify their supervisor For example, if a technician took a photo with a problem Did not take a new photo with correction within let's say 30 minutes We notify their supervisor with a link to the job So they can take a look and maybe contact that technician to figure out what's going on What's going on? What is wrong? If that's not something possible At the end of the day All of these jobs are flagged for back office review So that a human who is responsible for quality can have access to these jobs And assess if something was done wrong Over time, the goal is to change behavior So even if you don't prevent the technician from moving on If something is wrong That loop with people looking at what's going on And then communicating with them also changes behavior With long-term customer, we catch very little problems Because the technician knows that what they're submitting is being checked And they know that they can just conclude a job with everything being green And nobody will look at what they're doing And so that behavior change at the end is what really has a big impact as well So I hope I was exhausted with my answer And I hope that answered your question Great, thank you Related question Does the technician need to be mindful to stage the photo scene To not include any PII or other information the AI could grab And benefit from in other use cases? So that's a good question And I hear a couple different questions in this I'll start by addressing the PII question We are GDPR compliant We also have ISO certification to demonstrate our IT security responsibilities And one of the ways we're doing that is that we anonymize PII information When photos are taken in the field So we prevent the use of those type of information And when we do capture them We have customers who, for example, require a receipt of passage to be signed by a customer It includes their phone number, email address, name, etc. We handle this information as they should be as a PII to be GDPR compliant So they are siloed and only the right people have access to them So we take that very seriously So PII are protected in the system The other questions I hear in yours I don't know if that's what you had in mind But I'll take the opportunity to address this Does the technician need to be mindful of staging the photo properly? Yes, we try to standardize the way photos are taken on the field So if a technician is taking a photo of an asset from too far From an angle that's not correct Because then that prevents us from properly seeing what we want to see Or if it's blurry or too dark We usually flag and do not accept that photo at all And we require them to take the photo properly That's why we have this little example I showed during the demo Where we can show what we expect as the photo being taken That also solves a part of your question If a technician, for example, said take a photo of an access point on the side of a home For example, you could imagine they take it from too far And you can see people on the photo, cars, whatever it is We're actually just going to refuse that photo and not even keep it Because it's not taken properly So it does address a little bit of your PII question But mostly it's to increase the quality of the documentation and standardization And finally, you're saying, would it benefit to other cases? I'm not entirely sure what you mean by that But I'm guessing you're thinking of protecting the data that's being collected To train other AI to help other customers By default, you own your data So if you're purchasing our system to capture and analyze photos on the field This is your property And we do not use that data without your authorization And if you choose to seize using our technology You walk away with the data you generated that you own Great, thank you Does a technician need internet connectivity on their device running IQgeo at the time of the capture? Or is offline okay? Another good question because I didn't fully address it Offline is always okay So even if we're using a system that needs connectivity to be processed Where we do not have the AI on device The technician can still take photos that will be then stored in the cache Nothing is lost And the moment they're back online It's automatically processed and they automatically get the feedback In a scenario where we deploy visual agents that need connectivity And let's say they're working in a basement And they do not have connectivity They can still take all the necessary photo And then they walk back up Get connectivity Get that feedback And they're usually not too far from the job In a scenario where as I mentioned We're talking more rural jobs Where the impact of not getting that feedback immediately can be big Let's say you're driving for 30 minutes an hour with no connectivity Then it's not so great if you get that feedback so much later That's when we deploy those models directly on the device Which means they can run completely offline Great, thank you Let's see We already use Salesforce How does the Visual Agent Studio integrate with our existing systems Salesforce and ServiceNow? Great question So we have integrated with Salesforce and ServiceNow in the past As well as many other project management tools, GIS tools, field service management tools We have different ways to work this integration One simple one is when you schedule the jobs Have information related to what job it is We'll have a two-way API call that just retrieves the information from your project management system And we can send the information from the field from our system back into yours So you update your project KPIs metrics If you have a field application that you currently use in the technician's device with a workflow We would integrate in that workflow So you don't need to use our mobile application to have computer vision in your workflow We would integrate in the existing, for example, field service lightning from Salesforce In a way where if the technician has a job open with their form They would have a button to take a photo That's something that exists in this application today And we just do a deep link with that button So that for them it's a single pane of glass They inside field service lightning I'll keep this example Complete whatever they need to do in field service lightning If they go in field service lightning and then click on take a photo Our interface would naturally open on top of it Where they get the same UI The same feedback when they take a photo When they're done with taking the photo and they close it They're back into field service lightning And their form has been updated with the documents that I captured This is the way I mentioned earlier For example, you want to block a technician from closing a job If we detect any problem The way we would do it Because of the integration with the tool they're using on the field If we don't validate with the AI it's done properly We can prevent them from closing the job, for example So to conclude in very short If you have a field application We can integrate into it You don't need to have several applications for your technicians And you don't need to change the workflows that are already in place Great What type of quality issue can it scan and call out for splice trays and splicing in general? Well, that's a good question I have some specific examples But I'll start with a rule of thumb Basically, if an expert quality control controller can see it in the photo Then the AI can see it in the photo as well Which means the reverse is as well true If you have a very large account splice case with many strands of fiber completely entangled And you want to know to what cable, what tube a specific splice goes into I'm pretty sure nobody would be able to tell on the photo Which means the AI will not be able to tell as well So rule of thumb, if someone by looking at the photo can see it Then the AI can see it The AI will be slightly more precise Because we're looking at pixel level Imagine someone spending as much time as they need zooming in the photo That's the type of precision we can get A specific example Typically for splice use case We will validate that the tray is properly arranged So we'll flag if it looks like a red nest We can also validate colors So if you want to start building an inventory of your optical route We could match that red goes into blue, etc. But obviously it has limitation as I just said If you want to map what strand goes into what tube Most of the time this is not visible So if you want to build a complete splice case inventory It will require some manual intervention in order to map that full route On top of that, any physical quality indication or defect is something we'll check So on a splice case we can check that the trays are properly secured with a strap They're properly labeled When you close your can, is it sealed properly? In certain cases we have customers in cold areas in Canada That require their splice case to be pressurized We will check that they're pressurized So many different things Just keep in mind the rule of thumb If you can't see it, we can't see it Great I think this might be our last question We'll see if we get any more in But there's a few questions about pricing How is the pricing structured per user, per photo, per case? Is it relying on tokens? That's a good question And I can also see the question talking about predictability Well, it's very simple If you work with us, there is a cost per photo that's fixed So we can discuss that If we're looking at a use case together Where you basically pay on the usage The usage is fixed, it does not change And we handle making sure that our costs are in control And those are the innovation I was talking about earlier The model is the same Rather, whether you use the visual agents that are pre-trained, customized Or even the deep learning models that are available on device Our business model is on usage So that way you have predictability Especially since we will be looking at business case together Meaning, well, how many photos do you expect to process during the year? So that we can give you exactly a cost estimate that will not change Most of our customer commits on a three-year deployment And the price is fixed on usage throughout those three years So we really protect you against an unpredictable usage of the token While you're using the app But also over time We've seen how token costs just keep increasing That is not going to happen as well So thank you for this question Perfect Perfect Thank you So with that, I would like to thank you for presenting today I would like to remind the audience That to learn about upcoming, sure, FBA webinars They can scan the QR code on their screen Or go to the FBA website under the events section We hope everyone will join us for our Fiber for Breakfast webinar series Featuring FBA President and CEO Gary Bolton Which are live every Wednesday at 10am And past episodes are available on YouTube and SoundCloud Again, thank you, Thomas Thank you for our audience for attending today Have a great day, everyone Thank you, Lisa Thank you, everyone Bye-bye Thank you