How Visual AI Ensures Right-First-Time Smart Meter Installations.

Welcome back to Bitesize Electric where we unpack the ideas and innovations reshaping how utilities work one bite at a time. I'm your host Brandon Curkan and today we're talking about a challenge every utility running a smart meter rollout knows all too well. Making sure every installation is done right the first time. Smart meter programs operate at huge scales. Thousands or millions of installs, tight regulatory requirements, and very little tolerance for rework. A single missed photo, incorrect wiring, or a compliance issue can lead to repeat visits, audit risk, and rising costs. Today we're diving into how visual AI is changing that equation by bringing real-time quality assurance directly into the field. Instead of discovering problems days or even weeks later, crews get instant feedback while they're on site. To help us break this down, I'm joined again by Adrian McVelty from IQgeo. Adrian works closely with utilities and telcos to modernize field operations using geospatial workflows and AI-driven automation. Adrian, welcome back to Bite Size Electric. Hey, thanks, Brandon. Second podcast, excited, and it's a great topic. Alright, Adrian, let's start by explaining a bit more about our topic. In your words, can you tell me what real-time visual QAQC actually is and why it is a game changer for smart meter programs? Yeah, sure, Brandon. I think you know that there's a huge amount of work related to installation of smart meters today, whether they're electric or gas or water. And regardless of which geography, I mean, the Americas and me are APAC. So there's a mass program replacing older meters with new meters. And today what we see is a fairly manual quality assurance or quality check process. 10-15% of all meters installed have some type of manual check. So change in the game is automating that QAQC while the engineers are in the field doing the installation. And basically giving them the feedback there and then. So they know as they're on site that the job's been installed. But given a full digital record as well of all the checks they've performed. And utilities have been doing smart meter installs for a really long time. Why is this such a big deal for them right now? What's changed? Well, I think it's the volume of work that's happening right now. Depending on what part of the world you're in, the utility may own the meter or they may not own the meter. Typically it's getting deployed by contractors or engineers doing it on behalf of whoever the asset owner is at the end of the day. But just the volume of work. From talking to our customers who use this technology today. Anything from 6-10% of all installs have some type of failure. And most of them have nothing to do with the meter itself. It's just more of a process or there was an error during the installation process. So the more that we can reduce those to reduce truck loads and non-compliance in the future. It does a couple of things. It puts the meter in first time right and increases the volume. It gets the meter to bill quicker and reduced the cost of non-compliance. All which is good for whoever the asset owner is. So I've heard about AI analysis of photos as something that's done after the fact. Like a batch load or whatnot done in an office. But it sounds like you're saying this is happening on site. So how is that achieved and what kind of issues can it actually detect? Yeah, that's a good question. So what we see a lot of times right now is the installers or the engineers. They're actually taking images on the site, but they're not running AI models against them or they're not running automated checks against them. And there's two flavors which we've come across. One is where the installer does the job, the engineer does the job, takes images of pre and post installation, and they send them to a back office. The first flavor is they move to the next job. And then someone in the back office does a random QA QC check of those images. And if there's an issue, they have to roll a truck to address what issue they identify. The second flavor is the installer actually waits on site. They'll email the pictures back again. And someone will verify in the back office and say, yep, jobs complete, jobs safe, move to the next one. I think now, Brandon, I've got a work order application on my phone. And I open the ticket up. I'll go to my address and tell me what I'm putting in, all that type of stuff. From there, I launch the camera on my phone. I take a picture of the existing asset, the existing meter. And it's going to detect what type it is, the meter read. If it's electric meter, for example, peak polarity, right? Make sure that's good to go. Then they get to work. They do all the work. They do the install and they take a post, one or more post installation images. And that's looking at various different things. We'll pick up the actual asset type. What type of meter was it? The asset ID, the starting meter read, and the polarity test are all the seals on. And if there's any type of permits around, putting the job, the node stall in it. All detecting that automatically, but training the models to identify and give the feedback to the local engineer. If there's any issues, it will actually highlight on the image. I couldn't read the meter or blurry image. I should have mentioned, one of the first things it's doing is actually checking the quality of the image. Is it taken at the right time? Is it of the right asset in the right location? Does it have the right exposures? It's not blurry. You can actually see the asset. Many times when you're doing these manual images and sending them back, there could be general image quality issues that make it hard to do the QA, QC. So if you automate all of that, put it in the hands of the engineer on site, you're reducing the reliance on the back office and speeding up the process and improving quality as you go. So I can absolutely see how that reduces rework. But how does that help with other areas that utilities struggle with, like audit times or compliance documentation? If you're only doing 10% manual QA, you've got a 90% exposure potentially where other than the documentation of the installation, you don't have any checks to prove that was done correctly. Going from 10% to 100% digital QA, QC with a full audit trail, you know, then you've got a full digital record of everything. You can go back and look at every single image. Did it pass every quality control? You've got a timestamp. Who did it? When it was done? Basically from an audit perspective, then you've got all of the back end documentation. If you ever audited or if you've ever had an issue. Also, if you think from a more traditional perspective, having that document in a traditional file storage system is going to be time spent running reports to do the audits. If you've got that in what we have in Netlux where you've got the whole back end dashboard, you can go in and search via asset, via job number, via installation, via date. And you're going to get all of the images, all the jobs, all the controls, where they pass, where they failed, all basically through searches and filters. So it's very easy to run any reports post installation. One of our customers actually reported a 25% reduction of time in running reports and audits. So Adrian, in your last answer there, I heard you use the term Netlux. Could you explain what that is? Yeah, sure. I probably should have started with that at the beginning. About a year ago, we realized that AI in general is going to have a significant impact in both farm markets, telco and utilities. So we were looking at how do we leverage the technology. So we went through the whole build by partner analysis and really we were looking for what is complementary to our industry and what will scale and where would it have the biggest impact to our customers from a value perspective. So we acquired a company called Diplomatic based out of Paris in France. And they were very much aligned to IQgeo. First of all, they focus on telco and utility markets, which is identical to IQgeo. And they have field based visual AI. We have a field based application and they work at scale. That was really important to us. Some of our customers, the largest utility in telcos in the world. So it's important that we could find a technology that can scale to the same size as our traditional applications. So we rebranded the application that Diplomatic have and we call it Netlooks AI. So Netlooks AI is an application that sits on your phone and you can integrate into different field applications, IQgeo or other. And on that application, you build workflows to take images and train models against the images to identify and automate different processes. And this also works online and offline, which again is a massive differentiator compared to what else we see in the market. So Adrian, how long from when you start using this technology to it becoming actually useful and something that you're able to rely on? Yeah, that is a really good question because we all know the fundamental technology is based on deep learning techniques, which have traditionally taken a lot of information to train what's good, what's bad. So we've already deployed this technology many times to water, electric and gas. So we have an archive of models based on various different meter types and various different controls against the images of meters. So that's step one. We're not starting from scratch from the meter installation perspective. And then we'll refine with customer specific images. And then the full deployment is typically somewhere between three and six months. Additionally, what we can do is layer Gen AI techniques to accelerate that learning curve and then use the customer specific images to refine and get the high level of accuracy. Basically, we have a historical set of imagery of utility based data sets, Gen AI, and then refining with a customer specific image reduces that time and the time to value. Awesome. So you can use this to reduce rework, reduce your audit times, improve compliance documentation. But of course, at the end of the day, there's always one question that any executive at a utility is going to ask. And that's what kind of return on investment could you expect from a technology like this? First of all, the right first time. Let's say your failure rate is somewhere between six and 10%. And our customers are saying they're getting that up to 99% right first time. So significant reduction in non-compliance. That leads directly to dollars. So kind of rough rule of thumb, $200 for a truck roll in the US. If you're in the UK or a media, it's 200 euros, 200 pounds. You can do the math by the volume. So the ROI is quite easy to calculate. The other ROI, you reduce the need for the back office engineers. So if you're automating that task, so you could have two to five full-time employees, redeploy them to more value added tasks. Again, the significant ROI on there. Additionally, then you start to improve safety. You also improve the time from installation to billing. So imagine now that there's a manual input of a meter reading and the digits are not right. And then you start having billing irregularities. There's an ROI on that as well in improving those things. So the ROI in this use case is significant. And we've seen the payback of some of our customers as low as six months. Then another piece of the ROI or the value is in many cases, it's a contractor doing the installation. And they're competing for work. What we've been told from the customers we're working with, the contractors we're working with, is having this technology, having the ability to go 100% digital QA, QC, which reduces non-compliance jobs, improves safety, improves billing to time. It puts them in a market leading position to respond to competitive bids when they're competing against some of their peers. So that's another area of the significant benefit that this technology provides. And we talked about how this can be used when the technician is currently on site. Could you also use this visual AI on pictures you already have, maybe to detect where there are issues before they become a bigger problem? Yeah. I think that's the situation a lot of utilities have in. They've taken a lot of images already. What are our customers doing with all of the imagery they have today? They've got it in maybe some S3 bucket or in some SharePoint. Probably not the easiest to go back and search through if they ever need to get to it. So there's two things that we could do with Netlooks. One of them, the same models that we train to work on site, we can run them in a batch mode. Whether it's coming from a drone, whether it's from historical imagery, we can run those same models against the preexisting set of images, right? The second one is, think of it like an image hub, a repository with brains. You've got the ability to go and search through metadata, whether that's work order, asset type, date, engineer, see all the images, see what the controls were and how it passed. So what we could do is take all of your existing images and run it into the image hub. Clean out all the duplicates, clean out all the images that are not of a higher quality to use in the future and tie them to the assets and put them in the same structure that you can then index and search on. I think that's a significant value add. And then on top of that, then over time, you build more models against them. So yeah, that's a really great question. So, so far we focused here on meter installations, but what other workflows do you envision that visual AI could be useful for? I imagine there's a lot of utilities that could use this for other things like different equipment installations, maybe inspections. What else have you heard of that this could be useful for? Yeah. Now you're getting to the top of this really near and dear to my heart, Brandon. So as you know, depending on the workflow, the field crews are taking images on various different work types, whether it's an inspection, damage assessment, responding to an outage, whether they're doing an as-built. They're taking a lot of images today. So I think there's a lot of area where we can actually automate that process and get more value out of the images using techniques like OCR. So we're reading data out of the image, but also then running the models to look at things. So let's look at asset inspections first. A lot of IQ Geo customers do asset inspections today, whether it's a regulatory or non-regulatory inspection cycle. So we're taking images and attaching it to a digital form, which is all great, but we've got the ability to automatically populate the fields in the form based on taking an image. So now you train the model for, let's just take a wooden pole, for example, we're looking for pole lean. We're looking for a wood rot, woodpecker holes of a certain size, corrosion on the cross arm, damaged equipment. We can train the models to identify the same things that the inspector is doing on a digital form, but pull it through the image. And then you've got both the form and the image as a record. Take that a step further. If you're doing these inspections on an annual basis, you've got a view of the asset over time. So we can start to run models over time and actually see degradation over time. So it turns into this very proactive condition based assessment rather than this one time assessment of the asset. Some of the things we're working on that I think are really cool. So, you know, taking your phone using both the imagery and some phones have LIDAR capabilities. Not many people know that we can start to do the measurements of the pole, the lean of the pole, identify the attachments. So under the neutral, see what telco and fiber attachments they are, make sure they've got the right clearances as well as the assessments there. And doing that on two sets of data sets, we're seeing a lot of customers move towards, especially some of the larger ones, looking to procure their own drones. So we can do this at a point in time or as a more of a, like you mentioned earlier, Brandon, as a batch process, send the drone out. You've got span imagery, you've got structure imagery, and then run them through similar models. You know, you also mentioned different equipment. There's lots of different equipment types of installations that we could extend this to. Whether it's the utility assets we do in pictures of as-built, transformer installations, EV charging stations and things like that. Also in gas and water, there's lots of assets and installation scenarios where we can apply the same technology. And if you think back to the last time we talked, we discussed your concept of geospatial work execution. How does this capability fit into that larger vision? I think it's a significant enabler. You know, when we talked last time, we talked around the value of having the network model, native mobility and work laid on top of that. So we're attaching the work to the assets on a simple to use device working online and offline. And today, most of the time that's an iPad. What we're seeing is just as a natural reaction, field crews are starting to use their issued phones to do work. If we embed AI techniques, which is your visual AI, into those exact same workflows, and then it's not a different application. It's just embedded into the inspection workflow or the damage assessment workflow or the outage workflow. We start automating the population of some of these forms. We start automating the actual identification of things out in the field, which should lead to more quality, more in time. And then over time, we can start adding on even more technology and more techniques. So our vision at IQ Geo is to think of this now as a touchless interaction to your application in the field. Similar way to you may interact with large language models today. So you're speaking a natural language. It's basically prompt based into the application. Maybe the best way you can explain it is a scenario. Imagine that you get down to a poll and you say IQ Geo, where am I? And the application responds, hey, Brandon, you're at poll 1234. This poll has an open inspection order. Would you like to start that work order? Yes. Please take an image of the asset. You take an image and it identifies, oh, this is a broken cross arm. Do you confirm? Would you like me to do a follow on work order? That's the vision of where we want to get to is an application that is spatially aware of where you are. It understands the context of where you are in relation to the assets and any open work around it. And understands how to identify what it sees, put it in the context of the work and automate any follow up to other systems. So that sounds like a really powerful place for utilities to get to. What do you think are the steps to get from where we are now to that vision? That's a really good question, Brandon, because I think I would ask the same question. So today IQ Geo, we have a field mobile solution. Like as we mentioned earlier on the geospatial execution, we are mobilizing all of the GIS data layered with other work from asset management systems. All in easy to use iPad works online and offline. And we're seeing customers use that for different use cases from locating an asset tracing up and upstream downstream, asset inspections, outage mobility, all different types of use cases. How do we get that now to the vision of the future? So we are actively working on consolidating the application all into one easy to use mobile application. So taking what we have in Netlooks. So Netlooks already has techniques like OCR, voice to text, has the ability to train models against it. But that's on a separate application right now. So step one is embedding these techniques into a single mobile application. And step two from there, then building workflow specific agents. So, for example, we've been working on a prototype, what we call the pole agent. And the pole agent is able to go and through imagery is able to identify the type of pole, the attachments on the pole that we just talked about. We'll take that a step further and identify inspection criteria or damage assessment criteria into these poles. We also done some prototypes around layering larger language models on top of our applications to see how they react. And those results have been really, really promising. That gets us into the prompt based interaction rather than selecting an asset, selecting a form, pull down menu, dropping things in. So these are the steps that will get us there. I think this will happen a lot quicker than when we went from paper to first digital workflows, because the technology and enabling technology exists today to build that. And within IQ Geo, with the acquisition of Diplomatic and Netlux, we've got the opportunity to blend them together into a single app. That's what I see as the steps to get there. And I think the first step we'll see is that mobile application with embedded AI into it. Well, thanks again, Adrian, for walking us through this. It really highlights how visual AI is changing the way utilities think about quality in the field. Thanks for having me, Brandon. I enjoyed it. As you can tell, this is a technology that I think has got significant value to our industry, and I'm looking forward to see where it takes us. And I think if I could summarize the key takeaway for today, I think it's that visual AI brings instant feedback to the field, ensuring from right first time installations and complete compliance documentation at scale. To learn more, visit IQGeo.com and stay tuned for our next episodes, where we'll continue exploring how AI and geospatial workflows are transforming utility operations. I'm Brandon Kurgan, and this has been Bite Size Electric. See you next time.