![How APS dove head-first into AI [partner content] — Open Circuit cover](https://megaphone.imgix.net/podcasts/7ec68a0c-e24b-11ef-b277-7b3d3c3c73b9/image/796dd136afa5782eae8ad683dd946957.png?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress)
Loading summary
Transcript37 lines
- [00:03]
A
This is partner content from Latitude Studios. Most people now interact with AI every day. They use chatbots to summarize documents, answer questions or generate ideas. But inside industries like healthcare, manufacturing and energy, a different kind of AI is gaining traction.
- [00:22]
B
There is horizontal innovation or horizontal AI, cloud, OpenAI, Gemini included. But also there's sector specific AI, which is what we call vertical AI.
- [00:33]
A
Vertical AI is very different from the tools most of us use. It's trained on the data, workflows and constraints of a specific industry. It's narrow in scope with deep context. For Karthik Murthy and his team at Bijely, the focus is utilities and the enormous amount of intelligence hidden inside energy data.
- [00:51]
B
We build the foundational data layer for a utility, disaggregate and identify for each data point what is happening behind a household. Which appliance is your household running? Have you connected an ev? Is your H Vac saturating? Is there a pool pump? And then we take this rich data and build multiple applications for the utility.
- [01:11]
A
Karthik Murthy is a Chief Growth Officer at Bijly. He's worked at the intersection of software, AI and industrials for almost 25 years. And he says the next step change is coming as these two, two forms of AI, the broad horizontal tools and the deep sector specific models start to work together.
- [01:28]
B
So each industry has its own vertical AI. I think now's the time when those models, the mature models, are combining with horizontal AI to produce incredible output.
- [01:41]
A
That convergence is starting to play out inside utilities. Venkat Nimala is the director of Digital Transformation and Enterprise Architecture at Arizona Public Service. He's spent more than two decades working in the utility sector through many waves of digital transformation. He says this moment feels different.
- [01:59]
C
I think especially in the last 18 months, there's a lot of change, a big change that's happening, a lot of pivoting going on. The foundational models have improved so much. The big change for me was when I saw AI very seamlessly being implemented into the day to day tools that
- [02:15]
A
we are working on for a long time. Digital transformation in utilities often meant digitizing existing processes or making data easier to access. Now, Venkat says AI is making it possible to turn that data into decisions.
- [02:30]
C
Earlier it used to be more about transforming a process with bringing it from paper to making it digital or something like that. But now it's able to take actions on the data that we already have and bring a lot of insights so that it can actually directly influence the decisions that we make and the workflows
- [02:47]
A
that we are inside utilities. This AI convergence has a practical impact. For APS and Bidgley, it started with a very specific helping explain customer bills. But as that work expanded, the question got bigger. How do you scale AI across the entire utility for almost any application?
- [03:07]
B
So how do you democratize innovation and bring it to a place where the utility is able to kind of pick it up and build whatever they want in the format they want, in the place they want, and in the pace they want?
- [03:18]
A
The answer unlocks something much bigger than a single customer service tool, a new way to bring AI closer to the data and make those insights usable across the utility. Utilities everywhere are searching for new ways to manage rising demand and ease the pressure on customers. They're also exploring how to implement AI in the most effective and secure way possible. Through conversations with utility leaders and Bijley experts, this series explores how those tools are being applied in practice and what they reveal about the future of the grid. In this episode, our first in the series, Stephen Lacy talks with Venkat Nimala of APS and Karthik Murthy of Bitchly. They discuss how a tool to explain high bills became the starting point for a much broader conversation about data governance and AI across the utility.
- [04:13]
D
Venkai, you had a very specific entry point into adopting AI, which was explaining high bills to customers. Why did that process need improvement?
- [04:26]
C
When a customer's bill is high, it's an emotional thing. They're either confused or frustrated on why what is going on. And with utilizing AI visually was able to bring in the disaggregation for the energy consumed so that we could tell them what really caused that high to happen. Instead of having a conversation around, you know, your bill is high for whatever reasons. Now we are able to explain on what exactly caused that increase to happen, where you saw usage increasing, what is going on and we're able to guide them to be better and efficient. The other reason I would also say is whenever there's a high bill, I believe customers already do some kind of research because they're frustrated about it and when they give a call to the call center, the call center persons are really, you know, caught off guard but not having enough time to do the research on the account and being able to respond to them. I think these kind of tools also helped us to better explain a high wheel situation as well as guide them to what they could do to reduce the bill. Not just about your consumption is high, but we are able to pinpoint and tell what date and time what appliance would have caused it. That's a big change.
- [05:33]
D
Yeah. So Karthik, when APS is out looking for this solution, where do you step in and how do you begin the implementation and to bring disaggregation to this process problem.
- [05:44]
B
So over the last 12 years, we have invested in taking large amounts of meter data like 30, 50 million plus and doing disaggregation at scale with high accuracy. So it helps you understand what is happening from a DER perspective, from a household perspective, but also from an efficiency perspective, like which appliance is efficient? What is the lifestyle profile? We were like, okay, let's really give the insights to the people in the front line, be it the customers or be it the call center folks. Where first the solution was implemented as a SaaS solution, as in the data would come to us, we would kind of disaggregate it, turn it into an application which the call center folks can use in a very seamless fashion. It's designed when a call comes, it kind of pulls up what's happening at a customer, what happened last year, this time for this month, and helps you correlate between temperature and energy use, helps you compare total load, which has happened effectively giving the right information in the right frame and format so that a very healthy conversation can happen, which educates the customer, but also keeps them happy at the end of it and leaves them satisfied with the response and gives them tips, tools and tricks. So I think bringing that to life from an implementation perspective, but also from an impact perspective, was our first engagement before we started expanding into other things.
- [07:07]
D
Okay, so this worked well. You decided to explore expansion, but as you did this, you ran into some hurdles with data availability. What were they?
- [07:18]
B
I would call it less about problems, but also think of aspiration and scale. So first is the scale of the data, right? How do you make sure different use cases get lit up in doing so? So from a security and sovereignty perspective, shipping AMI data out every time for every use case and doing it differently is not the right way to do it. How do you containerize the model? How do you run it in the utility? How do you keep the data in a way which helps them innovate with the latest technology in hand? Those were the larger challenges. But I think the early part on implementing one solution was fine. But expanding the same for different applications, for different business units, in a way which is most meaningful was the larger challenge which we had to then focus on.
- [08:03]
D
Yeah, Venkat, speak to those challenges. And also when you started exploring broader use cases, what Were they?
- [08:10]
C
As Karthik said, when we started the journey, it was about providing these high bill related insights to the agents to be able to explain to the bill, et cetera. And then like I said, when I took it up 14 months back, we started to see this as a good data point in terms of good analytical insight into how our product is being consumed, which all appliances are using it at what date and time, how the patterns are, etc. There's a lot of information out of it. And then we quickly realized that I wanted to use this in our web chat interface or a enabled chat that we were building to be able to disaggregate the information to the customers. We wanted to use this into our load forecasting, grid reliability, grid resilience, etc. There are so many empty number of use cases where if you're able to disaggregate the information and be able to tell which appliance is being used when it might become a data point in different use cases. But this data could be used in multiple areas. That was the starting point for us to think data sovereignty is very key here. I can't be shipping data packet by packet to different all the models or all the vendors that are there chasing each use case. It might be prudent on our side to look at bringing models closer to the data as much as possible. And that was the turning point to see how this data could be used in so many other use cases. We were working on a power theft or revenue production use case and if the appliance disaggregation was there, we could tell whether there's a customer living there or not based on the seasons. We could see the consumption of air conditioning and other things. And when I went to Bisley and talked about how do we really get this data and then the conversation was okay, we could send me the data and then I can get it back. Then that back and forth itself is a delay and I'm not able to use it on demand when I needed it, et cetera as well for the consumption in last 15 days or whatever when it is built. That is when the conversations got triggered into what could visually do to give those models and deploy them in our cloud or in our environment so that these insights could be used across multiple use cases.
- [10:19]
D
Karthik, can you speak to technically how you do that? How do you bring the models to the data?
- [10:25]
B
First we containerize those models and we help them install it using the hyperscalers in the utilities environment. So what do I mean by that in APS's case they use Oracle cloud infrastructure. So these models don't necessarily now run only on visually, they can run on the Oracle infrastructure of aps. Second comes standardization of inputs which is what are the inputs these model needs? They need AMI data, customer data, weather data, demographic data. What are the inputs? How do you standardize it for them to run efficiently with the same quality and predictability which we have? Third is this is just the base where you're creating a data fabric and the data lives right next to all the other data which a utility has. So they will have SCADA data, outage data, vegetation data, and now the disaggregated data sits right next to it, which gives them the ability to to innovate at scale, which Venkat touched upon. So we not only brought the data from a security aspect into the utility environment through the cloud which they use, we built a layer which enables conversation, but also powers personalized applications and it powers all the agentic use case which you may need across the customer domain and the grid domain. So that provides a landscape for the utility to choose whichever is the high ROI path they want to take in their transformation journey.
- [11:50]
C
To add to that what's now possible, in my view, the biggest shift is that we are no longer thinking in an isolated use case situation, that there is one use case here or one different vendor for each use case and shifting the data to multiple vendors. Now with this, what has happened is disaggregation has now become part of our broader ecosystem. It's a good analytical data point that's now available, that's able to support our customer insights or bill analysis or even guiding customers with energy usage in our marketing programs or in our efficiency emails that we wanted to send. It's become a big part of that that's actually helped us to take insights to where it really matters to the business and the existing solutions. You can't just rep and replace all the existing technology. You need to be able to take these insights and data points to where it matters, where they're working currently as much as possible. And that's what we are able to do. We are able to connect the dots. We were able to bring these insights across multiple business processes that we look upon.
- [12:54]
D
And can you give me some specific examples of use cases that are emerging for you that are helpful for planning or operations teams?
- [13:02]
C
Yes, there are multiple use cases in terms of doing all the transformer and the load analytics to really understand how the load is currently being consumed, where it is going, what is Happening, get some diagnosis around it. There are multiple use cases around outage. We have use cases around grid resiliency, grid reliability as such. There are so many around propensity models, customer experience, enrolling them into different programs, being able to give interaction options for them to be able to interact with the data directly.
- [13:36]
D
So what does this experience tell us about like the way AI can scale in utilities generally? Are there any specific takeaways here that can be applied to the power business generally?
- [13:48]
B
A few patterns emerge for sure, right? One is 12 months back there was a conversation of should you use AI, how should you use it? Now that has been put to rest, like everybody knows the power of it. Now it's about how you unleash the power in a responsible way in a regulated industry like a utility. And the ones which are most ahead are treating data as the first area to cover. Like how do you have a unified data layer, a clean set of data as a foundational part to make a lot of their transformation happen. The second part is how are you not just procuring solutions which used to be historically like RFP driven? I'm going to buy the solution and the vendor is going to deliver to now what part am I going to build, buy and partner with? So clarity around that from a leadership perspective, which very quickly then translates into an operating model. What works best for your setup from a security operating model capability and people perspective. And the third part which is very, very important is the value realization story. What you do not want happening is start with a lot of intent and 12 months go by and there is no outcome or output which is good enough to be spoken about and reinvest in this exercise. So extremely important to move from pilot to production with a clarity on the highest value use cases. And there is multiple ways to do these things, but these are the key ingredients that need to happen to scale responsibly.
- [15:15]
D
Yeah. What about you Venkat? Are there any takeaways that you have from this experience that you think apply to other utilities?
- [15:24]
C
I think it will help to start really looking at a real business problem, then start to say that I'll do some AI work. You should never start from an AI point of view. You should actually start from solving a business problem, solve it and then you would learn where the data breaks and that's how you build the foundation and that's how then you scale up eventually. Data governance has always been there around, but not really given that importance. And AI actually fuel for the AI is the data it feeds on it. So it's very important for us to be able to connect the data, make it really ready for AI in a way so that it understands different aspects of the data and can consume it in the right fashion. These are very important things than chasing the technology.
- [16:13]
D
Going back to our discussion on organizational alignment with where does this push need to come from? Is it from the bottom up or is it from the top down?
- [16:23]
C
Venkat it needs to bottom up is always there, right? But their perspective is all about their productivity tools or something very tactical in, you know, that could help their business process, etc. I believe top to bottom is what is going to help to make it really transformative for the organization because then they'll be able to think of it at a scale of enabling people understanding for whom it's mandatory. What needs to be the expectations from a leadership point of view, how do leaders influence the mindset of the team members? Because not everybody is at the same spectrum. They would look at different aspects. They might be different in their journeys and how they look at this as well as AI, I think the fundamental for scaling it is also thinking of it from a process point of view, from an end to end process point of view and transforming that. That would only happen when, when it comes from the top, in my opinion.
- [17:18]
D
What do you think Karthik? Where are you seeing the opportunities inside utilities?
- [17:22]
B
Organizationally, the benefits of being centralized in the early stages, which is where we are, are pretty clear, right? Like as an organization you have a shared set of models and outputs which you have as an organizational asset. There is one cloud environment where these models run, there is one set of access, there's one accountable like owner of sorts. But also there's one roadmap of prioritized use cases which is then collected bottom up for sure. But it kind of tells the organization how to do this because one of the most important things from a utility context is you can't afford too much waste of things like duplicate SaaS. Spending inconsistency across department multiple duplicate things like this has an impact on ratepayers, it has an impact on speed and time to impact. But as long as these shared principles are put in place, it helps you get started and going on this journey. In the beginning it kind of helps to have a foundational framework, a way of working, a way of value capture, training, enablement and onboarding for a lot of these use cases.
- [18:27]
D
Okay, so you've embarked on this sort of centralized AI governance strategy. What does the future look like when AI is embedded in every core function of the utility, do you think it will be?
- [18:40]
C
Yeah, I do think every product vendor sitting out there or every deal that's going on, I believe AI is the heart of those conversations, right? Every vendor, every product will bring in AI into their solutions, but also they might be thinking very vertically around that particular product, etc. The story comes around when you look at more horizontally across the data across multiple systems. I do firmly believe that it's going to come up everywhere in every product. However, in the beginning stages people are really not that used to how you should actually develop or deliver an AI solution. How is it really different from a rule based to something more like that is generating. There's different testing methods that you should use for it, you should be caring about responsibility, how is it able to explain to develop trust, etc. That's all different. And with so much going on with every vendor bringing in solutions and things are changing also so fast. Every two weeks something that is new, I think it'll help to have initial days at least a centralized team that's able to bring all those learnings together. And as you get to scale then you could look at multiple models.
- [19:59]
A
Venkat Nimala is the Director of Digital Transformation and Enterprise Architecture at Arizona Public Service. Karthik Murthy is the Chief Growth Officer at Bijly. This is the first episode in a four part series exploring how utilities are using purpose built AI to solve operational challenges across the grid, customer programs and planning. To learn more about how Bidgley works with utilities through its utility AI platform, visit bidgley.com or click the link in the show Notes.