Using Machine Learning To Reduce Climate Emissions In The Physical World - A Chat With PlutoShift's Prateek Joshi

Prateek Joshi: 0:03

So we are at a point where we go to a site. We can reduce that water consumption by 15 and a half million gallons. Right. So that's just a start. That's just the average site. If you look at what we've done for electricity and chemicals and water, PlutoShift, if you scale it up globally, we have the capacity to reduce the carbon footprint or offset the carbon footprint of 268 million people every year

Tom Raftery: 0:29

Good morning, good afternoon, or good evening wherever you are in the world. This is the Climate 21 podcast, the number one podcast, showcasing best practices in climate emissions reductions, and I'm your host global vice president for SAP, Tom Raftery. Climate 21 is the name of an initiative by SAP to allow our customers calculate, report, and reduce their greenhouse gas emissions. In this Climate 21 podcast, I will showcase best practices and thought leadership by SAP, by our customers, by our partners and, by our competitors, if they're game, in climate emissions reductions. Don't forget to subscribe to this podcast in your podcast app of choice, to be sure you don't miss any episodes. The zoom one is there is a backup. Okay, here we go. Hi everyone. Welcome to the digital. No, this isn't. I, I do two podcasts. I actually do three podcasts and I'm supposed to be doing a fourth, but anyway, that's neither here and there. We'll try again. Hi everyone. Welcome to the climate 21 podcast. My name is Tom Raftery with SAP and with me on the podcast today, I have got my special guest Prateek. Prateek, would you like to introduce yourself?

Prateek Joshi: 1:43

Yeah, first of all, Tom, thank you for having me here. Really excited to have this conversation. I am Prateek Joshi, founder and CEO of PlutoShift. We are a company based in Palo Alto, California, and, pretty much through my entire professional career. I have built machine learning products and, uh, built machine learning systems that can, process a variety of data. At PlutoShift what we are building is data platform for industrial sustainability. We'll talk more about that, but that's, me, briefly.

Tom Raftery: 2:17

Okay. And why, why Pluto shift? Why machine learning? Why sustainable problems? You're you're trying to solve?

Prateek Joshi: 2:26

Yeah, I wanna start with machine learning first. I've been pretty much spending, my professional career in building virtual, infrastructure, meaning, images, text eCommerce, basically machine learning has done a lot of good in the ether. And through this process, I ended up writing a few books on the topic, ended up building products used by millions of people. What I realized is that while machine learning is ubiquitous in the, in the virtual world, like search engines. We use it every day. It's been around for a long time. They are heavy users of machine learning and we just, we don't realize it that much. We just use it. It just works. And we move on. In the physical world it's just not as ubiquitous as it could be. And that was the core motivation behind. Behind doing this is how do you bring machine learning to the world of physical infrastructure and, uh, within that, I started looking at how can we make this useful to the world? Like, how can we help, the Earth your fellow human beings? How can we help be better? And that's where, the topic of water came into picture. I, I grew up in a, in a small town in Southern India. A dry town, lot of water problems, like many I'm sure many towns in India. The water topic stuck with me. This was a chance to dive back in. So Pluto shift is a combination of a professional interest, which is machine learning and my personal interest, which is water. And, uh, we started with how can we help physical infrastructure? Not waste water. It's a precious resource it's worth saving. And how can we start that? So we started there and now it has expanded to how can we help physical infrastructure, reduce, their resource consumption, meaning energy chemicals, water, right? How do we, how do you help them reduce their carbon footprint? So that's, how it evolved over time.

Tom Raftery: 4:23

Okay. And what kind of solutions are you coming up with for these problems that you're finding? I mean, What are you able to do that couldn't be done without machine learning?

Prateek Joshi: 4:36

let's, I'll take a real example. Let's say you are a company that manufactures beverage and you have a lot of membranes, hundreds of membranes in series and parallel. And, and what you do is you buy raw water from the city, uh, from the, the government and you clean it. And then you use the clean water to make your product. Now through this process, there's a lot of, uh, wastage that happens like simple example would be if you buy a hundred gallons from the city, maybe 60 of that gets converted to useful product. What about the other 40? It's just, it just gets. Lost because there are so many inefficiencies in the system, like, you know, you don't track where it's going, the membranes foul, the membranes, stop functioning. So many reasons now, what happens in status quo is that are 200 membranes. The person, the operator comes in at nine o'clock in the morning and they do a round robin method, meaning they go to the first one. They see if it's doing okay, great. That's fine. Let's go to the next one. And once the, once a membrane is checked, it takes weeks or months for them to come back to it, to do another check. And during that time, if that membrane consumes five X more electricity, or if that membrane just stops functioning, you won't know for quite some time. And that's where all the resource wastage happens. With machine learning instead of doing a round robin, you come in and the system, the product that tells you, okay, membrane number 79 is where you need to look because it started clocking up five X more energy since yesterday. So you gotta take a look, right? So you kind of, it. Surfacing the right problems at the right time, detecting events of interest, predicting what's gonna happen in the near future. That's where machine learning. It's almost like a co-pilot that works with you to help you get to the problem faster so that you don't end up wasting electricity and chemicals and water. Right? So that's, that's one example. I'll take another example, many companies, they use, water to make their product, right. So mostly I'm gonna talk about chemicals. So let's say you're a company. You use a lot of water to make your beer and Ketchup or, or some physical parts. And, uh, the water is now contaminated. And before you just dispose it off into the environment, You gotta clean it, right? It's contaminated with chemicals. It's, it's just not, you're not allowed to do it by law. So what you do is, um, you have this giant treatment plans where dirty water goes in. The company has to clean it and the clean water goes out, but it has to happen at a certain speed because you are manufacturing continuously, it keeps going on and on. So you gotta keep up. So what happens is, dirty water comes in, you gotta put chemicals in it to clean it, and then it goes out. Now, if you put less chemicals than required to save money. Then you will be violating EPA laws because it won't be clean enough. If you put more chemicals, just to speed it up, you'll be wasting money. You'll have to spend more than required. So this is a very real time data problem where you need a machine learning system that can tell you, okay. At two o'clock today, you need the dosing is X right at two o'clock. The dosing is Y so. By doing that you'll stay compliant and you won't waste a lot of, of your capital and that's where, these, um, very useful outcomes happen when, when machine learning is infused into the, into the process.

Tom Raftery: 8:10

Okay. And you mentioned physical infrastructure as well. Is that a separate use case or is that part of this?

Prateek Joshi: 8:18

It's all part of it. So all of this happens in physical infrastructure. So when I say physical infra, I mean, membranes and pumps and cooling towers and heating systems, boilers, condensers, clarifiers. So all these physical assets, are part of physical infrastructure that, companies operate to make their product.

Tom Raftery: 8:37

Okay. Okay. And you are using machine learning in, various different ways then to make sure that you are eliminating wastage from processes, for example,

Prateek Joshi: 8:51

Yes, in our case machine learning is being used to take in all the data. And I say data it's temperature, pressure, flow rates, basically operational data. So all the data flows into the system and what the product does is it detects and predicts events of interest. And then that is showed to the operator who can then make decisions, based on what's actually happening or what's about to happen. Within physical infrastructure, the goal is to identify a specific process that can benefit from it. So in this case, you said resource waste stage that's, that's the primary outcome. And the eventual outcome is reducing your carbon footprint. There are multiple ways to do it the way in which we do it is we, reduce the wastage of resources so that per unit of product that you produce, you'll use less electricity, less chemicals, less water. So that translates to reduced carbon footprint, which is, um, a big outcome that we drive.

Tom Raftery: 9:52

Okay. Very good. Very good. And you mentioned, you know, you're sucking in a lot of data from systems. Is this part of the solution you give as well that you provide the sensors or are you taking it from existing sensors that your customers might already have in place? Or is it a combination of the two?

Prateek Joshi: 10:14

Yeah, we take the data from existing sensors. So that's part of our, our qualification process, just to understand, does PlutoShift, fit the needs here. So in this case, uh, our customers at a minimum, they need to have, some kind of sensors, sensors should be generating data. And, uh, if that's already happening, that's where we go and we plug in and, uh, we start, we get to work.

Tom Raftery: 10:39

Okay, that makes sense. Yep. Sure. and for your customers, is this a big, mind shift change that they have to get over, you know, for taking advice from essentially a machine?

Prateek Joshi: 10:57

Oh, that's a great question. And this is something that we have been thinking deeply about in terms of product design and you're absolutely right. Why should an operator who's been doing this for 20, 25 years? Why should they listen to a machine? So to address this, user behavior, question, what we did was we don't, we don't really. We're not black boxy. That was the whole, insight is the point is we look at data. Data comes in and we show clearly what our reasoning is, right. When we tell them, Hey operator at nine o'clock membrane, number 79 is where you need to go and they can just click. Okay. Why, why when day 79, then we tell them, oh, it's eating up five X more energy than yesterday. They're like, okay. Why, why does that have you can click a button and it says, oh, it's eating a five X more energy because of normalized Perme pressure or it's eating up because of X. So basically the multiple levels of why it makes it very clear why we are recommending or prescribing an action. So I, so once you provide the, the two levels, two to three levels of reasoning, it's it becomes very friendly and like, oh, okay. I get it. Yes, of course. I need to go to membrane 79. So compare that. With something, a machine just tells them, Hey, membrane number 79, do X. They won't do it. No. Why will they do it? I mean, no humans are not built that way. So we, we don't, we like to know why we do stuff. So that's how we, we have addressed this, issue.

Tom Raftery: 12:31

Okay. Very good. Very good. And Are there particular industries that, are more suitable to this solution or do you think it's kind of something that could be adopted across the board?

Prateek Joshi: 12:43

Yeah, we have consciously, uh, approached industries that are a little more suitable and more. open and more than anything, they have urgent problems that need to be solved. So food and beverage great example, uh, the processes that, that are continuous processes. They need insight on a, on a very continuous basis, as opposed to like discreet manufacturing where, you know, there is, there's more room in between. So that's one, food and beverage chemicals, another great industry where this is very, very relevant. So we approach industries where carbon footprint or resource consumption is a problem. And there are low hanging fruits where you go in and you, you help them get an immediate win. Right. And that's, that's a good combination for us. What it does is one, it, it helps them get over the hump hey, does this work? I, I don't know. The, the first initial success template just really opens up, uh, people's minds. And then they go from. The first deploying at the first site to the next one to 50, to a hundred and more. So that is how we have, uh, approached this market.

Tom Raftery: 13:52

Okay. Yeah, that makes sense. What about industries? Like. I mean, you mentioned process so steel or cement or any of these ones that have a very high carbon footprint. They, they sound like they might be ideal as well, no?

Prateek Joshi: 14:07

Yeah, definitely. It's we that's definitely on our roadmap. We just happened to not happen, but we, we talked to a few different companies and, companies in food and bev, and chemicals. They were open to trying out and deploying it. So, but yeah, absolutely. You cement steel, definitely industries that one very high carbon footprint and, it's definitely on a roadmap. The other thing to keep in mind is the company's data readiness also impacts our ability to work with them. Right. The, as you said, cement and steel, yeah definitely industries worth pursuing, but in terms of data readiness, I think food and beverage and chemicals, they're in a, in a good spot where high carbon footprint, they have the data, they have the intent and it just, it just comes together. But yeah, definitely our next industries are, are what you said. We are. We are on our way there.

Tom Raftery: 15:02

Okay, superb. Superb. And what about, companies being able to measure the return on investment of any investments they do in sustainability.

Prateek Joshi: 15:13

Yeah, it's a notoriously hard problem. And if you talk to the leaders, uh, like the VPs C suite across these companies, uh, they'll tell you that one of the biggest problems is if I put in a dollar into my sustainability initiatives, uh, I don't know what's coming out of it. And the reason is twofold. One is just measuring the ROIs, measuring the operational impact that is difficult by itself. And two, even if you were to measure it, it's almost impossible to get multiple stakeholders to agree that something happened. When I say stakeholders, I mean the operators, the managers, the finance people, the sustainability people, and the C-suite, all of them agreeing that, okay, this year we saved 12 million gallons of water. It's almost it's it's everybody has their own calculations in their head. They have their own agenda. They need the numbers to say something specific. So yeah, so the, the twofold reason makes it very difficult to measure the ROI. And that's why we've designed, um, our method around addressing these two key issues is one when you go in, when we, as in when PlutoShift goes into a company to deploy. We first measure the status quo of the last 12 months. Meaning without PlutoShift, here's what happened? Do we agree right? Until there is agreement on what they've already done in the past. There's there's no starting because it's just pointless. If, if you don't have a benchmark to compare, uh, it's just, there's just, there's no point starting. So step one, we established status quo and more than anything, more than us. We want the multiple stakeholders to agree that, okay. Yes, these are the numbers that we all agree upon. And then we get to work and say, okay, in the first 90 days, here's how we move the needle. And then we, we make sure that everyone has access to information on how we are computing it. And, uh, they have information on what's happening on, on a weekly basis. So that way. When we show up 90 days later here to show heads what happened? People are already in agreement. So that is how we have tackled this, issue of if you, if you put in a dollar, X comes out of it. And we usually like to keep it at about four to five X, meaning you put in a dollar in sustainability, it should get four to five X out of it. And the way you measure it depends on the company, but it's mostly a resource consumption, right? Same time. Last year we spent X on chemicals. Now we are spending Y great that that's the Delta. The carbon footprint impact is another key issue because it's kind of soft, meaning. You, if you reduce their carbon footprint, how do you measure the monetary value of that? So if every company has their own own measure, so we need, we need to kinda understand how they put a, dollar number on their carbon footprint. And, uh, yeah, so we include all of that. And then we present the number which, which, works for, for our customers.

Tom Raftery: 18:19

Okay. And you are US based, is that your market or are you expanding beyond the US in terms of your market?

Prateek Joshi: 18:29

Yeah, we are based in the us. All of our customers are headquartered, in the US, but what we have done is our customers took us once, once we deployed here and once they're successful, uh, they took us to south America and Asia. So at least for now, what we are doing is we'll start in the US and if an existing customer expands, and once we deploy at that Asian facility or south American facility, we'll do it but we're not gonna go approach a brand new company in Asia just yet. We are just not set up to, to do that because we need, people on the ground who can make sure the product is successful and the customer is successful. So that's what, we are doing right now.

Tom Raftery: 19:09

okay. That makes sense. Yep. Sure. Are there any successful customers that you can talk to?

Prateek Joshi: 19:18

Yeah, two of our biggest, customers Anheuser-Busch, Dow Chemical. We have been working with them for, for a while. They've been successful with, with our product we're expanding. So, uh, one, one for each Anheuser-Busch the use case is centered on water, water, recovery. Meaning if you buy a hundred gallons, how much of that can be converted to useful product in their yeah. Beer in their case beer, And also they're dealing with, water stressed regions as well. Meaning, in some, in the us, we, we just talk about dollars in efficiencies, but there are many parts of the world where water itself is not available. So every drop that you have, you gotta, you gotta use it. So here, for example, in the us, if you lose some water, fine, it'll go back. You'll buy more water from the city. If you are in, in south America or Asia or Africa, there's just, even if you wanna buy, even if you wanna spend money, there's just no water. So which means it cannot meet your throughput requirement for the day. So it's, it becomes extremely important that every gallon of water you have, you convert that to, useful product. So that's one. Dow chemical, another example, uh, it's it's focused on chemical usage. So basically, they, as, you know, Dow is a huge manufacturer of, of chemicals and they use a lot of water and the goal is how do you stay compliant while minimizing the spend and all of this translates to carbon footprint. So they use a product to make sure that the dosing is right and they keep track of it and they detect and predict events of interest so that they can avoid spikes of any kind, because spikes are the killers. Like basically if the system stays up and keeps spiking that's the system goes down. Great. Not great, but you'll know that, okay. The system went down, we need to fix it. It's not doing anything wonky. If the system stays up and keeps spiking every 24 hours, that's a giant bill waiting for you. Right. And that's and 2022, that's the bigger problem then systems just randomly going down. So yeah, we are working with, a few other customers as well, but those are the big ones.

Tom Raftery: 21:27

Nice. Nice. Yeah, very good. And where, where to from here? What, what are your next plans?

Prateek Joshi: 21:33

We are a point where, we are serving multi-year contracts with the customers. It's enterprise grade, we have great reviews. So our plan is to expand, and more than anything, we want the, the onboarding to be very, very fast. Meaning, uh, once you go to a new customer, the speed at which you can deploy, we are working on making that a lot better for our customers and also we're working with existing customers to, use our product and more of that in more of that infrastructure. So like existing customers, expanding with the product and getting new customers within these, um, these verticals that we're working in. So that's the, that's the plan from here. And really we based on the results that we have delivered, right? So we are at a point where we go to a site. We. We can reduce that water consumption by 15 and a half million gallons. Right. So that's just a start. That's just the average site. And, uh, if you look at what we've done for electricity and chemicals and water, PlutoShift, if you scale it up globally, we have the capacity to reduce the carbon footprint or offset the carbon footprint of 268 million people every year. Right. And that's just based on where we are based on today's product, what we've already done, the amount of reduction, that's what we are. And that excites me a lot. Meaning if you scale this up, if you, look at our target market and if every company uses our product, then every year, 268 million people, the carbon footprint is offset and that's, uh, that's a very exciting proposition. So that is our Northstar. We keep an eye on that as we continue to work on, on the product and working with our customers to achieve that.

Tom Raftery: 23:15

Okay, super, super we're coming towards the end of the podcast now, Prateek. Is there any question that I haven't asked that you wish I. Or any aspect of this, we've not covered that you think it's important for people to be aware of?

Prateek Joshi: 23:30

Yeah. Yeah. Oh, I'm gonna take a slight detour and, and talk a little bit about, why this, and why do I do this? Why does anybody choose a specific thing and just keep going? There's a very nice Japanese concept called Ikigai. I K I G A I I'm sure you've heard of it. Very it's very popular. and really at the heart of it, that's what I wanna highlight is finding your Ikigai or finding your purpose for existence. It just helps you minimize the friction. Meaning if you're doing what, you're, what you like to do, what you're supposed to do, what you enjoy, what gets you paid because you gotta pay rent. So, if you're on the sweet spot, the friction is very, very minimal, which means you can just keep going for a, for a long time. And when you do something for the long time, Compound interest kicks in, meaning it just, just compounds just by virtue that you didn't didn't die, or you just kept doing that thing for a long time. Compound interest kicks in and that's when magic starts to happen. So my, simple recommendation is in my case, it happens to be machine learning and climate problems and PlutoShift, right. But I think finding your Ikigai is a fantastic life hack. It'll just minimize all friction in your life and just, you just keep going for longer. So, yeah.

Tom Raftery: 24:48

Yeah. A hundred percent behind you there. Agree. Absolutely. Yep. Fantastic. Great Prateek. That's been really cool. If people want to know more about yourself or about PlutoShift or any of the things we discussed in the podcast today, where would you have redirect them?

Prateek Joshi: 25:04

You can visit PrateekJ.Com to learn more about me for Pluto chef, we can visit Pluto shift.com to know more about the company I'm active on LinkedIn. So if you have a question or a comment or any feedback, or if you just, just wanna brainstorm, reach out to me on LinkedIn, I'm happy to chat. One last thing, I write a weekly newsletter on machine learning where I just discuss concepts, topics, how to build ML products. And it's at PrateekJoshi.Substack.Com. So yeah, I many avenues to, uh, to reach out to me.

Tom Raftery: 25:39

Fantastic. Fantastic. I'll put those links in the show notes so everyone will have access to them. Prateek. That's been fascinating. Thanks a million for coming on the podcast today.

Prateek Joshi: 25:47

Perfect. Thanks Tom. I really enjoyed this conversation.

Tom Raftery: 25:51

Okay, we've come to the end of the show. Thanks everyone for listening. If you'd like to know more about Climate 21, feel free to drop me an email to Tom dot Raftery @ sap.com or connect with me on LinkedIn or Twitter. If you liked the show, please don't forget to subscribe to it in your podcast application of choice to get new episodes as soon as they're published. Also, please don't forget to rate and review the podcast. It really does help new people to find the show. Thanks catch you all next time.