The Science of Collections: How PG&E Uses Behavioral Science

[00:00] When we first heard and we were suggested to do this conference, obviously as myself and Ishita coming down from Calgary, a couple of days in Miami in March was very pleasing to us, so we really are happy to be here. Today during the presentation there's three of us to do it. Myself, Mark Brown, I look after sales for Symend, and then my colleague who I work with on lots of our tier one client accounts, Ishita. She's got the largest job description in the world there, but really looks after our team that manages our strategy and our applied science for our clients. And then also really pleased and proud to be able to co-present today with Rakesh. I think all of you know Rakesh. When I looked at last year's conference he was on like every panel that was here and every client presentation, so I think he's a dab hand at this and will guide us through. But I've been working on a kind of weekly basis with Rakesh for over the last 18 months up until this point, so really pleased to be able to do this. So without any further ado, we'll move on.

[01:32] Seems that our code is SCIENCE, so if you want to receive a copy of this presentation, there's the code, there's the text number, and we'd love to hear what you think about it afterwards. Okay, what we're going to go through today for the next kind of 40 minutes is we're going to talk about who we are, both PG&E and Symend, the reality of collections for specifically the PG&E use case, what we've tried to set the bar to, how we move then as a real case study of the partnership that we've been going on. So from plan to action, what the challenge was, then some of the results — and the results really, we did our first major insights call on Friday, so the results that you're going to see today are literally hot off the press from Friday where we did it to the team — and then what we're going to do as we go forward from this case study.

[02:38] All right, hey thanks Mark. So like everyone else, you know, you talked earlier about the struggles of utilities being real, especially if you're in California or like New York. Our struggles are way real, right? We have some of the most lucid regulations out there that are super customer-friendly. You know, me going to my executive VP and saying, you know, right now our overall accounts receivable even coming off of COVID is four times higher than pre-pandemic stages. Sit down for a second here — during the pandemic we got as high as like 1.2 billion dollars in outstanding AR. We've been able to bring it down to right now, roughly around 750 million-ish still outstanding, and there's a lot of reasons why that's happening, right?

[03:33] Well, we have limitation on the number of customers we can disconnect. We only can disconnect 3.5 percent of our eligible population. I was kind of smiling yesterday when some of the utilities were talking about high bills and pay plans. We are mandated to give customers 24-month pay plans that they can break twice within that. So basically just kicking the can, and then once they are no longer qualified for them, we're mandated to give everyone a 12-month pay plan, right? We cannot charge deposits to customers. And I'll get into the details, but you know, for us, customers' balances were around like $200 pre-pandemic. Right now, average balance is $700. So you know, this is me telling my EVP, you know, like here's all the struggles. And he's a very great motivational speaker — he told me to suck it up, buttercup. And I'm sure a lot of you guys probably have that, right? Just get it done.

[04:40] So part of what we wanted to do is, you know, we wanted to really get a partner with us that was going to help us focus on not just reducing AR, but the struggles are real, right? Because the more text messaging you send, the more call-ups you do, that's going to drive calls with the contact center. So one of the big solves was: we want to do a lot of this stuff but we cannot drive calls into the contact centers. We're going to have to kind of measure them, so we need a lot more insights on what's happening to the contact center. On the left side you see, you know, all the regulatory stuff that we have to do. I mean, it's California, it's crazy. We have limitation on not only the disconnection rate, but we only can do 30 percent in a zip code, right? Which makes it even harder because, I mean, most of you guys know you have populations, so you have unfortunately zip codes and areas that are more delinquent than others. And again, this is the commission's way of saying, at least in California, disproportionately you're not disconnecting the lower income customer base. So that's why they kind of put all these restrictions on us.

[05:51] But you know, the challenge is, again, no different than anyone else — legacy system, right? We can't get anything done. Our heavily customizable system that we have right now, you know, you can't touch it. Because if you touch the system, just testing any new product within it, it takes forever. And then, you know, our system by design is not configured with all of these changes. So kind of stuff what we want, and this is kind of my base requirement: it had to be a one-stop partner. What does that mean? Well, you know, we've talked to a lot of people and there are a lot of folks out there that are like, "Hey, we can do propensity modeling for you guys," or "We'll go to your customer and tell you exactly..." Okay, cool. Then what? Am I then going to find another partner to send text messages, and then a third partner to do dashboards for me, and a fourth partner to do this, and who's doing the analytic piece of it? I don't want that. I want someone to say, "Here's my delinquent accounts, you manage them for me," you know, with my requirements.

[06:50] The other big one was: must not drive contact center calls, you know. Help me figure out who's calling, why they're calling. We're gonna have to figure out different levels of solutions for them. And you know, you talk a lot about generative AI segmentation — we need that, right? We have to be able to segment our customers in a way, so again, don't segment it to me and give it to me. You segment it, you do something with them, you just let me know how well that's working or not. And then, you know, something that's going to be ongoing improvement, right? So I wanted that. We wanted a system that kind of looks at the customer behavior. So this is where Symend was really intriguing for me as a partner, was where it goes away from a little bit more of the propensity to pay. So you kind of know who's paying, but the behavioral science piece is, well, why are they not then paying, right? What is it that we're gonna have to change in the wording?

[07:44] So these are some of the big different things that we had. We want to improve our cure rates, reduce our inbound calls. We have like 13 different programs that we're offering the customer, so we want to integrate all of that into a solution, right? And it was a lot. And also at the same time, you know, we wanted to do it where we could test this out before integrating it into our system. So we kind of really had to figure out — and I'll show you in later slides some of the challenges that have happened. And you know, the reality is, as all of you guys know, with disconnection your cure rate is like 90-whatever percent. Like, you know, that's the best lever. But if you don't have that, what is my next best option?

[08:28] So just a really quick introduction to Symend and what our ethos is. We believed when we formed the company in 2018 that the traditional collections problems that Rakesh has just kind of highlighted there were not able to be solved by the traditional collection solutions available in that one-stop methodology. And we also believe firmly in the USP of behavioral science to move customers from inaction to action. Now I'm not going to try and steal any of Ishita's thunder as a senior behavioral scientist there, but that is the key, and the productized behavioral science is throughout our business and our solution.

[09:17] So we know that there are something like — and I've been working for Symend for six years now, working with collections leaders every single day — and that number in the middle, that 400 billion messages going out each day, was when I started was probably around about 200 billion. And so that message frequency is just getting more and more and more. So the ability to get the right message to the right customer at the right time with the right call to action and the right behavioral tactics becomes more tricky. And so being able to solution for that is difficult.

[09:50] We also know that beyond just the digital messaging or the omni-channel messaging, there are also numerous other channels where customers are making decisions every single day. So from the moment you wake up to the moment you go to sleep, you're making something like 35,000 decisions each day. Now we can just do the simple math and know that we can't make those decisions with thoughtful time and process. So what we do is we simply come back into what we'd call heuristics — in Ishita's world, "rules of thumb." So we just make decisions on what we've done before, what we think it's going to be, or whichever is the easiest for us. Or we just ignore making decisions in totality. So we don't pay our bill, we don't turn the TV off, or whatever it's going to be, because we just haven't got the capacity to make the correct decisions.

[10:45] And then finally, the last part of that puzzle that we've tried to solve is that in collections specifically, people do not always make the right decision for them. They'll make a small payment to kind of make themselves feel good by achieving a goal. That payment that they've made to one card or one utility or to their telecom provider, whichever it's going to be, may not have been the right decision. It may not have been the nearest bill that's due to them. It may not have been the one that's most urgent for them. It may not have been the one with the most consequences. So people make incorrect decisions within collections. So giving them the correct education, the correct empathy, the ability to clear their bills in the correct time, is really key inside that collections journey. And the last element of that is, once they start engaging in the collections journey, we need to change the behavioral outreaches as their behavioral indicators give us information. So that's what we try to do inside that delinquency journey. The rest of it is just kind of what you would expect: we only work with the most highly regulated industries — telcos, financial institutions, and utilities — and collect payments, reduce OpEx, and keep our clients' customers happy.

[12:12] All right, so Mark and Rakesh really set the stage of what the problem was, what the challenges were, what the boundaries were in which we had to operate. And where I come in and my team comes in is the how — specifically, how do we use behavioral science, or behaviorally-informed messaging, to change how a person acts when they see a message come in from PG&E?

[12:39] So there's a really simple version of kind of what this looks like on the screen here, and I know the text is a bit small so I'll tell you what it says. But on these two phone screens that you see here, we'll start with the left-hand one, the generic message. This is the kind of message that most customers, or utility customers, are used to getting from their providers. So it says the person's name, Sarah, "This is a reminder to pay your overdue amount," and then it says the number to call to take action on that. And there's nothing inherently wrong with this message — it's just inert. So it's not actually, if a customer is feeling anxious or overwhelmed or feeling stressed out about their financial situation or their bills, this kind of message isn't necessarily going to be what motivates them to change that behavior. And if they're feeling particularly anxious, it might actually make that worse.

[13:34] If you look at the message on the right-hand side, it's just a very simple example of what a behaviorally-informed message could sound like. So it's the same customer name, it's the same amount due, but let's just start with that first phrase: it says, "Sarah, you're usually on time with your payments." And now what this sentence is doing is something really particular. You might have heard of it — it's called cognitive dissonance. So it's really highlighting that gap between who that customer thinks she is, the identity of that customer — "I'm usually someone who pays on time" — and it's highlighting that gap between who they think they are and what the status of their account actually is. So that gap between who they intend to be and who they actually are in this moment. And that gap can be really motivating. It could be the thing that motivates that person to click that "Make a Payment" button.

[14:27] And again, this is just one example. I'll talk through some more. But something that I'm really proud of in this partnership with PG&E is that we were able to evolve PG&E's messaging strategy from three messages to 260 unique messages. And these messages are based on where someone is in the delinquency cycle. You can imagine someone who's only, you know, three weeks past due for the first time should probably be getting different language and different framing from someone who's six months past due and is frequently delinquent. It's also based on the persona or archetype that that person fits into, which I'll also go into more detail in future slides. And it's also based on the timing, and how the customer is responding to our outreaches in that delinquency journey.

[15:29] So I'll spend a moment on this slide as well because it's really foundational to what my team does and what Symend does as a whole. An assumption that I've heard from a lot of teams that I've worked with, both in utilities and other verticals as well, is that customers who won't pay can't pay, full stop. And that might be true for some customers, but there is a significant proportion of people who aren't paying for other reasons. And that might be things like shame, it might be avoidance, it might be feeling overwhelmed by the hundreds of messages and emails and texts they're getting every day. It could be feeling overwhelmed like, you know, a situation is just snowballing so they just feel like anything that they're going to do isn't going to be making a big difference anyways.

[16:27] Whatever that reason is, what I'm trying to get at here is that there's a group of customers who can engage but the friction of engaging feels worse than the friction of avoiding or ignoring that message. And that's who we're trying to address. So each of the behavioral tactics on the slide, as well as the dozens of others that we use, address a certain mental or psychological barrier or a facilitator that we can harness in these messages. So cognitive dissonance on the last slide was an example. Something like loss aversion is the concept that we feel the pain of losing something more strongly than we feel the pleasure of gaining something. So you can imagine that would really affect how you would want to frame messages to your customers, whether you're leading with consequences or offers. Even something as simple as empathy — just acknowledging that things are hard right now or you might be in a tough situation — can be what lowers that mental barrier enough to get someone to open that message.

[17:35] So here these are just some examples of kind of how we can address disengaged customers who might have the ability or willingness to engage, but we just need to meet them where they are. And before I pass it back, just the thesis of this, which I'll touch on again and again as we go through this, is that we can get customers to feel safe and engaged with an empathetic approach while also getting strong business outcomes for the provider, for the utility. Those two things don't have to be in tension. They don't have to be mutually exclusive. And that's something that we'll show you in our results slides later on as well.

[18:13] Okay, moving forward. So we can talk a little bit about now how practically we set up this engagement with PG&E and the work that went into it. And so after we'd completed the standard sales cycle, gone through the contracts, the onboarding, the security reviews, the model governance reviews — all of that stuff that is required to make sure that you have a solid foundational partnership that's trusted between the two teams, between the client and the vendor — we then really started working as a team. We set it up, I believe, tremendously for success by doing lots of discussions, weekly meetings with the team, setting up a correct project team that would work on this.

[19:08] And then really, as we moved into the kickoff, I'm trying to think of the day that we started. Where it was kind of from start to finish, from onboarding to first messages out, we had — I think Rakesh said there's quite a strict target. And we really wanted to be live before 2026 started. So I think we started kind of October with a real onboarding and testing and such like. And then he was insistent: "It's got to be 2025 that we're live in." So I think our first messages were due to go out the 26th of December, which was just an awful day. But I think we actually went out, I think we were three days delayed — the 29th of December we actually started sending first messages out, which was good to do.

[19:56] And the reason we got there is we had a standard kind of onboarding process and implementation plan where we gathered the teams, the stakeholders together — the multi-threaded stakeholders that they have within PG&E: the IT, the engineering, the different communication teams, the contact center teams, etc. — so they understand what the project was, how we were going to deliver it, the results they should be expecting, and the points of contact. Then we spent — Ishita and her team then spent a very good amount of time actually doing the discovery: the myriad of plans that PG&E have, the business rules that they have to manage those plans, how we were going to get the data to manage those plans and report back the information that they would need to start making judgments.

[20:54] So we spent a lot of time doing that discovery, looking at how the IT configuration should be, what it would be for the short term, what it would be for the medium term, and then the long term to make it really efficient for all of us. We then started creating the orchestration for each one of those journeys. I think Ishita said around about 260, and so they had three initial journeys and we started creating them for short-term delinquency, longer-term delinquency, plus also the hardship plans and the AMP plans they've got. So we've created journeys for all of those, plus also the messaging, the omni-channel messaging that's required for that. All for Rakesh and his team to approve and edit through standard process through compliance and legal, through copy packages.

[21:40] Once they'd approved that, then our team got to work inside our platform creating what that orchestrated journey would look like, creating all the copy, getting it ready, testing it, sending out with a small amount of batches to warm up the IP and ensure that there's nothing missed in the testing. And then I think on the 29th of December we went live. We had a 10-day kind of warm-up plan to make sure that we hadn't missed anything. And then on the kind of 6th of January, I think it was, our evaluation period — I think was looking at 50,000 customers. So by the 6th of January, I think we're up treating 50,000 customers per month for PG&E.

[22:38] And then the last element of that is, as I say, on Friday we do weekly calls to update, but we also do a standard insights call, a larger insights call, that we did on Friday with these folks where we took them through all their data for what's been happening for the last kind of 60 days. And we're going to show some of those results and insights and how we would iterate from them in a moment.

[22:54] So I won't go into too much detail on this slide because Mark covered a portion of it in the last slide, but in terms of how we actually design strategy, it's really important of course to start with discovery. And I think this is one of my favorite parts, because like Mark said, we're looking at the regulatory stuff, we're looking at the compliance and legal stuff, the key dates, the key consequences. But more importantly, or maybe more in-depth, we're also trying to understand the texture of the customer relationships that you have already. So we were trying to see how PG&E's customers respond to the current engagements that they have, the different kind of programs that they have. How do customers in hardship programs respond? Because it's really important to try to capture that brand voice and preserve those customer relationships, because we're a white-label service. So these are messages and outreaches that are going out to PG&E's customers coming from PG&E. So it was really important to us to keep going through cycles of review and approval and revision, because nothing goes in front of these customers until we have PG&E's sign-off on it. So that was a very important part of our process and something that is really important to help build a personalized and bespoke strategy rather than just generic message templates.

[24:40] So here's an example of what a strategy would look like. So here's just a 30-day example. We treat PG&E's customers for over 300 days. So you can imagine how many of these they had to approve. But essentially, I'll highlight some of the main levers that we can pull in terms of the engagement campaigns. So the first thing here is the channels. So we have email and SMS. The second thing that you'll notice is that we have intentional gaps between our messaging. We're not messaging people every day. We're also not messaging them on a repeated cadence like every second day or every third day. And that's because over-messaging and under-messaging can both lead to those avoidance behaviors that I was talking about earlier.

[25:31] So it's really important to be able to decide that cadence based on a person's archetype, based on how they're engaging with their previous outreaches. And the last thing I'll highlight here is the behavioral concept or behavioral tactic that we use for each and every outreach. And it's not necessarily one — sometimes we can use two or three in a single outreach. But the main thing to highlight here is that each message is building upon the last one and building upon the entire journey. They're not messages that you can just pull out and plug in at random. It's part of an entire coherent journey, because like I said earlier, an early-stage person, early-stage delinquency, is very different from late-stage delinquency. So we're really trying to build that as a journey through the playbook and through the campaign.

[26:27] So, you know, one of the things — and I figured you guys caught this — but we went from three standard text messages to 260. Again, three to 260 in the pipeline. And those guys, you know, text messaging, right? I'm sure you know what a pain it is to change those messages and all that stuff. So we actually went to our legal and marketing team and said, "Here's my 260 messages. Approve them all," right? No, but when it was said and done: "When are you gonna send them?" "I'm not sure exactly what time frame, but does it meet our standard? Is it flowing the way we want it?" So once you got everyone's approval, then it was cool — we're all set. Now we can do it any time. It's not going to a vendor each time to say, "Now I want you to update this, I want you to update that." We already have our 260. I'm sure it's gonna grow too as we get better, but that was one of the keys.

[27:16] But the other things I wanted to test out personally was, I don't just want early stage. I want to see if it works for late stage or not. What does that look like? So we did do that where we said, we're gonna do the early-stage strategy. I want to test it out if we're gonna do this for late stage, if it works or not. But we also wanted to test it out for our various payment arrangements, right? Hey, can I have maybe different messaging and kind of see uptake in the kept rate if we do this?

[27:42] The other big part for us was, and we're getting a lot of good data on this, is really doing a good champion-challenger. So as Mark mentioned, we are about doing three tranches. We're gonna have a total of 150,000 customers that are gonna go through this. But we took the same criteria for another 150,000 customers and we're not doing anything with them. We're monitoring them, right? And from there, that's where I was gonna point out that great data we're getting back, right? Because now we're able to see — one of the biggest things we saw was like, no matter what, these customers are gonna call back. Either they call back at the earliest part when we text message them, or they're gonna call back once they get their notices or they get disconnected.

[28:20] So the calls — we were able to prove out to our contact center that we're not really driving that many calls to you guys. These customers call you anyway. Now the flip side of this is, the customers that we were able to connect earlier and they still called you, the handle time was, I think it was like an 18-second difference. That might not sound like a lot for some of you, right? But you take 18 times five million calls we get, it quickly adds up, right? So now we started making that really good business case for them to say, this is how it's working.

[28:52] So those are the really big ones. I mean, again, everyone has kind of different plans. We have payment arrangements and budget billing, but that was one of our biggest things. I wanted to measure all of them. I didn't wanna just do one thing, the next thing, the next thing. I'm like, if I'm gonna get this in front of my teams, let's just do all of them at different stages so we can get the details on those.

[29:11] So I will say the complexity, like I was mentioning earlier, the data complexity — so all of this is not automated, and that's how we were able to get it up and going. We're still doing this behind the scenes, batching, et cetera, because again, we wanted to test it out, right? We don't wanna fully integrate this because you guys know, right, what a pain it is to do anything with IT. So we're like, okay, we'll do it manually. I just wanna make sure it works. I wanna prove it out. So we started manually.

[29:36] With the contact center, we gave them updated training, but one thing we did do differently with them was, I said, "You know what? You're gonna push back on this? I'll take the calls. I wanna get the calls back into my center." So now we're able to see these calls coming in. So we set up a little team within our instance. They contacted, fine, brought them back to me. So I set up like six agents that are getting these calls. I can see what they look like, what the impacts are. But we also started tracking from a customer transaction perspective, because the fear from the contact center and others were, "Hey, if you're gonna send all these messages, you're gonna upset these customers." And I'm like, "Yeah, but if they pay." But they were like, "No, no, okay, fine, we'll measure that." Because again, if you're at the contact center, they care about their customer transaction scores. That's what they measure. So we started doing that.

[30:26] So part of that was also, again, getting all the teams together, laying out our plan and how we approached this. And that was how we were able to kick this off really fairly quickly with the team. The other part of it — it does help. At some point, about three years ago, they moved our credit organization under our contact center. So that really helps with some friction between us and the contact center and our web team, because now we're all reporting to the same EVP.

[31:02] Okay, so we're just gonna go into some results and what we co-presented last week with PG&E. So I think as we said before, when we started off, what we always tried to do is obviously increase payments as kind of KPI number one. Every single client asks us to do that. Specifically for PG&E, you've looked at the complexity and the length of the payment plans that they offer and the regulatory compliance. So something that's really important there for PG&E is to keep people onto those payment plans and keep them making those regular payments. So that was really important to them. And then also, like everybody else, reduce the inbound calls. We didn't want any increase in inbound calls. We had the separate contact center. It was very carefully managed — how we do that ramp at that first stage to make sure that there wasn't a drive-in of a large amount of inbound calls. And also, I think maybe we didn't mention on this, but we actually, from our strategy, we actually removed the outbound calling. So it's a digital-only strategy to look at how that would prove.

[32:18] The results I'm gonna show you on the next slide — obviously we've abridged — but as we're going through the trial, we like to work on maturity of that data, of where the accounts are. So when we jointly showed the data on Friday, we really focused on that top set of results because that's at 80% maturity, 80% through their journey. So we have kind of 80% confidence that these numbers are tracking. Clearly, as we get to the bottom bars there, we're only at 15% maturity of people going through their journey. So it's still early results. We know that they can change.

[33:00] We know that some, when we build our strategy, our messaging, it's not a one and done. It's built in a really thought-out journey way where the messages build on each other and move different people towards different outcomes. And so when you start really heavily extrapolating with immature data, it can actually, if you try iterating from that, it can actually lead you to some results that are not informed by data. So we try to really focus on the mature data if we can.

[33:38] We took some details out of this slide. This is kind of almost the executive summary slide for ours. We took some commercially sensitive stuff out of it for PG&E. But for month one and month two, which is where we're at so far, we've been treating something like 50,000 accounts with an omni-channel digital strategy. On the day 30, the most mature data, where we're at 80% where we've got the real confidence, we're currently tracking it around — from the PG&E champion strategy against the Symend challenger strategy — we're currently, I'll use the word, "beating the champion" by around 18.8% with 80% accuracy, which has led to something like $1.9 million in extra collections for PG&E.

[34:32] You can see we're on the latest strategy, we're around about 1.7% down on that. However, that's a 120-day journey, so we're only around about 20% through that journey. So it's very immature at the moment. And then the February, the slightly less mature data, we know that it's kind of tracking, we're around about 40% maturity, and we're already around about $0.4 million above on collected payments above the PG&E strategy. And actually the day-120 journey, from some of the iterations that we've already done inside there — once again, our caveat, it's early, it's immature, so I'm not gonna make any judgments — but it's there. So in total, net value on January is around about $1.5 million. February, we're already tracking at about $1.5 million. So for every account that Symend is treating above the champion, we're currently earning PG&E around about $30 from what their collections process would be.

[35:43] All right, so now we can talk about insights. This is also — oh, I know I said discovery is one of my favorite parts, but this is one of my favorite parts as well. So just to level-set here really quick, something that I say often is that when we look at the results, like what Mark showed us on the previous slide, those aren't just outcomes, those are also signals on what we can do going forward, how we can iterate and the direction that we'll go in going forward.

[36:14] So I'll explain kind of what's happening on this slide, because I know it's a bit busy. So just to orient you, at Symend we use a delinquency archetypes framework, which is essentially just segmenting customers, creating personas for customers. And those are organized on two vectors. The first is readiness to pay, which you can see on this X axis down here, and then we've got capacity to pay. And those are pretty self-explanatory. Capacity is just the — do you have access to the funds to be able to make your payment? Readiness to pay is your willingness to engage.

[36:55] So I really wanted to highlight some insights that we can pull out of here. On these graphs, I have email open rate, email click rate, and payment rate. And when we look at all of these together in terms of how customers are behaving after they get each of these outreaches, we can start to pull out some group-level insights that are really helpful in terms of informing where we'll go next with our experiments.

[37:20] So something interesting that we can pull out right over here is, in those high-readiness groups, we see a dip in payment rate at around the same space, for around the same outreach for both of those high-readiness groups. So when we went and looked at the actual outreaches that were going out at this time, we saw that this is the first time that customers in this delinquency journey get offered alternative payment arrangements, pay plans, or financial assistance options. So we can see here that these high-readiness customers, they have pretty strong payment rates, but as soon as they're offered those alternative payment arrangements, that payment rate drops. These customers are willing to engage, but maybe those pay plans or deferred payments are an easier or less difficult path to take.

[38:14] So going forward, something that we could experiment with is maybe something like offering the high-capacity, high-readiness customers pay plans later, because they don't necessarily need it right now, or we can change how we're framing those pay plans. So we're specifically trying to target the customers who actually need it in that early stage. Another example here is our higher-risk customer group. This is the low-capacity, low-readiness group, so it's the highest risk. They tend to take three to five touches before we see them make a payment, and we see that pattern repeat twice, each time we get an influx of customers in this group.

[39:00] One more thing that we can pull out from here is this high-capacity, low-readiness group. They are engaging the least. They have the lowest email open rates, the lowest email click rates, but if you look at those email open and click rates in that line graph, they're almost on top of each other. So that means that almost every time someone is opening an email from this group, they're also clicking it. So what we can see from that is that maybe the barrier for this group is getting that first open, because they'll almost always convert to a click.

[39:35] So an iteration that we proposed for this group is doing a subject line test, an A/B experiment, where we can try to target these specific behavioral tactics that are more likely to get this customer group to actually open emails, because that's the biggest barrier that we see for this group right now. So that's just an example of how we pull out insights and how we continue to use the results that we get, even early results like this, to keep pushing and keep iterating, because it's really important to keep evolving to meet the customers where they're at, because the customer journey, the customer mindset doesn't remain stagnant, right? Someone's financial situation changes, their life situation changes, and with that, we need to be able to evolve our strategies to continue to meet them where they're at.

[40:29] Okay, so to conclude our presentation, I'm just gonna invite Rakesh, who's been our partner, stakeholder, champion, and protagonist for the last year, to kind of come up with any lessons learned, and then finish our presentation with his last thoughts, before we invite any questions from the audience.

[40:54] Yeah, and I wanna open it up to questions for you guys, like what you guys thought. I will say from the experience perspective, it has been where our big goal was, how do we wanna get this up and running quickly? It's not where we need to be, but it's getting good data, right? I mean, you're seeing it. You saw the slide where initial stages, but it's like, yeah, customers who are like 120-plus days, maybe text messaging's not working for them, and we're getting more insights into — those are the customers we were offering pay plans, and they're signing up for pay plans, and they have the highest break rate. So now it's like, well, maybe that's something I shouldn't offer them, and I'm gonna just do different kind of messaging for them, right?

[41:33] Those kinds of insights that we're getting just from this experiment, and measuring this at different levels, is really helping us kind of think through what kind of messaging, what kind of offering we want, because those are the ones that are calling the contact center, because they don't qualify, and then they do this. So we can take this data information back to our contact center folks and say, look, yeah, we're gonna create journeys that are — again, one of the big goals was to not drive calls to the contact center, right? Through this, as we get more sophisticated, we can really start tracking who's calling the contact center, what time are they calling in, what's being effective.

[42:06] And, you know, this is not gonna solve everything, right? And which, I 100% am going with this like open-minded, right? I mean, it's not like, oh, this product's gonna go and not everyone's gonna start paying because they were just waiting for that text message, right? We all know that, right? But we are seeing good incremental changes and updates on it, and the insights, and then we can go further along with reducing the call volume. To me, that's gonna be a big win, so we're communicating to the right customers. So with that, I'll open it up. Any questions?

[42:41] Q&A: You said something like 260 messages, right? So do you foresee any issues or concerns with the amount of messaging pushing out your collection timeline at all now? And could that be a reason for such a large debt, where it's taking you longer to collect on that customer because you're giving them more opportunity? And are you gonna choose any aggressive paths for certain customer accounts, where we're going straight to the point of termination as soon as I can, to prevent those bills from continuing to bill every month?

[43:10] Yeah, I would love to go to the aggressive path, but remember, we have our cap, right? So these customers that we're testing out, ideally what we wanna get to would be to integrate these messages as part of our collection cycle. So like you said — "So you're not prolonging the cycle?" — No, we're not prolonging it. For the champion, which is our current process, they're just going through the normal process, right? But the biggest challenge we have at PG&E right now is we never get to a disconnection, because we have like over a million customers delinquent, and right now, because of the cap, the most we can disconnect is like 190,000 a year.

[43:49] "How much of your arrears are associated with that 190,000 that you're disconnecting?" Yeah, so like most people, when we disconnect our customers, it's based on, we call it the oldest debt and the largest dollar amount. But again, with us at PG&E, we have so many different funnels that they have to go through — consumer protections, pay plans, AMP protection, zip codes — but by the time we get to them, we are disconnecting the largest ones. And what we saw was, again, that's where we were at the $1.2 billion, and now it's just steady at the 750 for the last three years, because we just can't get through all of them quick enough.

[44:50] "Did you do any analytics around which day of the week is the most effective to send your texts or your emails? Are you looking at that by customer?" That's a really good question. The first consideration for days of the week is the kind of send days that the client has. So, you know, some clients don't send on weekends, some clients, there's holidays and stuff like that. But in terms of analytics on the specific days, we've definitely done times of day that get the most results. And we do see that in terms of days of the week, like earlier in the week or later in the week, we do see differences. But our strategy is more — it's important for our strategy to be variable in terms of when we're sending out outreaches, because when it becomes a predictable pattern, that's when we see poor outcomes and poor engagements, when people know exactly when to expect that message. So an important part of the strategy is to keep it relatively feeling random for the customers.

[46:05] "Great presentation. It hits home, that's for sure. 260 templates — that's incredible. We have four, it's been four for three years, it's brutal. But that's why we have a new vendor coming in. So how often are you sending a text message to a customer? Our legal team is pushing pretty hard that it can't be more than one in 10 days."

[46:37] In terms of the frequency, we aren't restricted by PG&E to that degree, so we're definitely sending text messages more than once every 10 days. It's usually every few days, but like I said, we try to switch it up. Sometimes it'll be two days in a row, sometimes it'll be a gap of six days. From a legal team perspective, what we said was we're sending text messages to avoid the disconnection for the customer, so it's more of a preventive thing. And per that, it doesn't fall into the TCPA and other regulations because of how often in frequency, because this is — if you don't do this, the service is gonna get disconnected.

[47:47] "Are you guys monitoring opt-out at all on these? Are you seeing an increase with the number of messaging you're sending? Are certain customers opting out more than others?"

[48:03] Yeah, we do monitor opt-outs. We monitor opens and human opens rather than just machine opens. So we're not seeing any increase in opt-outs. We wouldn't expect to see that because our messaging — I think it's important, where we talked about the 260 messages there — our methodology is not to spam any customers. It's very much the opposite, actually. Those 260 messages are across multi-year campaigns as we've showed. We don't just send messages across one campaign. That's built for those different programs that we have, plus also to make it easier for the compliance and legal team when we create that initial package of messaging, we want to allow iteration. What our teams do is we already pre-plan for iteration of that period. So we'll create the package of messaging, make it very simplistic for the client to approve, because it's very simplistic. It's done in your tone and brand. But it also allows us then to build a sandbox of approved messages that, once we start doing that testing and going back to weekly insights calls with the client, saying, "This message isn't quite hitting home — we think it should be moved here, or we think we should maybe change the subject heading." It's already approved. It's already in the sandbox. So we're able then, with the client's approval, to swap one message for another one. It's not done on a weekly basis. It's very controlled. So as I say, we're very data-led, but I certainly wouldn't want to give the impression that we're creating 260 messages where they used to have three so we can spam. We send fewer messages. I don't know a client that we don't send fewer messages to, once our orchestration is embedded, than was with their original one.

[49:44] "So understanding that there's limited interaction between, let's say, automation and what you're doing today — are the agents aware of all the texts that each individual consumer is getting? So they know how to respond."

[50:09] Yeah, so thank you for the question, it's really good. I'd answer that in a couple of ways. The first thing we did — Rakesh and team were really quite concentrated and clever on this — they wanted the contact center team to be aware of this trial. They wanted them to be aware of what the implications were and also best practice for dealing with the customers. So part one, myself, Ishita mainly — I kind of do the royal "we," but Ishita mainly — we went up to Sacramento, I think, just before Christmas and we trained the contact center staff. "This is what we're doing. This is the behavioral messaging. This is how we're expecting the customers to call in." So it was a very collaborative thing. I think it was a wonderful day to be honest, where we were with those folks. So that was point one.

[50:58] And then point two, the platform itself has different export capabilities plus also a customer trail. So the agent is able to, whether it's account number or a different identifier, bring up that messaging that that customer has received, see it in the exact render they've seen it, see what other messaging they've received in that journey as well. So as they're having the discussion with the actual customer, they're able to see the digital, the omnichannel digital messaging, whether that be email, SMS, or if we enacted generated voice, they could be seeing what the transcript of that voice was as well for the warm handoff. We haven't done voice in this trial so far, but we could do that too.

[51:53] Well, great, thank you. That was fantastic. Thank you very much.