Meeting Title: Brainforge x Televero: AI Opportunities Date: 2025-09-17 Meeting participants: Scott_Harmon, Raywo, Uttam Kumaran, Amy Adams, Brian
WEBVTT
1 00:02:30.590 ⇒ 00:02:31.760 Raywo: Hey, Scott.
2 00:02:32.080 ⇒ 00:02:33.559 Scott_Harmon: Hey, Ray, how you doing?
3 00:02:34.130 ⇒ 00:02:38.029 Raywo: We’re good. We’re good. We’re keeping busy.
4 00:02:38.250 ⇒ 00:02:41.589 Scott_Harmon: I bet you are. Did you have a good summer? Or, yeah, summer’s over.
5 00:02:41.590 ⇒ 00:02:43.339 Raywo: Is… did we have a summer?
6 00:02:43.510 ⇒ 00:02:47.189 Scott_Harmon: Well, yeah, you’re right, it wasn’t that bad this year.
7 00:02:48.110 ⇒ 00:02:53.779 Raywo: It rained, like, every weekend, but… I don’t know. We just… I can’t believe we’re September…
8 00:02:53.900 ⇒ 00:02:56.749 Raywo: Getting ready for October now, this is ridiculous.
9 00:02:57.170 ⇒ 00:02:59.750 Scott_Harmon: Yeah, yeah, I know, it goes quickly.
10 00:02:59.920 ⇒ 00:03:03.069 Raywo: with the, I mean, Scott, probably the…
11 00:03:03.510 ⇒ 00:03:06.730 Raywo: A piece of noteworthy news is we…
12 00:03:07.520 ⇒ 00:03:11.909 Raywo: made the Inc. 5000 at, position 54.
13 00:03:12.310 ⇒ 00:03:13.190 Scott_Harmon: Yay!
14 00:03:13.380 ⇒ 00:03:15.890 Raywo: So, pretty, pretty good recognition.
15 00:03:15.890 ⇒ 00:03:19.940 Scott_Harmon: That’s fantastic. Congratulations. I… I love the model for the company, I…
16 00:03:20.540 ⇒ 00:03:23.409 Scott_Harmon: I’m a huge fan. Just… just think you’re…
17 00:03:23.840 ⇒ 00:03:27.279 Scott_Harmon: Thinking about it in a great way, and tell everybody about it, and…
18 00:03:27.640 ⇒ 00:03:29.329 Scott_Harmon: Can’t wait to see what you do.
19 00:03:30.120 ⇒ 00:03:30.970 Raywo: Thank you.
20 00:03:31.700 ⇒ 00:03:34.670 Scott_Harmon: Looks like we’ve got Utam coming on board.
21 00:03:35.130 ⇒ 00:03:41.089 Scott_Harmon: he’ll tell you all about it, but his team’s doing a great job on several different AI projects, and…
22 00:03:41.480 ⇒ 00:03:43.320 Scott_Harmon: Love to hear what you’re doing.
23 00:03:43.730 ⇒ 00:03:45.229 Scott_Harmon: Just kind of catch up. Hey, there he is!
24 00:03:45.230 ⇒ 00:03:45.960 Uttam Kumaran: You’re right.
25 00:03:46.160 ⇒ 00:03:47.330 Uttam Kumaran: Hey!
26 00:03:47.330 ⇒ 00:03:48.440 Raywo: Dude, how are you?
27 00:03:48.440 ⇒ 00:03:51.310 Uttam Kumaran: Hey, good, how are you? Good to see ya.
28 00:03:51.310 ⇒ 00:03:55.739 Raywo: Sorry, my Wi-Fi made me a bit choppy. I’m here in Chicago at a conference.
29 00:03:57.290 ⇒ 00:04:05.650 Uttam Kumaran: It’s… so we’ll see, it’s, I don’t know how good the conference Wi-Fi is, but yeah, awesome to be connected, and hey, Scott, it’s been a minute.
30 00:04:05.890 ⇒ 00:04:07.300 Scott_Harmon: I know, how you bet, buddy?
31 00:04:07.950 ⇒ 00:04:10.989 Uttam Kumaran: Good, good. Same old.
32 00:04:11.360 ⇒ 00:04:15.690 Raywo: Well, Amy’s joining at some point here, I don’t know… I don’t see her on.
33 00:04:16.500 ⇒ 00:04:21.750 Raywo: I just… I talked to her, so she’s going to join, but we don’t have to necessarily… weight…
34 00:04:21.860 ⇒ 00:04:28.340 Raywo: I mean, listen, I appreciate you reaching out. Excited to see any progress or breakthrough.
35 00:04:29.280 ⇒ 00:04:33.180 Raywo: That you made, We have…
36 00:04:34.420 ⇒ 00:04:39.130 Raywo: You know, we’ve been busy ourselves. We’ve probably implemented about 7 AI projects.
37 00:04:39.500 ⇒ 00:04:40.240 Uttam Kumaran: Great.
38 00:04:40.550 ⇒ 00:04:41.970 Raywo: As well.
39 00:04:43.050 ⇒ 00:04:52.259 Raywo: some of them are individual point solutions. We’re now moving to an AI platform internally that we could build and have greater control. But, listen, we…
40 00:04:52.540 ⇒ 00:04:58.899 Raywo: You know, we haven’t cornered the market on, you know, breakthroughs and great ideas, so we’d love to see what you got going.
41 00:04:59.140 ⇒ 00:05:00.350 Scott_Harmon: Such an impressive 7.
42 00:05:00.350 ⇒ 00:05:00.920 Uttam Kumaran: Yeah.
43 00:05:00.920 ⇒ 00:05:03.340 Scott_Harmon: About that, but.
44 00:05:04.170 ⇒ 00:05:07.949 Uttam Kumaran: What’s the platform that you guys sort of, like, arrived on, by the way?
45 00:05:08.700 ⇒ 00:05:12.110 Raywo: Well, it’s actually our own platform, building it in the door.
46 00:05:12.570 ⇒ 00:05:14.250 Uttam Kumaran: So, it’s gonna allow…
47 00:05:14.250 ⇒ 00:05:16.410 Raywo: A lot more customization.
48 00:05:16.720 ⇒ 00:05:21.950 Raywo: So, you know, we… We have worked with a number of people on point solutions.
49 00:05:23.390 ⇒ 00:05:24.040 Uttam Kumaran: Yeah.
50 00:05:24.040 ⇒ 00:05:28.129 Raywo: So, I think there’s… I’m gonna say there’s about 3 different
51 00:05:28.290 ⇒ 00:05:36.540 Raywo: point solutions, and I will say the platform that we’re looking closely at, Utum, is probably, It’s bland.
52 00:05:36.850 ⇒ 00:05:44.190 Raywo: B-L-A-B. Yeah. Land is a platform that we’re looking at, but we want to get greater control.
53 00:05:45.020 ⇒ 00:05:48.380 Raywo: We have implemented a number of voice agents.
54 00:05:48.820 ⇒ 00:05:55.540 Raywo: We have automated our scheduling, we’ve done some billing, all non-sexy stuff, but…
55 00:05:55.710 ⇒ 00:05:59.439 Raywo: You know, it contributes to our operational, efficiency.
56 00:06:00.190 ⇒ 00:06:01.729 Scott_Harmon: Which one have you seen the…
57 00:06:01.730 ⇒ 00:06:03.020 Uttam Kumaran: The bottle.
58 00:06:03.260 ⇒ 00:06:05.399 Scott_Harmon: Which ones have you seen the best impacts from?
59 00:06:06.700 ⇒ 00:06:09.479 Raywo: Really the patient outreach, so the voice assistant.
60 00:06:10.750 ⇒ 00:06:16.100 Raywo: So what we’re able to do, Scott, is rather, in addition to reaching out with text.
61 00:06:16.520 ⇒ 00:06:26.410 Raywo: we can reach out multiple times with a voice agent. We listen to the recordings. You know, I would say about 90% of the people don’t even know they’re talking to AI.
62 00:06:26.820 ⇒ 00:06:27.590 Scott_Harmon: Really?
63 00:06:27.840 ⇒ 00:06:28.860 Raywo: They really don’t.
64 00:06:29.910 ⇒ 00:06:34.770 Raywo: You know, some of the limitations that we’ve run up, oop.
65 00:06:35.320 ⇒ 00:06:42.519 Raywo: is taking credit cards. One of the platforms is working on that. I forget the credit card standard you have to hit.
66 00:06:42.830 ⇒ 00:06:44.079 Raywo: I forget the name of it.
67 00:06:44.380 ⇒ 00:06:45.290 Raywo: But…
68 00:06:45.550 ⇒ 00:06:51.979 Raywo: Taking credit cards to be able to complete the registration is the only… is the kind of the next thing that we’re trying to work through.
69 00:06:52.970 ⇒ 00:06:53.700 Scott_Harmon: Gotcha.
70 00:06:53.700 ⇒ 00:06:57.169 Raywo: And I’m trying to think of what that standard is, but I don’t recall.
71 00:06:57.170 ⇒ 00:06:57.850 Scott_Harmon: Gotcha.
72 00:06:58.340 ⇒ 00:07:03.969 Scott_Harmon: Well, Utam, do you want to… you should probably just, you know, give right an update on… you’ve been on a bunch of projects.
73 00:07:03.970 ⇒ 00:07:04.620 Uttam Kumaran: Yeah.
74 00:07:04.620 ⇒ 00:07:11.430 Scott_Harmon: I’m involved in one, so, you know, with one that he’s doing, but he’s doing others, so I’ll let you.
75 00:07:11.530 ⇒ 00:07:14.060 Raywo: Yeah. Catch. Oh, here, looks like, here’s Amy.
76 00:07:14.530 ⇒ 00:07:16.399 Raywo: Okay, great. Hi, Amy!
77 00:07:16.770 ⇒ 00:07:20.520 Amy Adams: Hi, sorry for the delay, I always have a hard time getting on Zoom.
78 00:07:21.420 ⇒ 00:07:22.020 Scott_Harmon: You what?
79 00:07:22.020 ⇒ 00:07:22.960 Uttam Kumaran: No problem at all.
80 00:07:23.310 ⇒ 00:07:23.869 Raywo: Where… where are you.
81 00:07:23.870 ⇒ 00:07:28.289 Amy Adams: I always have a hard time getting on Zoom, but no, I’m here, I apologize for the delay.
82 00:07:28.650 ⇒ 00:07:30.259 Scott_Harmon: No problem, good to see ya.
83 00:07:30.760 ⇒ 00:07:32.590 Amy Adams: Hi, Mary, good to see you!
84 00:07:32.590 ⇒ 00:07:34.669 Scott_Harmon: Yeah. Things good in Dallas?
85 00:07:35.360 ⇒ 00:07:37.260 Amy Adams: Very good, yes.
86 00:07:40.330 ⇒ 00:07:49.469 Uttam Kumaran: Cool, yeah, maybe I can share a little bit, you know, the kind of the impetus for, you know, thinking about y’all again is we started working with this company called Ellie Mental Health.
87 00:07:49.500 ⇒ 00:08:00.560 Uttam Kumaran: And it seemed very, very similar to a lot of the work that y’all were doing, and we’re actually coming and helping them, with a bunch of stuff on the data side as well, and then similarly, we have a lot of,
88 00:08:00.650 ⇒ 00:08:11.349 Uttam Kumaran: you know, AI work that we’ve done. So maybe I’ll just talk a little bit about some of, like, how the strategy has changed a little bit on the AI side, and I’ll just share a couple of
89 00:08:11.590 ⇒ 00:08:22.920 Uttam Kumaran: sort of slides that share a little bit of, like, how we’re thinking about AI adoption, and it’s really great that you guys have tried a bunch of stuff now. I think it’s probably clear what,
90 00:08:22.960 ⇒ 00:08:36.869 Uttam Kumaran: like, what’s working, and what, like, works, but you can’t get adoption, and what is sort of, like, vaporware, and, like, maybe the demo is great, but it doesn’t end up working in… in production. So one of the things that we’ve…
91 00:08:36.909 ⇒ 00:08:52.640 Uttam Kumaran: we started going to market with a lot of companies, is just coming in, and on the AI side, it requires a lot of discovery. And so we sort of have bucketed, our work, into a couple different areas in the AI side. So one is just, like, general workflow automation.
92 00:08:52.700 ⇒ 00:09:03.300 Uttam Kumaran: So this is coming and identifying a specific, workflow, identifying and documenting it end-to-end, and then building, basically, what is the
93 00:09:03.300 ⇒ 00:09:19.930 Uttam Kumaran: you know, AI-augmented version of it. Part of that includes, like, a ton of discovery and a ton of interviews with the actual stakeholders that are using the existing solutions. A lot of it is building SOPs. And then finally, there’s actually a huge measurement piece. So.
94 00:09:19.950 ⇒ 00:09:35.959 Uttam Kumaran: when we build… when we built originally a lot of the AI solutions, they would work in sort of a demo or prototype, but as you start to reach edge cases, as you start to sort of hit the limits of certain things, you need to be able to flag those and triage, and so we do a lot of evaluation.
95 00:09:35.980 ⇒ 00:09:45.519 Uttam Kumaran: And so what that means is we actually know and build, like, a dataset of all the correct answers, and basically we’re able to score the outputs of the LLFs.
96 00:09:45.590 ⇒ 00:09:56.599 Uttam Kumaran: But one of the challenges of language models, of course, is that they’re not deterministic, meaning they sort of can give you different answers, and so it requires a lot of tuning, not only on the prompt.
97 00:09:56.600 ⇒ 00:10:06.909 Uttam Kumaran: But on the data that’s coming in, on the integrations where it’s going out. And so, a lot of what we’ve done for clients is work really heavily with them up front to get
98 00:10:06.910 ⇒ 00:10:13.359 Uttam Kumaran: You know, a golden data set of the right and wrong answers, and then build, sort of, scoring for,
99 00:10:13.950 ⇒ 00:10:30.950 Uttam Kumaran: for workflows. Another thing that we also come and do within client now is just general, like, training on ChatGPT, on Claude, and so a lot of our companies, we’d walk in and, you know, they would ask us to automate a workflow, but their employees were just not using basic
100 00:10:30.950 ⇒ 00:10:47.939 Uttam Kumaran: ChatGBT or Claude to automate email writing, to automate, like, small menial tasks, and we found that, hey, like, that’s probably the earliest thing that you should try to chew off. It’s very easy to turn on. It’s, you know, you’re probably a couple training sessions away from getting folks onto that platform.
101 00:10:47.940 ⇒ 00:11:07.140 Uttam Kumaran: And starting to show them what use cases are best to use that. So, we also started offering, sort of training and sort of upskilling in that world. And then the last piece, you know, this is sort of areas where if we come into a company and they’re probably at the point where you were last time we talked, which is just, like, figuring out the
102 00:11:07.140 ⇒ 00:11:13.499 Uttam Kumaran: what opportunities there are, we started running workshops. And so this… these workshops are mainly to help you skip
103 00:11:13.500 ⇒ 00:11:28.910 Uttam Kumaran: you know, setting up an AI committee and it taking 6 months to arrive at, like, what the answer is. We get everybody in a room, and we sort of hash out, exactly what the opportunities are, what the challenges are in the business, and we arrive at, like, a few proof of concepts that we try
104 00:11:28.910 ⇒ 00:11:38.629 Uttam Kumaran: to then move into development. So, those are, like, some of the different ways that we’re doing things. A couple of things, a couple projects that, you know, come to mind that are interesting.
105 00:11:38.630 ⇒ 00:11:55.270 Uttam Kumaran: So as Scott mentioned, one of the clients that we collaborated on is with ABC Home. They’re, you know, a pretty big, you know, in Central Texas, and we’re, really augmenting a lot of the, difficulty that their customer service reps are having within,
106 00:11:55.270 ⇒ 00:11:58.940 Uttam Kumaran: you know, their office. So, these are things like, looking up.
107 00:11:58.980 ⇒ 00:12:13.650 Uttam Kumaran: basically being on the phone and having to look through tens of documents to find answers to things, we built them and have been continuing to build up a very sophisticated chatbot that not only takes in structured data from documents, but also takes in
108 00:12:13.650 ⇒ 00:12:25.250 Uttam Kumaran: Unstructured data takes in, you know, database information, and now we’re sort of building towards, like, customer information as well. So, these calls that can tend to get put on hold.
109 00:12:25.250 ⇒ 00:12:44.329 Uttam Kumaran: Maybe they… they have to, you know, really just get up out of their cubicle and ask somebody, and the customer has a bad experience, they’re now able to resolve these calls, you know, within… within the timeframe allotted. And additionally, now, the customer service reps can also pitch and upsell those clients. So what… what went… what was initially just a cost
110 00:12:44.330 ⇒ 00:12:55.049 Uttam Kumaran: center for ABC is now proving to be more of a revenue gain, where their customer service reps are able to actually, like, extract more revenue while they have the customer on the line.
111 00:12:56.100 ⇒ 00:13:07.030 Uttam Kumaran: And so that was directly worked with 8x8, which you may be familiar with, like, big for the call center… call center technology. So we’ve integrated directly into their APIs to pull transcripts.
112 00:13:07.030 ⇒ 00:13:18.169 Uttam Kumaran: We’ve… we’re loading all that data in so they can get a great view of which calls are augmented by AI, what parts of it are augmented, what are the outcomes from those calls.
113 00:13:18.170 ⇒ 00:13:29.160 Uttam Kumaran: And then additionally, we’ve also built in the feedback loop, which is that fact that their trainers, now look at those transcripts and look at the feedback and can update documents fast.
114 00:13:29.160 ⇒ 00:13:45.559 Uttam Kumaran: So if you were to ask a question over your knowledge base and the policy isn’t there, there’s nothing the AI agent is gonna do, right? And so there… in this company, we found there were a lot of knowledge gaps that weren’t filled out. And so we actually built this mechanism by which their trainers and the staff actually
115 00:13:45.560 ⇒ 00:14:00.779 Uttam Kumaran: whose task it is to create these documents, have an AI-assisted way of doing so, right? So we’ve kind of hit both sides, which is people are using AI to access the documents, and then also the documents themselves and the SOPs are being created with AI, kind of speeding up both sides.
116 00:14:00.780 ⇒ 00:14:14.060 Scott_Harmon: I’d like to… I’d like to just add just 30 seconds to that, because I… you might have seen this too, Amy and Ray, on some of your projects, but what… what I learned, and I… I’m not as deep in this, you know, it’s Utom’s team, I’m just kind of a advisor on it, but…
117 00:14:14.170 ⇒ 00:14:17.000 Scott_Harmon: The project he’s referencing
118 00:14:17.130 ⇒ 00:14:30.420 Scott_Harmon: you know, customer service, the problem is you’ve got to find knowledge to answer questions, and a lot of times in companies, knowledge is not… just not very well documented. It’s in different systems, and it’s different people’s heads, and…
119 00:14:30.630 ⇒ 00:14:43.850 Scott_Harmon: And it’s in a spreadsheet over here, or that spreadsheet hasn’t been updated, and it’s just… So, what was interesting is they started it as a knowledge management, project. And they said, can you help us take all this knowledge
120 00:14:43.990 ⇒ 00:14:46.539 Scott_Harmon: And, and just get it structured.
121 00:14:46.810 ⇒ 00:14:53.679 Scott_Harmon: And so we started by ingesting all these different files, and the first challenge we took was
122 00:14:53.790 ⇒ 00:15:08.080 Scott_Harmon: analyzing the knowledge data, we said, where are there gaps? Where are there inconsistencies? This spreadsheet said our billing policy was this. This SOP said our billing policy was that. We don’t have this for zip code this, like, incredibly
123 00:15:08.290 ⇒ 00:15:22.720 Scott_Harmon: mundane stuff, but it’s just the way we run our businesses. And it came… and there are something like 60 documents, you know, that it discovered that could be used by different people, and said, you know what, we can consolidate these into
124 00:15:22.840 ⇒ 00:15:25.780 Scott_Harmon: One master, well-structured document.
125 00:15:26.250 ⇒ 00:15:39.289 Scott_Harmon: And also, if there’s a gap, the EAI can tell people, hey, there’s a gap, you need to update, you know, you have it, this doesn’t exist, this knowledge for how to do this for these kinds of patients in this region, da-da-da.
126 00:15:39.420 ⇒ 00:15:40.340 Scott_Harmon: So…
127 00:15:40.620 ⇒ 00:15:48.439 Scott_Harmon: And then the other side of it is, of course, the customer service bot, which can now answer the question right. But, to me, the really interesting thing was…
128 00:15:48.740 ⇒ 00:15:51.549 Scott_Harmon: because it resonated with companies that I’ve worked for.
129 00:15:52.010 ⇒ 00:15:56.999 Scott_Harmon: You know, at any point in time, about half of what the company knows is written down.
130 00:15:57.280 ⇒ 00:16:03.980 Scott_Harmon: and you’re always going, oh, well, let’s form a committee to document, you know, this, and…
131 00:16:04.260 ⇒ 00:16:16.049 Scott_Harmon: And sometimes it’s not even… the workflow for how new knowledge gets approved isn’t even clear. Does Ray need to approve that? Does the CFO need to approve it? Does it need to be updated quarterly?
132 00:16:16.180 ⇒ 00:16:25.950 Scott_Harmon: it’s just a bit fuzzy. So I really thought that was interesting, and the CFO of that company is a friend of mine, Utam, I just saw him yesterday, and he’s… they’re really pleased, because
133 00:16:26.410 ⇒ 00:16:30.210 Scott_Harmon: They now know more about their own business.
134 00:16:30.630 ⇒ 00:16:33.249 Scott_Harmon: And where the gap… the knowledge gaps were.
135 00:16:33.380 ⇒ 00:16:37.259 Scott_Harmon: Like, they didn’t even know some of these knowledge gaps, you know, or they…
136 00:16:37.260 ⇒ 00:16:37.780 Uttam Kumaran: Yeah.
137 00:16:37.780 ⇒ 00:16:47.820 Scott_Harmon: And so, anyway, that was a learning for me personally, I think for the team. Utam’s team’s done a great job of it, and I don’t know if you’ve seen similar things where you discover
138 00:16:48.090 ⇒ 00:16:53.060 Scott_Harmon: To train the bot, you find out you’re missing data to train it with.
139 00:16:53.670 ⇒ 00:16:54.260 Raywo: Yeah.
140 00:16:54.640 ⇒ 00:16:59.780 Raywo: Let me go into something, though, Uten, let me ask you specifically, so… this…
141 00:17:00.900 ⇒ 00:17:05.710 Raywo: Let’s take the call center. So, you have a client calling into a call center.
142 00:17:06.480 ⇒ 00:17:12.359 Raywo: And you are essentially using AI to augment
143 00:17:13.210 ⇒ 00:17:20.920 Raywo: that call center. So, it’s not AI answering the call, it’s a live agent, but describe a little bit more, though…
144 00:17:21.510 ⇒ 00:17:27.400 Raywo: So, you’re gonna… let’s say I’m calling you right now, and I’m saying, listen, I need, I need pest control.
145 00:17:27.990 ⇒ 00:17:35.250 Raywo: Right? What’s happening… what’s happening in front of the agent at that time? What’s the augmentation look like?
146 00:17:36.670 ⇒ 00:17:39.689 Uttam Kumaran: Yeah, so I can just describe it… yeah, go ahead, Scott, go ahead.
147 00:17:39.690 ⇒ 00:17:56.209 Scott_Harmon: I’ll let you answer, but the short answer is, it’s a level 2. We started as, it’s a level 2 support agent, so it supports the support agents. We wanted to be very careful. They were very uncomfortable with having a lot of customer-facing stuff. They’re a very conservative company, so we started…
148 00:17:56.730 ⇒ 00:18:07.719 Scott_Harmon: with it being a Level 2 internal chat bot that an ABC employee… now, that could… that’s changing, but that’s where it started. And Utam, you could expand on
149 00:18:08.070 ⇒ 00:18:10.209 Scott_Harmon: That… the chatbot and where it’s going.
150 00:18:11.120 ⇒ 00:18:12.140 Uttam Kumaran: Yeah, so…
151 00:18:12.520 ⇒ 00:18:14.520 Raywo: Yeah, I’m sorry. Keep going.
152 00:18:14.520 ⇒ 00:18:31.309 Uttam Kumaran: Yeah, yeah, so we started off, basically, again, my goal is for this to get adopted, and so every time you implement a new tool or a new tab that someone has to go to, especially for, you know, the level of skill of the workers in the call center, it can be really, really
153 00:18:31.310 ⇒ 00:18:49.580 Uttam Kumaran: difficult. I mean, these folks are used to CTRL-F in 5 different Google Docs, hop over the cubicle, or, like, or, like, Google Chat a friend. That’s the level of, like, sophistication, right? So, introducing another tool in a workflow seemed a little bit overkill, and so one thing we started with, the constraint is
154 00:18:49.580 ⇒ 00:19:02.360 Uttam Kumaran: let’s meet folks where they are. So they’re already using Google Chat, they’re already in a call, they’re not able to solve a problem, they’re already sending a chat to another teammate. Right there is where we place what we call Andy, the AI agent.
155 00:19:02.380 ⇒ 00:19:08.430 Uttam Kumaran: And so you can actually chat directly in Google Chat with the agent. These are questions like.
156 00:19:08.430 ⇒ 00:19:22.549 Uttam Kumaran: who is the inspector that should go out here so I can schedule? This is what policies are covered under, this policy? What are our promotions right now? And again, as Scott mentioned, these are… this is, like, 60 to 100 pages of
157 00:19:22.560 ⇒ 00:19:40.549 Uttam Kumaran: documents. So, not… not very, like, easily consumable. Also, not very easy to control F over or find. And… and the other part was, this isn’t purely just, like, oh, every day was sort of relaxed. These folks are really stressed taking on
158 00:19:40.640 ⇒ 00:19:56.879 Uttam Kumaran: quite a bit of calls, and finding that their average call, you know, duration was so high, and they could never close these within the first call. And so, really, a lot of this was, one, reducing the amount of overflow that they need to have. Second was allowing people to resolve the call, like, within the first
159 00:19:56.880 ⇒ 00:20:13.199 Uttam Kumaran: call, not having to call people back. And then the bonus was, like, can you turn that opportunity when you have a client on the phone to upsell them on additional services? And so the use case is really, and the view from the CSR is they just have their Google Chat up, and they chat directly with
160 00:20:13.240 ⇒ 00:20:14.050 Uttam Kumaran: the bot.
161 00:20:14.500 ⇒ 00:20:16.989 Uttam Kumaran: Directly in their Google Workspace.
162 00:20:17.160 ⇒ 00:20:31.709 Raywo: Right. But is there empty… is… so let me just turn it a little bit more. Is there anything that is listening to their call and proactively raising up in front of them without the agent having to cognitively conceive?
163 00:20:31.710 ⇒ 00:20:32.360 Uttam Kumaran: Yes.
164 00:20:32.360 ⇒ 00:20:41.260 Raywo: question. Is there anything you’re raising up and say, hey, Ray, if Ray has pest control, has he done his annual,
165 00:20:41.680 ⇒ 00:20:48.499 Raywo: Termite inspection, or how about mosquito spray service? Is there anything that’s being generated?
166 00:20:48.950 ⇒ 00:20:53.790 Raywo: To augment that, like… like a cockpit display.
167 00:20:55.350 ⇒ 00:21:00.780 Uttam Kumaran: Yeah, no, this is exactly what we’re building right now for them, is sort of, like.
168 00:21:00.810 ⇒ 00:21:07.519 Uttam Kumaran: a standalone UI. And part of what I… when I propose this sort of something that’s outside of Google Chat.
169 00:21:07.520 ⇒ 00:21:21.419 Uttam Kumaran: One thing I told my team is this can’t just be a tit-for-tat replacement of that. This has to be something where the CSRs are going to want to use because it improves the process so much compared to the Google Chat. So what are additional features that we can consider?
170 00:21:21.420 ⇒ 00:21:36.800 Uttam Kumaran: And so part of this is, like, we just actually talked to 8x8’s team about 2 hours ago, learning about their real-time transcription, the real-time audio APIs, and sort of trying… one of the features, you know, we proposed is, can you build sort of a real-time cockpit?
171 00:21:36.800 ⇒ 00:21:43.459 Uttam Kumaran: And so there’s a couple of challenges there. One, you do need to have some type of audio streamer understanding what’s coming in.
172 00:21:43.460 ⇒ 00:21:51.689 Uttam Kumaran: But given you have that, it’s actually pretty easy to dynamically pull up sections of a document that could be helpful in the moment.
173 00:21:51.700 ⇒ 00:22:08.999 Uttam Kumaran: And so that is something that we’re actively pursuing for them now. But on top of that, what is the benefits of something, you know, that sits a little bit standalone? One, we can actually do a couple more powerful, things, like actually showing the reference back to where in the document
174 00:22:09.020 ⇒ 00:22:24.480 Uttam Kumaran: you know, it’s searched from. So, typically, when you do things like RAD, you have retrieval, but the AI, what it does is it’s taking a piece of a document and summarizing it for me. I actually want to not only summarize, but actually share with the CSR in the moment where in the document it came from.
175 00:22:24.480 ⇒ 00:22:30.410 Uttam Kumaran: So that maybe while they’re reading it, they can actually go reference and look just next to that and see other policies.
176 00:22:30.480 ⇒ 00:22:46.229 Uttam Kumaran: The other piece is actually training. So, we want to build some helpful features around, understanding improvements in a transcript, proactively making changes to, the database, the knowledge store, or proposing changes
177 00:22:46.230 ⇒ 00:22:55.729 Uttam Kumaran: without having a trainer in the loop, or necessarily proposing those themselves, and really basically helping provide every single CSR much more personalized
178 00:22:55.730 ⇒ 00:23:09.110 Uttam Kumaran: feedback on all of their calls. So you could see a benefit of having all the history of chat logs and all the transcripts in one place, being able to actually much more richly propose what is missing from the knowledge base.
179 00:23:09.110 ⇒ 00:23:25.489 Uttam Kumaran: And why are, sort of, calls now, breaching, sort of, hold time thresholds? And what else should be, like, promoted in training? And so, there are these several features that we’re now working on that sit squarely on, can we get all the transcripts in one place?
180 00:23:25.490 ⇒ 00:23:30.659 Uttam Kumaran: And can we build, sort of, like, something that’s more real-time, or almost, like.
181 00:23:30.750 ⇒ 00:23:36.470 Uttam Kumaran: sort of like a, in-front-of-you HUD that, like, brings up relevant information as, as you’re…
182 00:23:36.470 ⇒ 00:23:37.280 Scott_Harmon: How’s your hand?
183 00:23:37.280 ⇒ 00:23:38.380 Uttam Kumaran: Interacting with the customer.
184 00:23:38.380 ⇒ 00:23:44.880 Scott_Harmon: I just want to just land the plane real quickly, Ray, to your original question. What the thing does right now.
185 00:23:45.200 ⇒ 00:23:48.730 Scott_Harmon: and I don’t know if this is what you’re talking about, but it’s Maine…
186 00:23:49.040 ⇒ 00:23:53.210 Scott_Harmon: UI is responsive, so somebody’s saying, hey.
187 00:23:53.360 ⇒ 00:23:58.789 Scott_Harmon: You know, the client wants to know this, that, or the other thing. So his first job is to try and answer.
188 00:23:58.930 ⇒ 00:24:01.599 Scott_Harmon: And what they’ve added now is…
189 00:24:01.930 ⇒ 00:24:19.280 Scott_Harmon: kind of an inference to say… they call them, oh, by the ways, they’re using it as cross-sell, but they’re saying, based on who this customer is, what they bought, when they bought it, what they have, you should suggest this proactively to this customer. And so, that’s kind of your example, Ray, like.
190 00:24:19.470 ⇒ 00:24:24.419 Scott_Harmon: This customer and this situation suggests they could buy an additional service
191 00:24:24.610 ⇒ 00:24:28.869 Scott_Harmon: And so they’ll actually, at the end of the conversation.
192 00:24:29.020 ⇒ 00:24:31.420 Scott_Harmon: They’ll literally pop up and say.
193 00:24:32.150 ⇒ 00:24:38.009 Scott_Harmon: you know, tell the customer about this, because it’s very likely to be something they need. So, they’ve added a…
194 00:24:38.210 ⇒ 00:24:42.150 Scott_Harmon: At the end of the conversation.
195 00:24:42.330 ⇒ 00:24:52.659 Scott_Harmon: you know, helpful additional things that the AI has generated as being applicable to that client. They’re measuring upsell right now. I mean, the measurement for that is how many people bought new
196 00:24:53.070 ⇒ 00:25:03.029 Scott_Harmon: promotions or new services. We’re just starting to baseline that, but that’s one of the things, Uten, that Matt’s real excited about, is he’s like, hey, I can see revenue, like.
197 00:25:03.370 ⇒ 00:25:09.150 Scott_Harmon: That’s great, because I’ll be able to measure up incremental revenue, you know, when that happens, so that’s…
198 00:25:09.310 ⇒ 00:25:12.480 Scott_Harmon: That’s what they’ve built in the current product.
199 00:25:14.260 ⇒ 00:25:15.060 Raywo: Okay.
200 00:25:18.750 ⇒ 00:25:29.929 Uttam Kumaran: Yeah, and then, you know, kind of some of the work that, you know, I mentioned we’re doing some work for Ellie. For them, we’re doing a lot of, sort of, product analytics and funnel measurement work. So, for them, we’re measuring… they run a lot of paid
201 00:25:29.930 ⇒ 00:25:39.870 Uttam Kumaran: paid media and targeting, so we’re building a lot of, like, lifecycle measurement, for people that enter the funnel. Funnel measurement. We’re using Amplitude to build product analytics.
202 00:25:39.870 ⇒ 00:25:46.849 Uttam Kumaran: onto their… their, you know, their platform that… that ingests, that people go in through and measure, so that’s all…
203 00:25:46.850 ⇒ 00:25:59.769 Uttam Kumaran: data. And so, for us, we’re building out their entire, like, product analytics measurement, and building out their first understanding of, like, the life cycle of one of their, leads to customer, and then what happens after.
204 00:25:59.820 ⇒ 00:26:03.979 Uttam Kumaran: And so that’s what we’ve come in and started doing,
205 00:26:04.050 ⇒ 00:26:09.369 Uttam Kumaran: You know, for them, and it’s been… it’s been really effective. We’ve just been working with them for about a month so far, so…
206 00:26:11.250 ⇒ 00:26:11.820 Raywo: Okay.
207 00:26:12.700 ⇒ 00:26:14.340 Raywo: Yeah, the thing, you know, what…
208 00:26:15.370 ⇒ 00:26:22.260 Raywo: You know, an area that we’re… We’re still, attacking…
209 00:26:23.670 ⇒ 00:26:28.420 Raywo: Is really, you know, where we’re taking… we take 99% of our calls alive.
210 00:26:29.530 ⇒ 00:26:30.760 Raywo: Okay, and…
211 00:26:31.160 ⇒ 00:26:37.719 Raywo: Most of these people already have the ability to go into an AI chat and get every bit of information they want.
212 00:26:37.890 ⇒ 00:26:45.220 Raywo: So, the passive piece to go look up, all that information is available, we’ve already done…
213 00:26:45.550 ⇒ 00:26:50.160 Raywo: Okay, the next thing we want to do, though, is, is really…
214 00:26:50.600 ⇒ 00:27:05.009 Raywo: there’s, information that should be… we want to make sure, from a consistency, or a reduced variability, that certain pieces… certain things are covered. So, for example, we want to make sure that
215 00:27:05.260 ⇒ 00:27:11.000 Raywo: you know, the appointment is set. We want to make sure the charge card has been collected. We want to make sure that
216 00:27:11.270 ⇒ 00:27:20.549 Raywo: They’ve confirmed that they have the right hardware to go be able to do this call. We want to make sure that they understand what their copay is.
217 00:27:21.050 ⇒ 00:27:24.999 Raywo: We want to make sure that they understand that there’s testing requirements, so…
218 00:27:26.140 ⇒ 00:27:31.609 Raywo: You know, these are the things that we would be looking for, kind of ambient listening in a call center environment.
219 00:27:32.330 ⇒ 00:27:43.200 Raywo: To make sure that certain things are done, and if they’re not, you know, there would be some sort of heads-up display that the… we don’t have an agent, their care coordinators would know.
220 00:27:44.220 ⇒ 00:27:45.120 Raywo: So.
221 00:27:45.120 ⇒ 00:27:58.070 Uttam Kumaran: And then, tell me how’s it been working with Bland so far? I mean, we’ve done a lot of work with 8x8 now, with VAPI. We did some testing with Bland last year, but very familiar with that team. How has it been testing? Yeah.
222 00:27:58.070 ⇒ 00:28:07.400 Raywo: Yeah, we’re not… so Bland is kind of the next generation of what we’re doing, so we probably have 7 or 8 point solutions that we’re currently… Bland is kind of the next generation for us.
223 00:28:07.620 ⇒ 00:28:10.609 Raywo: But the point solutions are working really well.
224 00:28:10.790 ⇒ 00:28:15.640 Raywo: I mean, we’ve implemented… like I said, we’ve implemented at least 7 projects.
225 00:28:15.830 ⇒ 00:28:22.289 Raywo: And we continue to go. We just want greater control over it. These point solutions that third parties are building for us.
226 00:28:23.070 ⇒ 00:28:31.600 Raywo: You know, our ability to… customize it, is somewhat limited in…
227 00:28:31.700 ⇒ 00:28:35.450 Raywo: You know, plus we want one system informing another system.
228 00:28:35.770 ⇒ 00:28:43.830 Raywo: So that’s where we’re going. But the call center one is of interest to us, because we’re looking at our care coordination and
229 00:28:44.180 ⇒ 00:28:51.090 Raywo: what’s happening. We have, you know, 12 people that interact with patients, and we’re looking for there to be a consistent experience.
230 00:28:51.100 ⇒ 00:28:52.979 Uttam Kumaran: Make sure certain points are…
231 00:28:52.980 ⇒ 00:28:54.200 Raywo: are handled.
232 00:28:55.380 ⇒ 00:29:13.590 Uttam Kumaran: Yeah, maybe what I can do is I can share with you, sort of, like, what the UI of… you know, we’re developing this right now on top of the 8x8 data. I can sort of give you a sense of, like, what that UI would look like from our end. I mean, I’m curious to know, sort of, like, where the calls are coming on, and if you can get the live transcription or the live audio.
233 00:29:13.590 ⇒ 00:29:22.240 Uttam Kumaran: If you can grab those, this is definitely possible. Of course, there may be a little latency, but I think it’s certainly achievable to…
234 00:29:22.300 ⇒ 00:29:42.200 Uttam Kumaran: dynamically scan on some sort of schedule and look back through the call and proactively present information, whether it starts as just make sure that these things are hit, or it goes to something more dynamic about upselling, or churn prevention, or, you know, intervening in a call,
235 00:29:42.320 ⇒ 00:29:53.640 Uttam Kumaran: So I’m curious, like, do you know what the software that is actually, like, bringing in the calls? And that’s something I can look into to see whether even, you know, APIs or the data stream is available for
236 00:29:54.150 ⇒ 00:29:57.500 Uttam Kumaran: for that live, because I assume they’re taking it on a headset.
237 00:29:57.800 ⇒ 00:30:00.120 Raywo: Yeah, it’s coming in through RingCentral.
238 00:30:00.470 ⇒ 00:30:01.360 Uttam Kumaran: Okay, okay, fair.
239 00:30:01.360 ⇒ 00:30:02.050 Raywo: week.
240 00:30:02.830 ⇒ 00:30:06.519 Raywo: Or is it Microsoft Teams, Brian? Is it RingCentral? It’s RingCentral?
241 00:30:08.470 ⇒ 00:30:10.909 Brian: Which solution are you talking about?
242 00:30:10.910 ⇒ 00:30:14.679 Raywo: We’re talking about the calls that come in to Sarah’s team.
243 00:30:16.520 ⇒ 00:30:21.570 Brian: The calls that come into Sarah’s team today are, in the current state.
244 00:30:21.800 ⇒ 00:30:24.820 Brian: Being managed by a very simple
245 00:30:25.110 ⇒ 00:30:32.170 Brian: AI receptionist that basically just does call routing based upon.
246 00:30:32.170 ⇒ 00:30:33.710 Raywo: phrases that…
247 00:30:33.710 ⇒ 00:30:36.820 Brian: The patient says, and make sure that it gets sent to the right queue.
248 00:30:37.110 ⇒ 00:30:40.259 Raywo: Yeah, but the system itself, Utam, is RingCentral.
249 00:30:40.260 ⇒ 00:30:47.049 Brian: It’s RingCentral. It’s very unsophisticated, but we don’t need a lot of sophistication for it right now. But yes, that’s what we’re using.
250 00:30:48.170 ⇒ 00:30:49.980 Brian: It’s called their AI Receptionist.
251 00:30:50.830 ⇒ 00:30:51.430 Raywo: Okay.
252 00:30:51.610 ⇒ 00:30:59.530 Uttam Kumaran: And then, I guess, Ray, do you have a sense of, like, what percent of calls usually have this type of mistake? And then if you could just confirm, like, what is the…
253 00:30:59.530 ⇒ 00:31:07.759 Raywo: I don’t know, because I’m not listening or recording calls or anything today, so I don’t know what the variability is.
254 00:31:08.200 ⇒ 00:31:12.449 Uttam Kumaran: And what do you think… what is the remedy, like, if you miss one of these? Do you have to call folks back, and then there’s.
255 00:31:12.450 ⇒ 00:31:24.530 Raywo: The only thing is, is it just, it could lead to variability in the experience. I mean, we’re at 97% patient satisfaction, but doing 8,000 patients a month…
256 00:31:24.710 ⇒ 00:31:26.000 Raywo: 3%.
257 00:31:26.610 ⇒ 00:31:27.080 Uttam Kumaran: Yeah.
258 00:31:27.080 ⇒ 00:31:32.970 Raywo: a lot of people. So, we’re trying to… that’s what we’re trying to… you know.
259 00:31:34.080 ⇒ 00:31:35.310 Raywo: Trying to handle.
260 00:31:36.870 ⇒ 00:31:53.609 Uttam Kumaran: Okay, so one thing that we can do on our end is I’ll… we’re developing this right now. As soon as we sort of have something, I’m happy to sort of send over a little bit of a demo. We’re building sort of this, like, kind of, like, heads-up display for people that are actively on calls, and so that’s something that I’m happy to share.
261 00:31:53.640 ⇒ 00:32:06.569 Uttam Kumaran: and I’m definitely… we’ll explore RingCentral a bit. But yeah, if sort of augmenting continued to improve that customer satisfaction and continued to attack that 3%, and this is, like, you know, a keyhole where maybe at minimum.
262 00:32:06.590 ⇒ 00:32:17.760 Uttam Kumaran: you just want to have visibility into, like, what people are saying on those calls, and then at maximum, hey, can we turn… reduce some of that churn, and can we expand? I think that’d be a great opportunity. Yeah.
263 00:32:20.530 ⇒ 00:32:29.360 Raywo: All right, listen, this is a good discussion I have to run. I gotta get on another call, but listen, I definitely appreciate…
264 00:32:30.870 ⇒ 00:32:34.210 Raywo: Yeah, and listen, we’d love to see if you got something there.
265 00:32:34.680 ⇒ 00:32:39.119 Raywo: You know, we’re… Like I said, we can only advance so much on our own.
266 00:32:40.110 ⇒ 00:32:44.919 Uttam Kumaran: Yeah, no, I’ll shoot it over to you, you can let me know what you think as soon as it’s ready, so, perfect. Sounds good.
267 00:32:44.920 ⇒ 00:32:46.470 Raywo: Alright guys, thank you.
268 00:32:47.120 ⇒ 00:32:48.699 Uttam Kumaran: Thank you so much, I appreciate it.
269 00:32:49.370 ⇒ 00:32:50.270 Raywo: Take care.