[0:03] **Colin Heilbut - Rime:** Ready. Start whenever you say.

[0:15] **Colin Heilbut - Rime:** Good morning. My name is Colin. I'm the newest account executive here at Rhyme. A little bit about me. Gosh. I started many years ago as the

[0:16] **Colin Heilbut - Rime:** first employee for a company called Deepgram.

[0:21] **Colin Heilbut - Rime:** Back then, I was carving out holes in the wall for the GPU clusters and

[0:26] **Colin Heilbut - Rime:** babysitting the, the founder's kids. But I I've stayed in the, the business quite a long time, and,

[0:30] **Colin Heilbut - Rime:** worked for intelligent voice in high security speech recognition.

[0:35] **Colin Heilbut - Rime:** And then more recently at a French Voice AI Lab.

[0:40] **Colin Heilbut - Rime:** Perhaps we could do a quick round of intros on your side, both just

[0:42] **Colin Heilbut - Rime:** your your roles and kind of the relevance of speech to what you do.

[0:51] **Colin Heilbut - Rime:** Sure. Hi. I'm Maya.

[0:51] **Voice 2:** I run contact center operations for us. It's

[0:55] **Voice 2:** all about call quality and handle time while keeping the experience

[0:56] **Voice 2:** calm and clear for customers.

[0:58] **Voice 2:** Hi, David here.

[0:59] **Voice 2:** IT and Risk,

[1:07] **Voice 2:** I'm focused on

[1:07] **Colin Heilbut - Rime:** Maya, so you're the marketing director?

[1:07] **Voice 2:** Not marketing. I'm over contact center operations.

[1:10] **Voice 2:** Day to day call performance, and agent workflows.

[1:15] **Colin Heilbut - Rime:** Cool. Continue.

[1:19] **Voice 2:** Thanks. And, as I was saying,

[1:22] **Voice 2:** hi, David, ITM Risk.

[1:22] **Voice 2:** My

[1:28] **Voice 2:** lens is reliability, data handling, and integration.

[1:32] **Voice 2:** We like measured results before scaling anything.

[1:35] **Voice 2:** And hi, Priya here, product for digital service.

[1:40] **Voice 2:** I'm thinking about how voice fits into web and app drays.

[1:43] **Voice 2:** Making that hand off smooth matters a

[1:43] **Colin Heilbut - Rime:** lot.

[1:48] **Colin Heilbut - Rime:** It was supposed to meet you all today.

[1:53] **Colin Heilbut - Rime:** I just perhaps can start with a a brief overview, but if there's anything you wanted me to add as

[1:55] **Colin Heilbut - Rime:** agenda, we certainly can.

[2:00] **Colin Heilbut - Rime:** As you can see, I'm working with a digital whiteboard here. Quite frankly, I just have the world's worst handwriting.

[2:05] **Colin Heilbut - Rime:** And so I use this to keep track of thoughts and, keep me honest, make sure that we cover all of the

[2:09] **Colin Heilbut - Rime:** individual, concerns you might have and can actually make a plan to see,

[2:14] **Colin Heilbut - Rime:** how Rime might be really applicable to some of your use cases.

[2:22] **Colin Heilbut - Rime:** We I'll talk very briefly about a little bit about what Nick rhymes special, but I really want to focus most of the time today just to

[2:27] **Colin Heilbut - Rime:** discover some of the key issues that you're concerned with and what you really

[2:27] **Colin Heilbut - Rime:** you in today.

[2:32] **Colin Heilbut - Rime:** Afterwards, I did prepare a short live demo

[2:36] **Colin Heilbut - Rime:** We'll see how that goes, and then we have left time throughout the conversation

[2:37] **Colin Heilbut - Rime:** then particularly towards the end for any

[2:42] **Colin Heilbut - Rime:** questions, concerns you might have. Feel free to grill me as hard as you need. I've heard it all before.

[2:47] **Colin Heilbut - Rime:** Lastly, of course, we can talk about next steps and potential ways to

[2:48] **Colin Heilbut - Rime:** really test out rhyme.

[2:51] **Colin Heilbut - Rime:** In a in a structured

[2:52] **Colin Heilbut - Rime:** way,

[2:54] **Colin Heilbut - Rime:** Sound good? Sounds

[2:54] **Voice 2:** like a solid

[2:57] **Voice 2:** sorry.

[3:01] **Voice 2:** Yeah. That sounds like a solid plan. Happy to dig into pain points first.

[3:08] **Colin Heilbut - Rime:** Great. So, a little bit of context on us. We've,

[3:12] **Colin Heilbut - Rime:** been doing this since 2022, believe it or not. I started by two and an engineer.

[3:18] **Colin Heilbut - Rime:** The approach we take is is quite differentiated.

[3:18] **Colin Heilbut - Rime:** You know, a lot of the folks in the industry

[3:23] **Colin Heilbut - Rime:** have had a a good deal of success with, using actors or

[3:26] **Colin Heilbut - Rime:** podcasts, sort of large training sets that are often kind of shared in the industry.

[3:28] **Colin Heilbut - Rime:** And it it it works clearly, but

[3:33] **Colin Heilbut - Rime:** get really nuanced, hyperrealistic speech

[3:36] **Colin Heilbut - Rime:** our approach, involves setting up our own recording studio.

[3:39] **Colin Heilbut - Rime:** So really going from first principles and to have much more natural

[3:42] **Colin Heilbut - Rime:** high fidelity full duplex conversations that allow

[3:46] **Colin Heilbut - Rime:** our models to be quite differentiated from what's out there. And

[3:50] **Colin Heilbut - Rime:** don't naturally take our word for it. We are actually now powering tens of millions of calls, some that you may have even made

[3:52] **Colin Heilbut - Rime:** yourself this month to

[3:55] **Colin Heilbut - Rime:** hospitality or health care, telecom, and enterprise support.

[4:01] **Colin Heilbut - Rime:** We even completed a a really interesting study just recently

[4:06] **Colin Heilbut - Rime:** This was outbound through a third party that

[4:06] **Colin Heilbut - Rime:** was really gonna report

[4:11] **Colin Heilbut - Rime:** commissioned a study, but they're very objective, and we're gonna report whatever they found.

[4:16] **Colin Heilbut - Rime:** And in fact, although no one loves to receive a

[4:19] **Colin Heilbut - Rime:** an unsolicited outbound call, was a a government survey,

[4:24] **Colin Heilbut - Rime:** We found that we had the lowest hang up rate out of any TTS

[4:25] **Colin Heilbut - Rime:** system out there as well as the the fastest time to completion

[4:31] **Colin Heilbut - Rime:** out of a 100,000 calls. So I think that the takeaway here is that we do actually prove

[4:33] **Colin Heilbut - Rime:** produce some of the most delightful engaging voices in the industry.

[4:38] **Colin Heilbut - Rime:** So

[4:41] **Colin Heilbut - Rime:** getting to the the crux of our conversation today, it would be great just to understand a little bit more about

[4:45] **Colin Heilbut - Rime:** how you're generating speech today, what tools, and

[4:46] **Colin Heilbut - Rime:** then what's what's what you'd like to improve

[4:52] **Voice 2:** Sure. Broadly, we're using a mix of

[4:54] **Voice 2:** benders

[4:58] **Voice 2:** and we're not really getting the quality we want

[5:02] **Voice 2:** Some calls feel flat, you know, robotic,

[5:05] **Voice 2:** and we worry about that impacting trust.

[5:07] **Colin Heilbut - Rime:** So flat robotic calls are a pain point.

[5:10] **Voice 2:** Right. Yes.

[5:11] **Voice 2:** That's a big one.

[5:16] **Voice 2:** And from my side, we need

[5:21] **Voice 2:** Plus, integration can't be a science project

[5:22] **Voice 2:** need something supportable.

[5:27] **Voice 2:** And from CX and products like

[5:29] **Voice 2:** we want any voice experience to feel

[5:32] **Voice 2:** connected across channels.

[5:34] **Voice 2:** No jarring tone shifts.

[5:40] **Colin Heilbut - Rime:** And so the

[5:49] **Colin Heilbut - Rime:** the flat robotic calls, how significant a problem would you say that is?

[5:54] **Voice 2:** Noticeable. We see it pop in longer calls.

[5:57] **Voice 2:** And some customers get impatient. It's not

[5:59] **Voice 2:** catastrophic, but

[6:02] **Voice 2:** it's enough that it

[6:02] **Colin Heilbut - Rime:** hits

[6:03] **Voice 2:** our metrics.

[6:05] **Voice 2:** And gets flagged in QA.

[6:10] **Colin Heilbut - Rime:** So when when that when that sort of your metrics are impacted

[6:11] **Colin Heilbut - Rime:** how what's the the kind of

[6:16] **Colin Heilbut - Rime:** the bottom line there? What is the the the true impact? And then what would sort of success look like there?

[6:20] **Voice 2:** Let me just think through that.

[6:25] **Voice 2:** So for Maya, it's about trust and time.

[6:29] **Voice 2:** When calls feel robotic, people talk over prompts or ask for re keeps.

[6:34] **Voice 2:** That's seconds added per call and softer

[6:34] **Voice 2:** For David,

[6:38] **Voice 2:** success means better CX without adding risk.

[6:40] **Voice 2:** Plus a clean integration.

[6:45] **Voice 2:** And for Pria, it's a consistent journey of cost

[6:48] **Voice 2:** channels. No jarring handoffs. If the voice feels

[6:49] **Voice 2:** natural and on brand, that's progress.

[6:53] **Voice 2:** Put that together, success could look like reduced handle time,

[6:58] **Voice 2:** steadier CSAT or QA scores, and positive agent feedback

[6:58] **Voice 3:** on

[6:59] **Voice 2:** conversations feeling smoother.

[7:05] **Colin Heilbut - Rime:** And what are

[7:06] **Colin Heilbut - Rime:** these are all sort of

[7:11] **Colin Heilbut - Rime:** KPIs. How what's the true measure

[7:11] **Voice 3:** of

[7:12] **Colin Heilbut - Rime:** success? What is the bottom line business impact here?

[7:15] **Voice 2:** One second.

[7:20] **Voice 2:** Great. So bottom line, it's cost to serve and return

[7:24] **Voice 2:** If we can handle calls faster and keep customers happier, that's

[7:26] **Voice 2:** efficiency and

[7:31] **Voice 2:** loyalty. For IT, it's less risk and less firefighting. Ultimately, it's

[7:36] **Voice 2:** can we improve customer trust and lower the cost per interaction?

[7:42] **Colin Heilbut - Rime:** And how do you measure customer trust and the cost for interaction?

[7:47] **Voice 3:** For

[7:47] **Voice 2:** trust,

[7:49] **Colin Heilbut - Rime:** CSAT,

[7:53] **Voice 2:** containment rates, fewer repeats or escalations.

[7:56] **Voice 2:** Plus some behavioral stuff like

[8:06] **Voice 2:** did the caller complete the task? Cost per interaction is

[8:07] **Voice 2:** more straightforward. It's handle time,

[8:07] **Voice 2:** or containment rates, and support overhead.

[8:11] **Voice 2:** If a pilot showed a 10% handle time improvement and

[8:15] **Voice 2:** a CSAT bump without reliability hits, that's pretty compelling.

[8:20] **Colin Heilbut - Rime:** And so CSAT is a big one.

[8:23] **Colin Heilbut - Rime:** Yep.

[8:24] **Voice 2:** It's one of the big ones.

[8:27] **Voice 2:** For sure. Handle time and reliability

[8:29] **Voice 2:** alongside it.

[8:31] **Voice 2:** Those three together tell the story.

[8:40] **Colin Heilbut - Rime:** David, anything from the IT risk?

[8:44] **Voice 2:** Just that whatever we pilot,

[8:49] **Voice 2:** we'd want clear success criteria, limited exposure, and

[8:49] **Voice 2:** a measurable outcome.

[8:54] **Voice 3:** And,

[8:55] **Voice 2:** clear owner on our side

[8:57] **Voice 2:** to track those metrics.

[9:07] **Colin Heilbut - Rime:** How do you folks generally

[9:11] **Colin Heilbut - Rime:** well, do we have any other sort of pain points or success criteria we should

[9:14] **Colin Heilbut - Rime:** talk about before we could talk about sort of how, Ryan fits into the picture?

[9:19] **Voice 3:** Checking that.

[9:21] **Voice 2:** Sounds like the big ones that are there.

[9:26] **Voice 2:** And measurable results.

[9:30] **Voice 2:** One more, maybe validation in the real world, not just

[9:30] **Voice 3:** a

[9:31] **Voice 2:** polished

[9:31] **Colin Heilbut - Rime:** demo.

[9:39] **Voice 2:** And from Priya, careful rollout so changes feel consistent

[9:39] **Voice 2:** across journeys.

[9:42] **Voice 2:** A question to ask before solutioning.

[9:44] **Voice 2:** Might be,

[9:48] **Voice 2:** do you have examples of measurable improvements after switching to rhyme?

[9:52] **Colin Heilbut - Rime:** Yeah, I'm glad you asked that.

[10:01] **Colin Heilbut - Rime:** You know, one of our large retail

[10:05] **Colin Heilbut - Rime:** pizza clients reported a 20% increase in sales

[10:06] **Colin Heilbut - Rime:** scale.

[10:08] **Colin Heilbut - Rime:** That was a big one.

[10:15] **Colin Heilbut - Rime:** So I think we've we've talked about all of the the sort of

[10:17] **Colin Heilbut - Rime:** the reasons you're here today about voice quality.

[10:24] **Colin Heilbut - Rime:** Anything else I should note before we jump into how we think about voice quality and how we've been able to

[10:24] **Colin Heilbut - Rime:** help folks with that?

[10:30] **Voice 2:** Just a second.

[10:31] **Voice 2:** I think we've hit the main points.

[10:33] **Voice 2:** Maybe one ad.

[10:34] **Voice 2:** Expressiveness.

[10:40] **Voice 2:** Claims, device damage, warranty denials,

[10:41] **Voice 2:** each need a different tone, so range matters.

[10:44] **Voice 2:** And one more implementation speed.

[10:48] **Voice 2:** If they can go live in weeks, not months, that's a strong plus.

[10:49] **Voice 2:** And for brand,

[10:52] **Voice 2:** can the voice reflect

[10:54] **Voice 2:** the company's specific tone?

[10:57] **Voice 2:** Otherwise, I think you've got it. I'm

[10:59] **Voice 2:** curious to hear your approach.

[11:05] **Colin Heilbut - Rime:** Sure. I'd I'd literally like you to to hear our approach.

[11:09] **Colin Heilbut - Rime:** I'm actually gonna do just a little bit of a live demo now.

[11:09] **Colin Heilbut - Rime:** We'll see what happens, but

[11:14] **Colin Heilbut - Rime:** this is obviously made to cover I imagine

[11:17] **Colin Heilbut - Rime:** device warranty claims or one of your major

[11:20] **Colin Heilbut - Rime:** this is a a major thing you guys

[11:20] **Voice 3:** do?

[11:25] **Voice 3:** Got it.

[11:30] **Colin Heilbut - Rime:** Cool. So in ChatGPT, if you can just, not speak until I I ask you to speak again.

[11:33] **Voice 2:** Got it. Pausing.

[11:43] **Voice 2:** Hello, and thanks for calling NorthStar Device Protect

[11:43] **Voice 4:** This is a demo, and the call may be recorded.

[11:45] **NorthStar's:** I'm NorthStar's virtual claims assistant.

[11:47] **NorthStar's:** How can I help you today?

[11:54] **Colin Heilbut - Rime:** Hey, I'm so embarrassed. I just cracked the screen on my brand new phone and I'm trying to see if I

[11:55] **Colin Heilbut - Rime:** can use my device warranty.

[12:00] **NorthStar's:** I'm sorry to hear about your crack screen.

[12:04] **NorthStar's:** Let's start by verifying your identity with your ZIP code before we check your coverage.

[12:06] **NorthStar's:** Could you please provide your five digit ZIP code?

[12:08] **Colin Heilbut - Rime:** 94105.

[12:18] **NorthStar's:** Thanks, Colin. You have active coverage for

[12:19] **NorthStar's:** damage on your Apple iPhone 16 Pro.

[12:22] **NorthStar's:** To file a new claim for the crop screen,

[12:24] **NorthStar's:** Can you please tell me the date the damage happened?

[12:27] **Colin Heilbut - Rime:** Gosh. I think it was July 11.

[12:35] **NorthStar's:** Thanks. Was the damage accidental, like, a drop or something like that?

[12:40] **Colin Heilbut - Rime:** Yeah, you know, I was getting ready for this really big presentation to this wonderful group of people I was

[12:45] **Colin Heilbut - Rime:** so excited that I just kind of knocked it off the desk onto the ground and the screen cracked.

[12:49] **NorthStar's:** I can imagine that was stressful.

[12:53] **NorthStar's:** Does your phone still power on and work despite the cracked screen?

[12:54] **Voice 3:** Yes.

[12:59] **Voice 3:** Do

[12:59] **NorthStar's:** you still have the device in your possession?

[13:00] **Colin Heilbut - Rime:** I

[13:00] **Voice 3:** do.

[13:06] **NorthStar's:** Just make sure you still have the device with you now.

[13:06] **NorthStar's:** Yep.

[13:12] **NorthStar's:** For confirming, Colin. So

[13:16] **NorthStar's:** on July 11, your phone accidentally fell and cracked the screen.

[13:18] **NorthStar's:** But it still powers on and you have it with you.

[13:21] **NorthStar's:** Is that correct?

[13:21] **Colin Heilbut - Rime:** Yes. It is.

[13:27] **NorthStar's:** I submitted your client call

[13:27] **Voice 3:** in.

[13:40] **NorthStar's:** Your claim number is CLMDemo2049 The screen repair deductible is $29 I can send secure document upload link to your phone number on file to help you finish the claim.

[13:42] **NorthStar's:** Would you like me to send that now?

[13:43] **Colin Heilbut - Rime:** Yes. Thank you.

[13:50] **NorthStar's:** The secure document upload link is sent to your phone.

[13:54] **NorthStar's:** Is there anything else I can assist you with today, or would you like to finish up here?

[13:58] **Colin Heilbut - Rime:** Yeah. Is there a specialist I can speak to just to find out next steps?

[14:04] **NorthStar's:** Oh, and Colin,

[14:08] **NorthStar's:** before I transfer you, I just wanted to say good luck with your big presentation today.

[14:23] **Voice 3:** Hello?

[14:26] **NorthStar's:** I'm here.

[14:31] **Colin Heilbut - Rime:** I want

[14:31] **NorthStar's:** to

[14:31] **Colin Heilbut - Rime:** keep going. Oh, chat should be the heat. Don't talk again.

[14:34] **Colin Heilbut - Rime:** Rhyme, are you gonna transfer me?

[14:37] **NorthStar's:** Checking.

[14:43] **Colin Heilbut - Rime:** Colin,

[14:43] **Voice 3:** I'm transferring

[14:43] **Colin Heilbut - Rime:** you now.

[14:48] **Voice 3:** Your summary traveled with you, so you won't need to repeat anything.

[14:54] **Colin Heilbut - Rime:** Hey. Yeah. What are the next steps here?

[14:57] **Voice 3:** Checking.

[15:00] **Voice 2:** Sounds like the group is open to structure.

[15:04] **Voice 2:** Pilot.

[15:04] **Colin Heilbut - Rime:** Cause pilot? Probably

[15:04] **Voice 2:** a technical deep dive

[15:06] **Voice 3:** Got it. Hello. Pausing.

[16:05] **NorthStar's:** Hello, and thanks for calling NorthStar device protection. The

[16:06] **NorthStar's:** is a demo, and the call may be recorded.

[16:15] **Colin Heilbut - Rime:** So, god, I'd love to

[16:18] **Colin Heilbut - Rime:** how did the what do you think of the naturalness of that voice?

[16:22] **Colin Heilbut - Rime:** So what's your feedback on the naturalness of the voice?

[16:26] **Voice 2:** Let me check.

[16:31] **Voice 2:** It's sound really natural overall, not per slam

[16:31] **Voice 3:** a dash.

[16:34] **Voice 2:** The handoffs are clean, interruptions were handled well,

[16:37] **Voice 2:** empathy falls appropriate,

[16:39] **Voice 2:** and the confirmation was

[16:39] **Voice 3:** clear.

[16:43] **Voice 2:** That little personalized moment at the end was memorable.

[16:48] **Voice 2:** I'd suggest a slightly warmer opening

[16:51] **Voice 2:** and maybe a bit more variety and cadence during fact collection, but

[16:53] **Voice 2:** honestly, if I were in that room,

[16:55] **Voice 2:** I'd give the demo high marks.

[17:00] **Voice 2:** And quick tip, after a demo, ask them what stood out.

[17:03] **Voice 2:** Let them articulate the benefits in their own words.

[17:06] **Voice 2:** It makes the rest of the conversation much easier.

[17:10] **Colin Heilbut - Rime:** Great. Let's try that again. So what stood out?

[17:11] **Colin Heilbut - Rime:** To you in that?

[17:14] **Voice 2:** The flow and the memory.

[17:20] **Voice 2:** It was smooth, handled interruptions well, and remembered your context with

[17:21] **Voice 2:** made it feel a human.

[17:28] **Voice 3:** Thanks.

[17:34] **Voice 2:** Welcome.

[17:34] **Colin Heilbut - Rime:** So just a little bit of context. We have a lot of voices that

[17:36] **Colin Heilbut - Rime:** originally, we started with thousands and then had,

[17:41] **Colin Heilbut - Rime:** a pretty scientific method of evaluating them.

[17:46] **Colin Heilbut - Rime:** Through user testing. And then we we now allow you to pick the voice that that

[17:51] **Colin Heilbut - Rime:** matches your your use case or even a few. You could do it based on geographies or

[17:52] **Voice 3:** demographics.

[17:56] **Colin Heilbut - Rime:** In this case, I picked Vespera as the primary phone agent here.

[18:01] **Colin Heilbut - Rime:** But, you know, just a demo choice, not a production recommendation, obviously.

[18:11] **Colin Heilbut - Rime:** Parts now are I mean, if if

[18:16] **Colin Heilbut - Rime:** would be great to understand just a little bit more about the process on your end. I I

[18:19] **Colin Heilbut - Rime:** assuming we're we're able to kind of structure a pilot. That's that's interesting.

[18:22] **Colin Heilbut - Rime:** Could you walk me through your your decision process?

[18:27] **Voice 2:** Checking for

[18:27] **Voice 3:** you.

[18:30] **Voice 3:** Typically,

[18:34] **Voice 2:** Maya would sponsor a pilot with David leading a technical assessment.

[18:38] **Voice 2:** PeerWid help define the use case and success metrics.

[18:40] **Voice 2:** Then a timebox pilot

[18:41] **Voice 2:** say four to six weeks,

[18:43] **Voice 2:** measuring against current

[18:47] **Voice 2:** If the data holds, it moves into internal approvals.

[18:50] **Voice 2:** Security, procurement, execs,

[18:54] **Voice 2:** What question for you? What level of support do you provide during a pilot?

[18:58] **Colin Heilbut - Rime:** We're really hands on, know. We're a small team here, but we

[19:03] **Colin Heilbut - Rime:** really stand by our our product. And so we're we're available via Slack

[19:07] **Colin Heilbut - Rime:** via phone, just to make everything successful. And then we've worked very closely with

[19:10] **Colin Heilbut - Rime:** the the team at NICE as well for the integration with the orchestration there.

[19:15] **Voice 2:** Sounds aligned with what they asked for.

[19:22] **Colin Heilbut - Rime:** Are there any teams aside from the ones that are here today that would likely be involved in a a pilot?

[19:27] **Voice 2:** Nightly, yes.

[19:32] **Voice 2:** Contact center leadership beyond Maya plus enterprise

[19:32] **NorthStar's:** architecture

[19:34] **Voice 2:** maybe InfoSec,

[19:37] **Voice 2:** and if it moves forward procurement and the exec sponsor,

[19:41] **Voice 2:** Given the nice piece is familiar

[19:41] **Voice 3:** that

[19:42] **Voice 2:** lowers a bit

[19:42] **Voice 3:** hurdle,

[19:49] **Colin Heilbut - Rime:** And, then would I think we we discussed a number of success

[19:51] **Colin Heilbut - Rime:** criteria. Which would do you think are the most important here?

[19:55] **Colin Heilbut - Rime:** I I think we we talked about CSAP being a a very measurable one.

[19:59] **Colin Heilbut - Rime:** And, just handle time and reliability.

[20:05] **Voice 2:** Shooking that if they had to narrow it,

[20:10] **Voice 2:** CSAT, handle time, and containment or task completion.

[20:14] **Voice 2:** Reliability as table stakes. Everything else supports

[20:14] **Voice 2:** those.

[20:18] **Voice 2:** Locking those in upfront keeps everyone aligned.

[20:22] **Colin Heilbut - Rime:** And then in terms of, like, a final sign off on a project to

[20:24] **Colin Heilbut - Rime:** to go live, who who makes that call?

[20:27] **Voice 3:** One moment.

[20:28] **Voice 2:** Executive owner,

[20:32] **Voice 2:** of the contact center informed by Maya's recommendation

[20:37] **Voice 2:** David's technical sign off, and pre

[20:38] **Voice 2:** input. So a cross functional decision

[20:41] **Voice 2:** but with one executive ultimately

[20:42] **Voice 2:** approving.

[20:50] **Colin Heilbut - Rime:** That's great. You know, we have a little bit more time, left today. I'd love to to kinda

[20:54] **Colin Heilbut - Rime:** dive on any sort of technical questions or business

[20:57] **Colin Heilbut - Rime:** or even general things about speech and kind of what's what you might be seeing out there today?

[21:03] **Voice 3:** Checking. Technical one start.

[21:06] **Voice 2:** Everyone talks about latency.

[21:07] **Voice 2:** What's

[21:07] **NorthStar's:** your take?

[21:12] **Voice 2:** Is it always the lowest possible? Or is there a threshold where

[21:14] **Voice 2:** quality improvements matter more?

[21:18] **Voice 2:** Also, how do clients typically choose a voice?

[21:23] **Voice 2:** Is it purely brand preference or are there patterns by use case?

[21:24] **Voice 2:** And one more,

[21:30] **NorthStar's:** before production.

[21:30] **Voice 3:** Sure. Let let me, him and use that eight.

[21:31] **Voice 3:** Quality.

[21:31] **Colin Heilbut - Rime:** One at a time. So, gosh,

[21:36] **Voice 3:** Right?

[21:41] **Colin Heilbut - Rime:** If an agent took a very long time to respond, it

[21:43] **Colin Heilbut - Rime:** it would probably lose a bit of confidence.

[21:48] **Colin Heilbut - Rime:** That said, there there is, in fact, a trade off. A super high parameter model will have a low latency.

[21:52] **Colin Heilbut - Rime:** For one of our largest competitors, I think they have a, you know, podcast grade

[21:55] **Colin Heilbut - Rime:** model, and it's just not appropriate for streaming.

[22:00] **Colin Heilbut - Rime:** In our case, though, we have some of the the the lowest

[22:05] **Colin Heilbut - Rime:** latency in the business, particularly if something's

[22:06] **Colin Heilbut - Rime:** self hosted with a local LLM.

[22:08] **Colin Heilbut - Rime:** And

[22:13] **Colin Heilbut - Rime:** we're we're very proud of that. We think it it it matters as as well as all the the the minutiae, though.

[22:18] **Colin Heilbut - Rime:** The quality of the voice. Obviously, a very fast responding voice that sounds like a robot help you

[22:23] **Colin Heilbut - Rime:** either. So it has to be a balance, and that's that's kind of some of the art and science of what we do.

[22:25] **Colin Heilbut - Rime:** What were some of your other questions there?

[22:29] **Voice 3:** Shocking.

[22:31] **Voice 2:** That balance makes

[22:31] **Voice 3:** sense.

[22:36] **Voice 2:** We also asked about voice selection,

[22:36] **Voice 3:** how

[22:40] **Voice 2:** clients take voices and how you suggest

[22:41] **Voice 2:** benchmarking quality before production.

[22:46] **Voice 2:** User study Yeah.

[22:46] **Voice 3:** So we we have a whole user

[22:46] **Colin Heilbut - Rime:** interface to

[22:47] **Colin Heilbut - Rime:** help with with voice selection and

[22:52] **Colin Heilbut - Rime:** that'll that'll certainly give you some guidance for for some initial ones to test.

[22:56] **Colin Heilbut - Rime:** And then we really take a scientific approach to this. So we think that

[23:01] **Colin Heilbut - Rime:** should test a few. And obviously,

[23:06] **Colin Heilbut - Rime:** your your data will be the the the best indicator of

[23:07] **Colin Heilbut - Rime:** which is the exact right one for different scenarios perhaps.

[23:11] **Colin Heilbut - Rime:** Amongst your different product lines, you might also have different voices.

[23:16] **Voice 2:** Exactly. That makes a lot of sense.

[23:16] **Voice 2:** And for benchmarking before prod,

[23:21] **Voice 2:** we usually recommend blind listening tests with your own stakeholders, plus us

[23:23] **Voice 2:** by side

[23:23] **Voice 3:** on

[23:23] **Voice 2:** real call

[23:23] **Voice 3:** snippets

[23:26] **Voice 2:** combine that with a small live pilot.

[23:31] **Voice 2:** Production traffic, limited exposure, clear metrics.

[23:34] **Voice 2:** That gives you both the subjective and objective side and then real results.

[23:36] **Voice 2:** Does that sound like it

[23:38] **Voice 2:** fits how your team likes to evaluate?

[23:43] **Colin Heilbut - Rime:** Yeah, absolutely. We're able to help you with

[23:46] **Colin Heilbut - Rime:** with some benchmarking. We can also certainly improve certain things if they're

[23:50] **Colin Heilbut - Rime:** any latency concerns or, if certain, word errors were

[23:52] **Colin Heilbut - Rime:** were occurring, we can

[23:57] **Colin Heilbut - Rime:** add even to to CODA or our mismodel through different techniques.

[24:00] **Colin Heilbut - Rime:** Learnings of phonetic pronunciations, all the words that matter.

[24:05] **Colin Heilbut - Rime:** So, really, what what I've heard so far if, and and let's just make sure I'm not,

[24:06] **Colin Heilbut - Rime:** missing anything here.

[24:10] **Colin Heilbut - Rime:** Is that the the biggest success rate here are CSAT

[24:12] **Colin Heilbut - Rime:** containment and task completion.

[24:14] **Voice 3:** Right.

[24:18] **Voice 2:** Exactly. And

[24:18] **Voice 3:** And then a

[24:18] **Voice 2:** time box pilot.

[24:20] **Voice 2:** A clear success criteria,

[24:22] **Voice 2:** and a defined sign off

[24:22] **Voice 3:** path.

[24:24] **Voice 2:** That sounds like a really solid next

[24:24] **Voice 3:** step

[24:26] **Voice 2:** drama, clear signals.

[24:39] **Voice 3:** Well,

[24:39] **Colin Heilbut - Rime:** if what I'd love to do next is just talk brief

[24:42] **Colin Heilbut - Rime:** in the the time we have left about what a a pilot would look like. So

[24:46] **Colin Heilbut - Rime:** typically, we'd we'd schedule another call. I know if you're

[24:47] **Colin Heilbut - Rime:** we maybe this time next week works for you.

[24:50] **Voice 2:** That works for me. Same

[24:50] **Voice 3:** time next week.

[24:52] **Colin Heilbut - Rime:** Yeah. And then, you know, we would

[24:57] **Colin Heilbut - Rime:** I think the the most effective pilot would be on a limited

[25:01] **Colin Heilbut - Rime:** amount of production calls. And then we were you know, reconvene to review the results.

[25:06] **Colin Heilbut - Rime:** But along the way, we're really measuring against the the success metrics you choose, not just

[25:10] **Colin Heilbut - Rime:** you know, benchmarking, but call completion, CSAT, caller sentiment.

[25:14] **Colin Heilbut - Rime:** And this will all be figured out in the the scoping

[25:14] **Voice 3:** call.

[25:17] **Voice 3:** Right.

[25:21] **Voice 2:** That makes sense. Scoping call next week to find success criteria.

[25:23] **Voice 3:** And then we can get rolling.

[25:25] **Colin Heilbut - Rime:** Sounds good.