Terminal Value

LightSpun founders discuss AI-native automation replacing legacy dental insurance back-office systems

42 min Listen to the original Episode page
AI AutomationHealthcare AdministrationDental InsuranceSaaSLegacy ModernizationVenture Capital

Summary

Co-founders Shaju Puthussery and Deepak Ramaswamy explain how LightSpun uses an AI-native architecture to automate dental insurance administration, contrasting it with legacy systems reliant on fax and paper. They claim their platform reduces claim adjudication time to under a second and cuts administrative costs by 50%. The company entered the market via credentialing products and is expanding into vision, pharmacy, and behavioral health using a LAMP strategy. The speakers argue that regulated industries can modernize rapidly and that failure to adopt AI-native automation poses an existential risk to incumbents.

Key takeaways

  1. 1
    LightSpun has raised $18 million to build an AI-native administrative platform for dental insurance.

    Unidentified speaker (speaker_0) · 0:51

  2. 2
    An AI-native solution can cut administrative costs by 50%, a figure described as conservative by the speaker.

    Unidentified speaker (speaker_2) · 35:37

Episode sections

  1. 0:00 – 0:51

    AI-Native vs Legacy Infrastructure

    Speakers contrast LightSpun's AI-native architecture with competitors who bolt AI onto legacy platforms, arguing that legacy systems are difficult to retool for AI agents.

  2. 0:51 – 1:44

    LightSpun Introduction and Market Context

    The host introduces co-founders Shaju Puthussery and Deepak Ramaswamy. LightSpun has raised $18 million to build an AI-native administrative platform for dental insurance.

  3. 1:44 – 7:13

    Dental Insurance Legacy Pain Points

    The speakers describe the dental administrative space as operating 10 years behind other sectors, with key issues including reliance on fax machines and paper-based EOBs.

  4. 7:13 – 11:20

    Automating Claims and Payments

    LightSpun automates the entire back office, claiming to adjudicate routine claims in under a second and pay them shortly after, contrasting with legacy systems where providers wait 30-60 days.

  5. 11:20 – 17:51

    Deterministic Rules vs Clinical Judgment

    Founders explain their approach: deterministic rules are automated 100%, while clinical decisions remain with licensed professionals. AI agents handle repetitive tasks, with humans reviewing low-confidence outputs.

  6. 17:51 – 22:27

    Model Selection and Data Organization

    The speakers discuss the balance between using frontier models and fine-tuning, emphasizing that data organization and reducing token usage are critical challenges.

  7. 22:27 – 30:00

    Go-to-Market and Credentialing Entry Point

    LightSpun entered the market via a credentialing product, solving a major pain point for the ADA and DSOs, which helped overcome change management hurdles.

  8. 30:00 – 35:17

    AI-Native Migration Advantages

    Founders detail how AI-native architecture simplifies migration from legacy systems, with AI agents handling schema mapping and data inconsistencies.

  9. 35:17 – 36:58

    Expansion into Adjacent Benefits and LAMP Strategy

    Speaker 2 explains LightSpun's expansion from dental into vision, pharmacy, and behavioral health using a LAMP strategy to replace legacy systems incrementally.

  10. 36:58 – 38:52

    Go-to-Market Expansion via Trust and Consolidation

    Founders describe a go-to-market approach where success in dental earns trust with the payer, allowing expansion to other units within the same enterprise.

  11. 38:52 – 42:20

    Contrarian View on Regulatory Speed and Modernization

    Deepak Ramaswamy argues that regulated industries can modernize faster than expected, challenging the view that compliance is a major barrier to speed.

Claims worth verifying

LightSpun has raised $18 million to build an AI-native administrative platform for dental insurance.

Unidentified speaker (speaker_0) ·0:51

How to verify: Check LightSpun's press releases, Crunchbase, or PitchBook for funding announcements and amounts. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Dental insurance providers can wait 30 to 60 days to get paid under legacy systems.

Unidentified speaker (speaker_0) ·1:10

How to verify: Review industry reports on dental insurance payment cycles or provider surveys regarding reimbursement times. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

LightSpun can adjudicate a routine claim in under a second and pay it seconds later.

Unidentified speaker (speaker_0) ·1:10

How to verify: Review LightSpun's technical documentation, case studies, or third-party performance audits. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Dental administration is operating 10 years behind other areas of operations.

Unidentified speaker (speaker_2) ·2:12

How to verify: This is a subjective industry assessment; verify by comparing technology adoption rates in dental vs. other healthcare sectors via industry reports. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

18 cents on the dollar is spent on administration in US health insurance, totaling around $17 billion.

Unidentified speaker (speaker_2) ·5:46

How to verify: Calculate based on total US health spending data from CMS (Centers for Medicare & Medicaid Services) and administrative cost studies. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

The Congressional Budget Office (CBO) publication for 2025 ramps up $21 billion additional annual savings available from the market.

Unidentified speaker (speaker_2) ·6:11

How to verify: Search the CBO website for publications from 2025 regarding health insurance administrative savings. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Credentialing and onboarding a provider takes an average of 60 to 150 days.

Unidentified speaker (speaker_2) ·6:50

How to verify: Review industry standards for provider credentialing timelines or surveys from dental associations. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Every dental office loses around $87,000 per year due to credentialing lag.

Unidentified speaker (speaker_2) ·6:50

How to verify: Verify the calculation method (days * daily revenue) against average dental practice revenue data from sources like the ADA or MGMA. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

LightSpun is SOC2 Type II certified and NCQA compliant.

Unidentified speaker (speaker_2) ·26:19

How to verify: Check LightSpun's security documentation or compliance badges on their website and verify with SOC2/NCQA registries. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

LightSpun was built in 2022.

Unidentified speaker (speaker_2) ·29:47

How to verify: Check LightSpun's company history, press releases, or SEC filings (if public) for founding/incorporation dates. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

LightSpun believes it can cut administrative costs by 50%, which is described as a conservative estimate for an AI-native solution.

Unidentified speaker (speaker_2) ·35:37

How to verify: Request LightSpun's case studies or white papers detailing cost savings for specific clients; compare against industry benchmarks for dental admin costs. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

LightSpun is expanding its AI-native eligibility and payment solutions into vision, pharmacy, and behavioral health.

Unidentified speaker (speaker_2) ·36:17

How to verify: Check LightSpun's website (lightspun.com) for product offerings; review press releases for announcements in these sectors. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Payers often operate dental, vision, hearing, and life insurance as separate units with independent legacy systems.

Unidentified speaker (speaker_2) ·37:53

How to verify: Review annual reports of major US payers (e.g., UnitedHealth, Cigna, Aetna) to see segment reporting; analyze industry reports on payer IT architecture. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Regulated industries can speed up compliance and modernization significantly using AI, contrary to the belief that regulation is a major barrier to speed.

Unidentified speaker (speaker_1) ·39:15

How to verify: Compare implementation timelines of AI tools in healthcare vs. other regulated sectors; review regulatory guidance on AI adoption in healthcare. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Companies that do not rapidly adopt AI and automation will exit the marketplace.

Unidentified speaker (UNKNOWN) ·40:01

How to verify: Track market share data of dental admin software providers over time; identify companies that have exited or been acquired. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

LightSpun's strategy is to earn trust in dental first, then expand to other benefit lines within the same payer enterprise.

Unidentified speaker (speaker_0) ·38:05

How to verify: Review LightSpun's customer list and case studies to see if they have expanded accounts beyond dental. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

The 'LAMP and expand' strategy involves giving good ROI for a single product, then adding more products to replace the legacy system.

Unidentified speaker (speaker_2) ·36:58

How to verify: Analyze LightSpun's pricing model and sales deck; interview current customers about their adoption journey. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Dental insurance handoffs still rely on fax, paper, and legacy software, creating an opportunity for end-to-end rebuilding.

Unidentified speaker (speaker_0) ·41:09

How to verify: Conduct primary research with dental providers and payers to assess current technology stack; review industry surveys on dental admin tech. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify LightSpun's $18 million raise via SEC filings (e.g., Form D) or press releases.
  • Check the Congressional Budget Office report referenced for the $17 billion administrative cost figure.
  • Research the average provider credentialing timeline and costs in the dental sector via industry reports (e.g., ADA surveys).
  • Compare LightSpun's 'under a second' adjudication claim against industry standards for legacy claim processing times.
  • Investigate the LAMP strategy's application in other healthcare verticals to assess cross-selling viability.
  • Review LightSpun's technology stack for compliance with HIPAA and other healthcare data regulations.
  • Analyze the competitive landscape for AI-native administrative platforms in dental and ancillary benefits.
  • Assess the scalability of AI agents for schema mapping in legacy systems with varying data quality.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): A lot of our competitors are still bottling AI or their traditional legacy platform, whereas we are more AI native.

0:07Unidentified speaker (speaker_1): We have this AI tools that'll automatically help you migrate that data from the source system to to our system without really requiring an army of data engineers. If you don't think of that from the beginning to make your platform friendly to AI agents, It's it's gonna going to be very hard for them to retool it. And that's the advantage we bring.

0:25Unidentified speaker (speaker_0): So if you don't rapidly change and adopt, you're going to be extinct, in my opinion. You modernize or you exit. With AI moving at a rapid speed, with intelligence, with automation, if you are not open to that change, you will die.

0:50Unidentified speaker (UNKNOWN): Welcome

0:51Unidentified speaker (speaker_0): to Terminal Value. Today I'm joined by Shaju Puthasseri and Deepak Ramaswamy, co-founders of LightSpun. LightSpun has raised $18 million to build an AI-native administrative platform for dental insurance. Dental insurance still runs on fax machines, paper explanations of benefits, and core systems built decades ago. Providers can wait 30 to 60 days just to get paid. LightSpun says a routine claim can now be adjudicated in under a second and paid just seconds later. So we get into what it takes to rebuild the entire back office from eligibility and credentialing through claims and payment. Why deterministic workflows should stay deterministic while clinical judgment remains with licensed professionals,

1:32Unidentified speaker (UNKNOWN): how

1:32Unidentified speaker (speaker_0): AI changes the work of migrating customers off legacy systems, and why dental may be the wedge into vision. pharmacy, and other benefits. Now, on to the episode. Deepak, Saju, thank you so much for joining the podcast.

1:47Unidentified speaker (UNKNOWN): Thank

1:47Unidentified speaker (speaker_1): you so much, Nick, for having us here. Glad to be

1:50Unidentified speaker (UNKNOWN): here.

1:51Unidentified speaker (speaker_0): So before we get into LightSpun, I'd like to set some context on the dental administrative space in general and how the legacy environment works. So could you walk through how uh provide provider coverage setup works, how claims setup works, and decisioning and the payment and service process today, so we can better understand the legacy environment.

2:12Unidentified speaker (speaker_2): Nick, I'll be happy to take that. Nice to be on your show. You're asking a great question about the legacy systems. Dandali Dental is one of those ideas areas we are still operating 10 years behind. compared to other areas of operations. In Dental, if you really think about from a provider side, from a payer side, from a member side,

2:39Unidentified speaker (UNKNOWN): a

2:39Unidentified speaker (speaker_2): lot of things are still old-fashioned, so-called old-fashioned. You know, the provider and the payer

2:46Unidentified speaker (UNKNOWN): uh

2:47Unidentified speaker (speaker_2): quite often communicate with tax fax emissions. missions.

2:50Unidentified speaker (UNKNOWN): Uh

2:50Unidentified speaker (speaker_2): EOBs, queries, patient information is transmitted to fax missions. Um the member still gets EOBs. We all get EOBs in our mail when you go to a dentist. Instead of talking about um what's the outcome for the member or the patient, we talk about how much should we paid and how much should we what is that difficult? for and so again dealing with paper. that comes through your front door every day when you have it to sit with the dentist. The payers still um adjudicate claims on legacy systems. We have some payers who still use a scobold or rather uh old operating system. Not they are uh bad, but very difficult to integrate, you know, difficult to communicate, exchange. information. Also, the source of the amount of resources available to maintain are very hard. Also, even though we have AI and agents, it still takes a lot of time and effort to maintain those kinds of systems. So the infrastructure is still outdated. Data, We are still banking on 10-year-old technology. The operations that manual. The experience factories are

4:06still old-fashioned. Take the case of payment.

4:10Unidentified speaker (UNKNOWN): The

4:10Unidentified speaker (speaker_2): dentist gets paid probably 30 to 60 days after a visit. We still send um printing of all the bills,

4:21Unidentified speaker (UNKNOWN): uh,

4:21Unidentified speaker (speaker_2): line by line, manual process, paper copies are used to reconcile the payment. Even though we have a fintech uh evolution and revolution happening. Uh so the infrastructure itself is old fashioned, the operations are manual, time consuming, error prone. Um and that's the s the state of uh a lot of tender insurance, administrational education.

4:48Unidentified speaker (speaker_0): Gotcha. And so

4:49Unidentified speaker (UNKNOWN): you

4:49Unidentified speaker (speaker_0): understand the personas a little bit is can you help me understand at the insurer, who's taking the brunt of this pain? And um is that the right way to think about it, or is it also being felt at the uh dental office as well? Like how how can I think about that dynamic?

5:05Unidentified speaker (speaker_2): Yeah.

5:06Unidentified speaker (UNKNOWN): I

5:06Unidentified speaker (speaker_2): can take this and maybe I will ask the fact to Your own uh kind of uh pitch kitchen. in yeah. You know, think about we have around three hundred 300 million Americans have some form of insurance. It could be Medicare, Medicaid for the elderly population, the young children, the Medicaid. Um, a lot of the middle value is for employer group based, you know, because they have employer group sponsored, we call the commercial group business. Then you have the small population that leverages individual that helps, they'll work, right? So some form. you have it that you'll cover. And 300 million, that's more than the number of people who voted in the last presidential election. You know, if you were to just put some of these numbers in in space. Who is taking the brand of it? 18 cents on a dollar is spent on administration. Right? If you were to total that up, it is around seventeen 17 billion. in purely administration, not directly going to the uh care. And if you look at the CAQ

6:11publication for 2025, they ramp up 21 billion additional annual savings available from the market. So the cost is not only the payers side of expense, it's also borne by the employer group who is sponsoring the cost of care, as well as the providers. You know, they get very lacked payments. For example, credentialing, one of the issues that we have solved. You know, in order to credential and on board a provider, take an average of 60 to 150, 120 days, large for each. And every dental office or every dentist does around 87. 87K per year if there's a lag in Ortology. So everyone feels the pain point in it. And most important, if you really look at the patients, us, the members who receive the care, we don't get access to care. You know, we are not focusing on go to outcome, oh build and we be a are focusing focus on even administration.

7:12Unidentified speaker (speaker_0): And so I I'd love to then take the chance to pivot into light spun's offering i think you guys are um have built a pretty expansive platform that handles a lot of the administrative components and so if you could walk through that journey from provider setup coverage setup the actual claims decision process and you know down to the payment how does where's light Lightspan spun involved and how does that uh drive innovation to that process that you've outlined?

7:39Unidentified speaker (speaker_3): So our job, our platform is basically,

7:42Unidentified speaker (speaker_1): you know, our goal is to automate the entire back office, right? So which means right from the moment uh a claim in, uh, right from the moment uh an eligibility file comes in, or uh a provider office roster comes in, or we get the initial set of credentials for a provider for credentialing purposes, right from the that point all the way to the end point, which is like you made a decision on the claim or the payment. So the entire entire back office, entire that is basically where we where we work on and we are looking to automate, right? So uh and and that's the best way to think about it. Uh because the end of the day we are not taking it, okay, let me solve one problem right here. You you really have to look at all the different pieces of the workflow and see what you can do to impact everything. So it's um the you know, the the medal model I would leave with is like, okay, I I've done a I'm a I'm a provider, I've given care to my patient. What's next? What what do I do next? So right from the transmission of

8:47the claim all the way to them getting paid, the pro the member uh getting proper communications, that entire thing needs to be automated. And that's where we basically work on.

8:57Unidentified speaker (UNKNOWN): Uh

8:57Unidentified speaker (speaker_1): Sharjugan, if you want to add something more to that.

8:59Unidentified speaker (UNKNOWN): Yeah,

8:59Unidentified speaker (speaker_2): if I may say what I have been in this industry twenty plus years, I watched the technology gap, the operating gap in providing services, both from the payer side as well as uh while we were both at Overjet, uh you know, we did the innovation for the clinical side, but the technology and the administration side was left behind. And that's when Deepak and I decided to start um LightSpun, uh purely to address this administrative problem. So when we built what we have built over the last four years. is the core of insurance administration platform, the backend. Deep enumerate, you know, the eligibility issue, train submission, train payment. How do you bring in a new group? You know, when you acquire a new group, onto To a platform for a pair, credentialing, how do you onboard a provider, network management?

10:03How do we stack different network? You know, quite often, how do you bring different fee schedules for for a provider payment? The actual fintech payment, how do you integrate our platform with call centers? So that instantaneously using AI, you can provide information about that call, as provider call or or uh an employer call or a member call. All of those things are integrated. So now you see it is not a single product point solution, it's a platform.

10:34Unidentified speaker (speaker_0): You

10:34Unidentified speaker (speaker_2): have integrated that end-to-end experience from the time you receive the claim to the time you pay and settle back instantaneously.

10:44Unidentified speaker (UNKNOWN): Now

10:44Unidentified speaker (speaker_2): Deepak will go into much more detail. We are able to adjudicate a normal claim you know, typical claim, less than one second. And pay in a couple of a couple of seconds, if you add, we can pay that claim. So we are really looking at that real transformation, right? In order to do that, we have to re-rebuild the core of the infrastructure, the back end, using an AI first, API first, agentic core. that is helping us to adjudicate instruments.

11:19Unidentified speaker (speaker_0): I see. What one of the um points that you've both spoken on in prior podcasts is this fine line between what LightSpun owns in terms of automation and then the human decisioning aspect. And I think that's very fascinating. I've in multiple conversations in the past, uh

11:36Unidentified speaker (UNKNOWN): I

11:36Unidentified speaker (speaker_0): often kind of probe pro-guest guests to understand. how automated can this process get, both from your own product and then the customer implementing their own automation on top. And so I'd love to use the claims adjudication setup and processing as that example to just understand your points of view is how do you think about where your solution or what it can automate and then what points you decide to give back to the customer. And equally importantly is. how does a customer think about that themselves?

12:06Unidentified speaker (speaker_1): Let me take that sharp. That's a great question, Be. Let's uh step back a little bit and talk about sort of the the the traditional challenges and for example something like fence processing, right? So one is um

12:21Unidentified speaker (UNKNOWN): you

12:22Unidentified speaker (speaker_1): know you have the problem of volume and speed, the usual sort of scaling scaling issue issues. But I think the fundamental problem is there are just a lot of variables uh in processing a claim accurately. And these can be very complex and if you make a single mistake, it can have repercussions for the patient, for the provider, and of course for the pay. peer as well. So you have um this uh complexity uh you know complex universe when you process a claim. And then secondly, you will get uh information, you know, sometimes uh uh you know peer or providers will attach all kinds of uh

13:08Unidentified speaker (UNKNOWN): you

13:09Unidentified speaker (speaker_1): know disjoint disparate information and if you're doing case processing you have to make sense of it right You have to know if this condition was present uh last time or or this payment was made previously. So you have to it it's a world where fundamentally it should be driven by a sort of a skeleton of uh you know well-defined deterministic rules, but then you have this complexity where the cost of doing a mistake is is is is great, and then you have different data sources. So that's the balance. backdrop backdrop of the world uh we we live in. Now in terms of how we think about it is um anything that is um you know sort of rules based deterministic, we will automate it one hundred percent. Uh you you know if you do one plus one and get two today, you should get that same number tomorrow as well, right? Um and uh that that we are very unwavering on. So we should we we don't want to compromise and introduce AI and workflows which have which should have a deterministic outcome. Now, uh the approach we have taken is uh, for example, uh when we do data ingestion, there could be some ambiguity when you have millions of documents for when or we are onboarding a payer, or you y know, you have sort of disparate uh data. coming in

14:27which contradict each other, right? And that's where we cede control to to our customers and say, I I got data X from this source, I got data Y from this source. Uh

14:38Unidentified speaker (UNKNOWN): what

14:39Unidentified speaker (speaker_1): what do you want us to do uh with that point, right? Do you want us to simply always prefer X or do you want to have a built built-in in audit mechanism and a sort of staging area where you can bless it? and so on. So it we have taken an approach of being very pragmatic and seated or given controls to a user so they are in full control of the automation, right? And wherever possible, if the AI is uh a little suspect or in terms of having low confidence, we'll explicitly point out this is what we think it is, and you, the human uh reviewer, should actually review it and make sure it's okay. Right? Because you know, end of the day, this is a very compliance heavy industry, right? We have to account for everything that you know our our system does, including AI. Everything should be documented because uh if you if you document it, then you're able to explain it. So so I would say in a nutshell, our approach is to automate it whatever possible.

15:39but make the make our users, our customers, be aware of the decisions you're making. And also ultimately give them control so they can control it. So um uh yeah, I think Shaju, if you want to add something small to it, I think uh that's that's how I would characterize.

15:55Unidentified speaker (speaker_2): Yeah. In in nutshell, we have a number of agents that work can concurrently, can only. AI agents that concurrently to automate the process. Lots of models deploy. But anything that involves clinical decisions, anything that involves diagnosis, even though it's an adjudication process, what is allowed and not allowed,

16:22Unidentified speaker (UNKNOWN): we

16:23Unidentified speaker (speaker_2): lead up to the clinical expertise. Right? But purely the repetitive administrative tasks, combining all the different pieces of information required, or looking back the history of a patient, the tool, the surface, the type of treatment that are done, and the effect of that, those repetitive manual tasks can be automated through AI agents. Or things like you know, you know, different habits of the patients that are uh captured through the process. Uh or the payment. Payment is a good example. You know, how much is the deductible, how much is the coinsurance, is there a copay, reconciliation of payments to the dentist, EOPs. your peas. They're all in manual repetitive. tasks, which relies a lot on paper today or pro-digital as we advance. All of those items are agency, and that's the new uh mechanisms that we have built, because we are built from the scratch through API, but leave the

17:33clinical decision uh to the clinicians. So those are those type of claims that require clinical. review. We allow to be experts. But even though you can still use AI agents to assist clinical review process, we leave them to the licensed clinicians make that it doesn't

17:50Unidentified speaker (speaker_0): understood. One of the interesting bifurcations that I often hear uh around model usage is a lot of the

18:00Unidentified speaker (UNKNOWN): uh

18:01Unidentified speaker (speaker_0): founders that I've interviewed have gotten to the point where they say, hey, you know, I use an eval. Once my models reach that eval, there's no need to, you know, continue to uh need to push to the frontier. And then on the other side, there are evals that because the frontier models continue to get better, they just keep rising on the evals. And so I'm very curious as you guys think about your own product architecture. Do you see model usage as unlocking incremental use cases? Or are you at the point where you feel like, hey, you know, I'm able to go drive cost reduction because the incremental model is not driving much more value for my customer in the in this uh context?

18:43Unidentified speaker (speaker_1): I I think there's uh truth in both both of those, but I would say leaning more on the first one, we we see increasingly uh you know, as the frontier models get better and as we fine-tune to our use cases, right? Because at the end of the day, the models have to understand our domain, the insurance domain, the general domain, the vision domain, whatnot. And as as they get better,

19:06Unidentified speaker (UNKNOWN): uh

19:06Unidentified speaker (speaker_1): we unlock more use cases. So for example, uh you you know, I'm just putting in putting an example out there, like maybe the first word general. gener Gen Gen Zero, Gen 1 models would look at uh a claim and make an assessment. But now you you can look at the entire claim history for that patient. You can look at the history at the wire offices and make maybe a different decision, right? Uh or you know it it doesn't have to be a clinical review, it could be some advanced. administrative uh things that are there. So as those models become more powerful, as their ability to reason become powerful, and as we fine-tune and as we learn from our own sort of experience, and uh so they are continuously um on unlocking for us. Now, of course, uh there'll be some cases where we we but get to a really good steady state and then we don't have to touch that case. For example, if there is a particular form that we use a modus for and it has a ninety-nine point nine percent accuracy with very little errors, you know, pragmatically we don't want to be spending investing a lot of time on squeezing all the coins zero zero zero one if there is very little incremental value for the customer. Right. So so there will

20:26be some endpoints as we reach for some of the models, but really it's on taking the multiple points of data, looking back across many things. Is that where it gets very exciting for us to unlock those pieces?

20:39Unidentified speaker (speaker_0): Yeah, to that point, do you find that the challenge today is more on getting access to the data, organize the data, than it is actually model selection? Like, how do you think about the the like if you you were to pick one product? problem that was the core focus of your thought process, say D Bac, like how do you think about that?

20:55Unidentified speaker (speaker_1): I said that's an excellent point, Nick, and that's a really good question. That I would say is very important. It's really how you organize the data and and you know end of the day make it make so that you burn as fewer tokens as possible, right? And then have the right tooling in place uh so that you don't re-thread things that that have been already been solved. I think that is fundament very, very key for us to build uh you know an efficient agent experience for us. Um and then then of course asking the right questions and really uh you know really understanding the business. Uh because as Shaji was saying at the beginning of the uh you know, in the beginning, a lot of the things like uh network stacking, scheduling, uh, you know, which feast. fees to use, they become quite complex. And if you build the right if you train your models appropriately, then those questions become easy to answer. But if you don't, then it becomes very hard, right? So

21:52Unidentified speaker (UNKNOWN): uh

21:52Unidentified speaker (speaker_1): data organization is extremely key for us, and that's you've been uh um I would say a point of emphasis at Light LightSpun Spun since day one.

22:02Unidentified speaker (speaker_0): Makes sense. I'd love to shift a bit to just the the go-to-market side. Uh my understanding of the the landscape that you guys target, both on the insurer side and on the dental service organization side, is that they're pretty uh consolidated, you know, very top-heavy uh buyers in a sense or or customers. And so how do you guys think about targeting uh folks who are you know so consolidated in that sense?

22:27Unidentified speaker (speaker_2): yeah uh I can take that um

22:30Unidentified speaker (UNKNOWN): uh

22:30Unidentified speaker (speaker_2): so Nick when we began to build a product obviously we we built a platform but the platform didn't come day one we started off with the first product which was a shilling product and we were um solving a problem of tiny timing, right cost delay.

22:52Unidentified speaker (UNKNOWN): And

22:52Unidentified speaker (speaker_2): we noticed this problem existed both on the payer side as well as the dental service organization side. Obviously, there was as a as a startup company a lot of fear of adoption for a large payer, or even on the dental service organization side, that change management has been very hard because we're they're all used to doing a certain work in certain order. Here comes something automated, you know, we go to 11 different sources, API, agents, through AI, extract information, consolidate and give you a reporting to that part. Not used to it, it's a sticker shocker. Not now. Right now, a lot of uh industries really attuned to the AI. Um but for us to reach that, uh we really focused on what the pain points were and what the actual solutions would do. And that got the attention of ADA. ADA is not um a very mature large organization. They have almost every dentist in the country as part of their membership. Credentialing was a core problem. when we were talking to DSO. ADO called up and said, Shadju, this is the same problem we have to source. And

24:10Unidentified speaker (UNKNOWN): you know,

24:10Unidentified speaker (speaker_2): they don't usually partner up with startups, but they did. They they we were even the delegated credentialing to CHUH, another larger healthcare controller, we were able to solve that problem of credentials. credential.

24:25Unidentified speaker (UNKNOWN): So

24:26Unidentified speaker (speaker_2): each time.

24:27Unidentified speaker (UNKNOWN): for

24:28Unidentified speaker (speaker_2): us to the go to market was deploying high value, high quality product, but change management was in the heart of it. And you know, we had some clients who couldn't take that faster change. There was a pace of change. I'm sure Nick, you would have probably gone going through when you built build the the you know the the the payment solutions, right? The CRM solutions. So that's the exactly the same process that we use with the nation's benefits. Through heavily uh concentrated with um compliance using combining real-time adjudication and fintech, bringing it together, right? They had the fintech engine, which is you know instantaneously pay the providers. We have the auto-adjudication within a one second, one second, two seconds, three seconds, in order to adjudicate and pay a while. Right? And whenever we did this production ready, each of our sales became easier and easier Again, if you go back to Smile America Partners, they they service over 500,000 school children,

25:478,000 schools across many states. They had delay in onboarding at the wider 60 to 90 days. We cut short less than every,

25:58Unidentified speaker (UNKNOWN): right?

25:59Unidentified speaker (speaker_2): That's money safe, more access to the kids. Obviously, the heart of it is platform, the code platform, the NTDK, the the automation that we created for credential. Now, we were able to do all of these. We also paid attention to compliance. Yeah, SOC2, type two certified, high trust certification, NCQA compliance, all of those pieces to get you know pieced together. on an automated platform is the value that we can give on top of the business value that we provide to our consumers.

26:34Unidentified speaker (speaker_0): Makes sense. How do you think about deployment and how individualized does it have to be? Uh, you know, a lot of the uh uh more recent applied AI companies that I speak to, they have a pretty intensive forward-deployed Ford deployed engineer who will go in and help build the contextual layer or add to the harness. Like, how do you guys think about that deployment process? in in your industry?

26:54Unidentified speaker (speaker_1): Very, very, very similarly to exactly what you said, Nick. Uh, because the AI has enabled a lot of the deployments, uh, you know, made it more easier, right? So it it also depends on the type of customer, the approach that we take, for example, for large pair, peer may not be the same as we take for for a general service organization. But the the our approach have always has been, even before the advent of the AI, is to meet uh the customers where they are, right? We we understand some of our customers have some of you know mature systems, well-defined integration points, but some of them don't. But the advantage we have with this uh same forward the correlate engineer model is to kind of build it out pretty quickly and build all those integrations pretty quickly. So while we have uh your standard API, your standard endpoints, which will make integrations easily, we we recognize that customers may not have uh their data in the best possible format. So But that's where our engineers kind of come in and build that uh layer to make that easier. And that's that process

28:09is getting increasingly easier, uh, you know, uh in uh almost in real time for us.

28:15Unidentified speaker (speaker_0): That's fantastic. I'd love to shift a bit to to just the competitive landscape and um and and and even equally importantly, is just the the state that customers are in today. Like when you are uh negotiating a deal with customer in an RFP process, are you finding more often than not that you're dealing with just a insurer's own tool set that they've built themselves, or are you ripping and replacing a a a legacy um competitor?

28:44Unidentified speaker (speaker_2): It's both. Uh what we see. A lot of payers pairs have home homegrown systems. built homegrown, very comfortable over the years.

28:54Unidentified speaker (UNKNOWN): A

28:54Unidentified speaker (speaker_2): lot of these pairs have built a system in early 80s or late 90s, running on client server. Some of them still hosted internally, even though a lot of them are moving to cloud, or some of them have already moved into cloud. So that is one segment. Then we have three. third party uh solution providers who provide this as a software as a service to payers and TPAs in the marketplace. So we we see both of them. What we see in a sense is um none of at least uh as of today, none of the offering has an AI AI-native native solution, right? And we were a little more fortunate that we built a system in 2022, took advantage of the technology at hand, but a lot of the existing vendors bolt AI onto a traditional platform, right? And that has put its own repercussions and issues and synchronizations, Organizations,

29:58integrations, as well as a talent pool that is required. to maintain and take to the next level. Right. Not very easy to, even though with Quad and other tooling, you can do that today, but still harder to integrate into like a file-based system. Right. So a lot of competition competitors are still bottling botting AI on the traditional legacy platform, whereas we have more AI. native. That also reduces the amount of time that we will do for large enterprise implementations. Again, we have more tooling at our disposal to analyze and integrate and convert traditional data with it.

30:40Unidentified speaker (UNKNOWN): I see.

30:41Unidentified speaker (speaker_0): Shahju and Deepak, I'd love to dive into that point because I it's a big point that a lot of the founders make on this um channel. is like we're we're built natively for this and it changes in every scenario the the impacts of that. And so could you help walk through like you're sitting with a customer who has the problem of claims adjudication or credentialing, how does an AI native architecture tactically help you deliver value like what does a customer see when they're comparing it to like a legacy provider like a Skygen or whoever or their own homegrown tool? Like what does that look like tactically?

31:13Unidentified speaker (UNKNOWN): Yeah

31:13Unidentified speaker (speaker_1): let me take a let's stab stop it at that. Nick, that's an excellent question. So, you know, let's take a scenario where you have, let's say, a legacy platform, right? And now it the business says, uh, hey, we have these new groups that we want to onboard. My plane's processing volume is going to grow by 2X or 3X or 4X. Now you you have a problem which is like you have a legacy system. You may need new APIs. You know, there are new compliance APIs coming all the time, and you will need new data volume. Uh that will probably make it very difficult to work with a legacy system, right? Because uh compliance requirements change all the time, customer requirements change all the time, vendor requirements change all the time, and then you may have a legacy system where you're not able to scale, skip. right? So the AI. tooling, like if you compare that scenario with what we offer, is that when we have a customer says that, hey, we I am going to get end of this year, I'm going to get a 5x small volume, great. We have this AI tools that will automatically help you migrate that data from the source system to our system without really requiring an army of data engineers to do all the you know schema mapping. You still need to have a human in the loop right to kind of validate, but you don't need uh that heavyweight

32:32uh experience of of uh of uh data migration for example. The there's a concept of business rules mapping uh that can be automated as as well through AI. And fundamentally you have uh uh you know AI pointing out uh data inconsistencies for example are the duplicate data, dupes, you know, the standard sort of data issues. So that process doesn't become uh easy if you have uh uh sort of a legacy system, right? Because of fundamental limitations on what APIs can you have bought on to it, or or uh uh what kind of data a schema, for example, you we can change in the legacy system. For example, that's uh that becomes an issue. So when we come in and say is that, you know, your system migration, rear migration, we have an army of uh agents kind of handling that. And then once you process the claims, uh, you know, we have the automation built in and then sort of the AI cardrails against each step so that you're not paying to the wrong provider or not sending communications to the wrong member. that'll prevent those things. So those you if you don't have an AI native solution, if you don't think of that from the beginning to make your platform friendly to AI agents, you it's going to be very hard for them

33:52to retool it. And that's that's the advantage we bring.

33:56Unidentified speaker (speaker_0): Understood. Uh I'd love to to shift a little bit just to the future uh maybe kind of two bits of the future futures. is one and One is just the near term, where is Lightspan focused today in terms of expanding the platform or or or the offering more generally?

34:09Unidentified speaker (speaker_2): Yeah, I can take that, Nick. Um great question. How how does the future look good uh in the AI uh area or with the AI revolution that is happening? Um obviously the administrative costs cost can be really cut down. using AI or show. That's simply as that, right? Um, what we are doing in Dental is a proof part, right? Um, how do we using Dental as the base? How do we reduce the cost? How do you tackle that cost problem? Now, with this AI native um infrastructure, uh, we are also able to extend into other areas like wishing, pharmacy, the other ancillary areas. That will obviously is a pretty heavy expensive area to solve. So we are already doing uh integration with the fintech. You know, that was the example of nation's benefits platform where adjudication and fintech come together by right? the payment. Ten years ago,

35:17you know, we were paying so much for processing and payment, the fintech has solved that problem and it has made it more instantaneous. So we are expanding naturally, taking the core of administration, solving the eligibility problem automatically, solving the payment problem, solving the payer provider, network, onboard. onboarding, credentialing, contracting. We believe that we will we can easily cut down 50% is very conservative, right? But as the technology progresses, that administrative cost is going to shrink as long as you have an AI-native solution. But the reason we say is the company that refuses to do it, refuses to modernize, so slowly they will exit the marketplace because you won't be able to compete, compete on cost, right? So naturally we want to expand to larger enterprises within the devil.

36:17At the same time, we are also expanding similar theses into vision, pharmacy, all of other, that, even medical, some areas. for medical, right? And you know, like behavioral health

36:30Unidentified speaker (UNKNOWN): and

36:31Unidentified speaker (speaker_2): so on and so forth. You know, many of those places we can expand.

36:34Unidentified speaker (speaker_0): Shaji, you mentioned something very interesting, which is the expansion into other components of the healthcare, which I don't think I've heard you guys talk about in the past beyond dental. And so I'd love to get your um point of view on are do you have go to market advantages from your existing customer base or are the insurers that you work with say are they primarily dental? Like how do you think about that go-to-market expansion, you know, medium long term?

36:58Unidentified speaker (speaker_2): Yeah, that's a great question. Because we follow a LAMP and expand because it's a platform solution. We we give you good ROI for a single product. Um then add more and more of those products with the AI native solution. and eventually replace your your net your native old-fashioned legacy system. However, there are adjacent products in the same pair. For example, a lot of pairs will have dental, vision, hearing, life insurance, absence,

37:33Unidentified speaker (UNKNOWN): you

37:33Unidentified speaker (speaker_2): know, you name it, accident. So all of these ancillary areas have to the same issue, just like we enumerated in that, right? There's an eligibility, there is a payment, there's another duplication. So those are all natural expansions for us, building on the strong thesis of what we have achieved. Any pair, any large pair, you know, they have all these multi line of products, but they all operate independently with the separate unit within their there whole for enterprise uh business.

38:05Unidentified speaker (speaker_0): That's fascinating. So you because these guys are so consolidated, they basically have a bunch of different providers and your thesis is hey, if we can do dental exceptionally well in an AI native manner, we can earn the trust that unit talks to the next unit and we we expand. That's correct. That's very cool.

38:19Unidentified speaker (speaker_2): But we have to crawl before you right Raj.

38:22Unidentified speaker (speaker_0): Yeah. yeah we we had a uh a saying in our company, um, crawl, walk, run marathons. Which I'm not sure if marathon applies because it may not be faster, but it's uh understand where you guys are coming from. So I I want to thank you guys both. This has been fantastic. I'd love to leave each of you with one question. Uh what is one opinion that you have that is contrarian that the market doesn't yet understand today? Maybe Deepak starting with you and then moving to Shaju? Shaji?

38:52Unidentified speaker (speaker_1): Hey, Nick, that's a that's a great question. And uh I'm sure you know if I answer this, some someone will have the same view. But uh I I do think uh my my view of this is the you know I I do think the market understands this, but they they probably don't appreciate the you know the the traditional dodgy or traditional industries. Uh you know, we can kind of see change real happening in real time, right? I I don't know if this is a contrarian view, but at least it's the view I had. It's like, okay, it's a very uh compliance-oriented industry, very regulated industry. You know, you have to do things a certain way. Well, you can do things that same way, but you can also speed up it tremendously, right? And I I don't think it's it's that much of a barrier uh as what some might think. At least I I didn't, you know, that that's my own view.

39:51Unidentified speaker (speaker_0): I like that.

39:53Unidentified speaker (speaker_2): Dick, I will go. Um change is a new constant. Okay.

40:01Unidentified speaker (UNKNOWN): A

40:01Unidentified speaker (speaker_2): lot of folks, lot of our client-based think that change can take like the change that they did 10 years ago. That's not going to work, right? So if you don't rapidly change and adopt, um, you're going to be extinct, in my opinion. That could be contrary in view.

40:24Unidentified speaker (UNKNOWN): Uh,

40:24Unidentified speaker (speaker_2): you modernize or you accept. exit. Right? Especially with AI moving at a rapid speed, with intelligence, with automation. If you if you are not open to that change, you will die.

40:39Unidentified speaker (speaker_0): This was fantastic. Thank you, uh, Deepak and Shadu. Thank

40:42Unidentified speaker (UNKNOWN): you

40:42Unidentified speaker (speaker_0): both very much.

40:44Unidentified speaker (speaker_2): Thank you. Thank you for having us on your show.

40:47Unidentified speaker (speaker_0): Thank you for

40:47Unidentified speaker (speaker_1): having us today. Nick. Thank

40:48Unidentified speaker (UNKNOWN): you.

40:48Unidentified speaker (speaker_2): Bye-bye. Bye bye.

40:49Unidentified speaker (speaker_0): That was my conversation. with Saju Puthaseri and Deepak Ramaswamy, co founders of Lightspun. I took away three themes. First, many of the best applied AI opportunities live in the administrative plumbing that no one has rebuilt. Dental insurance connects provider data, benefits, claims, payments, and member communications, but the handoffs still often rely on facts, paper, and legacy software. Rebuilding the workflow end end-to-end to end can create more value. than just automating a single task. Second, the right boundary for AI depends on the type of decision. LightSpun automates repetitive, rules-based work, uses AI to organize messy inputs and surface ambiguity, and leaves clinical judgment to licensed professionals.

41:30Unidentified speaker (UNKNOWN): The

41:30Unidentified speaker (speaker_0): goal is not autonomy for its own sake. It is speed without losing control or auditability. Third, an AI-native AI native architecture can change both implementation and expansion. Agents can help map old data, reconcile inconsistencies, and configure business rules, reducing the cost of replacing the core system. Once LightSpun earns trust in Dental, the same administrative building blocks can extend into vision, pharmacy, hearing, and other benefits. LightSpun's Lightsfund's bet is that healthcare administration can move at software speed without forcing clinical judgment into software. Machines handle the repetition while people retain the consequential decisions. As Saju put it, the choice for legacy platforms may be stark, modernize or exit.

42:13Unidentified speaker (UNKNOWN): Thanks

42:13Unidentified speaker (speaker_0): to Saju and Deep for joining me and thank you for watching Terminal Value. I'll see you on the next episode.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights