theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems | Giovanni Pizza, BonusX | AI Luminaries with Neo4j

More from theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems

Giovanni Pizza

CEO & Founder

BonusX

John Furrier

Co-Founder & Co-CEO

SiliconANGLE Media, Inc.

Info

Giovanni Pizza, BonusX | AI Luminaries with Neo4j

Giovanni Pizza

CEO & Founder BonusX

John Furrier

Co-Founder & Co-CEO SiliconANGLE Media, Inc.

HOST

search

(INTRO)

John Furrier

>> I'm John Furrier, host of theCUBE here in our Palo Alto studios in California, in the heart of Silicon Valley. This is our AI Luminary series with Neo4j. We talk to experts and people in the field working on cool AI projects, platforms, who are building the next generation AI services, who are leveraging graphs, knowledge graphs, and actually putting together the platforms. Giovanni's here. Giovanni Pizzi, CEO of BonusX. Giovanni, thank you for coming all the way in from Italy. How are you doing?

>> All good. Thank you, John, for the invitation. I'm really happy and thrilled to be here.

John Furrier

>> Yeah, well, you're a luminary, so it's just self -identified by us. Congratulations. Let's get into it. Talk about what you guys do, because you guys have a platform, you have services, you have services that tie to relationships and context. Talk about what what you guys do?

>> Basically, at BonusX, we developed a platform that enabled people to discover the tax and welfare benefits they're entitled to and manage the application process end to end. Today, in Europe, four out of 10 people that are entitled to a specific benefit from government, so we're talking about public money, fail to access it because the system is so complex that they are not even aware of these benefits and often they are discouraged by the bureaucracy. At BonusX, basically, we enable people to impute data, create their own household profile, impute financial information within the platform. Based on this data, we match people with hundreds of eligible benefits that may be entitled to in Italy at the national, regional, and municipality level, and are able to manage this application process and bureaucracy end -to -end.

John Furrier

>> So you guys have a complicated, old, antiquated, bureaucratic system out there that people have to deal with, which we know anyone who's worked with any kind of government knows it's hard and certainly, you know, getting those services, there's a lot of old data, but you know, they're in old databases. So talk about what you guys are doing with graphs and knowledge, this knowledge layer, because if you look at the trends right now in AI, it's really turning out to be a core crown jewel in the architecture. Talk about how you guys are implementing.

>> I think it's a very interesting, actually, application of graph. Because if you think about it, when you have a regulation, like tax regulation or welfare regulation, you need to understand if someone is eligible for something, you need to have different kind of information data about the person, about the household. But actually, you also need to understand the connection, the relationship between these data. So, for example, if you have a family allowance, Family allowance may depend on the fact that Alice, a person, has or not a child, and what is the overall income situation of Alice. So a graph database actually is able to provide a much more direct access to this kind of relationship among entities, which in turn we can translate in eligibility rules to match people with the relevant benefits they entitled to. So it enables us actually to represent reality and tax regulation in a much more efficient way than an old -style relationship database.

John Furrier

>> Yeah, I mean, the data business has changed. I mean, you're seeing the search feature become great, but also when you have agents and knowledge layers built in, you see the benefits of accuracy, speed, tying in unstructured structured data. So, you know, Knowledge Graphs has got huge traction And you see in companies like Palantir in the US, entrepreneurs putting it into their AI systems, and it kind of works well when you can tie it together. So I have to ask you, it is gaining traction. How would you define what a knowledge layer is? And why is the concept important for organizations to understand as they look at the massive amounts of data they got coming in and how it relates to AI, not just search and finding something, but like making something actionable?

>> Well, if you think about it, it's often really matter to provide the right context to AI so that AI can actually enable people to reach a specific outcome. So if we think about regulation in a country, in our instance, a lot of regulation is actually based on things like income, financial assets, household, this kind of information. information. And actually the way we are able to connect together all these facts or these data and provide this kind of knowledge to an AI can enable the AI to answer important questions to a person like, what do I need to check to understand if I'm eligible to a scholarship? What kind of data am I missing? Why am I not actually eligible to this specific income, universal income, for example? And this is the kind of support we can provide to people when we have this kind of knowledge layer, because actually we make this information accessible better to AI, know that AI is much more context and much more understanding of the user situation to provide a reliable answer and really support the end user.

John Furrier

>> Yeah, the reliability is key. Talk about the alternatives. You didn't have knowledge layers. What would be your approach? Take us through what would have been the opportunity or what would be the solution? How would you tackle the data problem if you didn't have a knowledge layer?

>> Actually, it would be pretty difficult. For example, I can think about how our platform used to work two years ago. So before we were using Neo4j and having this kind of citizen relationships. So we had to use very complex joints, basically to understand the situation of specific user and then being able to match this user to relevant information, relevant benefits at the government level.

John Furrier

>> So basically, you basically do SQL queries, basically, right?

>> Exactly, so it was possibly SQL that we are still using, but just on some operational things within the platform, not to represent the full citizen profile.

John Furrier

>> Talk about the impact. What's been some of the results? Okay, so take us through the discovery phase. Okay, you have complex joins, they're error prone too, obviously in their complex rows and tables that are great for certain use cases. But what was some of the impact? What were some of the discoveries? Was it an oh, wow moment? Was it, hey, this is amazing? Did it move the needle for you?

>> Yeah, actually enabled us also to create a system by which we can create deterministic rules much faster. So when we were tackling this problem, we divided it into realms. So on one side, there was the legibility and joints where we decided to have a ai automatically translating regulation into deterministic rules but still human checking these deterministic rules in order you know to provide the clear outcomes without any risk for hallucination and then we have the eye that actually is enabling the user to understand why is eligible or not for something so basically having this kind of graph representation of real data enable us to to create also a system where through AI we can create this a very much faster. And for example, now integrating a new benefit within the platform takes 30 minutes. While in 2025 used to take four to five hours.

John Furrier

>> Wow, major time savings. Talk about the data iterations, because I imagine when you have bureaucracy and these old systems, you've got good data and it's laying around in a system of record. And what was some of the data challenges? You got companies, we see companies all the time, hey, I got Snowflake, I got Databricks, or I have some legacy data. And how do you stack that together? And how did you look at that integration? How did you partner and how does AI fit into that? Because I'd imagine when you start building these knowledge graphs, the AI really connects in once you do that. What's that transition? Because a lot of people are in the same boat right now. They have a vision, they go, okay, I got Databricks or whatever. You got a data lake, or they have a database, they've been doing SQL queries, but now the knowledge layer comes in. What's been some of the data ecosystem things you guys have done? How should people think about that?

>> Well, John, I will give you first a number. Today we have 880 ,000 users in Italy, but each of these users has own household on BonusX. So basically, we have data around 2 .5 million people. We collect data about healthcare condition, financial assets, income, real estate, earnings, because all this data are relevant to really match people with eligible tax benefits from government, even the European Union, and then, you know, all relevant services for them. So understanding the connection between these data is actually key to really provide user -relevant information. We are now moving one step further. So we are partnering with a major banking group in Italy to actually integrate financial data, especially financial transaction, to trigger automatic notification of new benefits. So, for example, if you're using your bank account in Italy to pay for kindergarten benefits, I can actually trigger an eligibility check that is testing if you're eligible or not for a kindergarten benefit, which is a national benefit that provides you to up €3 ,600. So putting together all this data is key, actually, to support people to handle bureaucracy. And today we are leveraging on one side Neo4j, so in terms of graph database to have all for this connection. But on the other side, we are also embedding more and more AI at different layers within the platform from document validation to data extraction, to really supporting user and understanding eligibility and how actually to track all this money and unlock this money for their own families.

John Furrier

>> Talk about the AI piece there, because I think the knowledge layer is clear, great benefits, time savings on your side, great user experience, it can get results faster. But when you think about the AI piece, you have ingestion of a lot of content, right? You are ingesting a pre -existing content. Benefits have policies. They have rules. How are you looking at that piece now that you have the knowledge graph? Because you've got the data connection. You've got the AI kicking in. I'm sure agents are on the horizon. Why not just eat up all that data? And then the reasoning starts kicking in. How are you seeing that play out, the reasoning piece?

>> case? Well, basically, first about the data point, there is something very important in our case, which is data access, no? So we need to make sure that the user can have AI only accessing his own information and not other user information, when basically they are interacting with our AI to get support. On the other side, we have, you know, memory, we have all the knowledge which pays about the regulation, national, local, municipality level. So everything provides AI the right context to really help the user tackle the issue. Bureaucracy is a major friction point in everyday people's life. So from a user perspective, a lot of people, especially from low -income families, today are cut out from the welfare services and financial opportunities because they do not understand which kind of services and opportunities they can access and how to get there. So AI here is providing a major help. But in order to really provide help, we need to not give a generalist answer to the end user. So we need to give very personalized support and move the user one step at a time over the application process to really enable them to access this kind of support.

John Furrier

>> So you're saying they give up because it's a pain in the butt to deal with. It's like, okay, they just abandoned. Like, that's what we see, right? You're solving that problem. So you just shortcut the speed to value.

>> Exactly. So we want to make, and we are making economic and social policies much more efficient because people that today are eligible to a benefit created by new policies are automatically notified by the system on how to access it. And then through this conversational system, they can input the data that are relevant to manage the application and having AI that support them step by step. So in understanding the data, in understanding the documents to upload, in validating these documents, in filling in the modules. So we cover different steps over the overall application process to get this public money.

John Furrier

>> Giovanni, it's great to chat with you. Talk about the situation where people now are learning about graphs and knowledge graphs and all the benefits. It's great to see Neo4j because now that the market wants to see this AI integration but a lot of people are leaning in and people have gone down this road and had to rebuild. They still had errors. The trust wasn't there. What's your message to leaders that are scratching the surface right now with knowledge graphs? What would be your advice? How should they think about what that does for them? Maybe potentially be luminary, but also their organizations. I mean, the power that comes out of this. What's your advice to people who are looking at this just from your personal perspective and what you've learned?

>> So thinking about the BonusX experience, we are really happy of the choice we took. We migrated all this data less than one year ago, so end of 2025. So this is something really fresh in our memory, in the product team memory. And this actually lent us to several advantages in terms of time in integrating new benefits, its reliability, about AI answer, and a lot of functionalities and value that now we can unlock for our user. I may also mention a bit about the migration process, because we still had to migrate data about half a million people at the time. So it was actually a process that we had to cover across different weeks and manage really carefully to avoid the data leak, to avoid data loss. But in the end, I think it was definitely worth it. and now, I don't know, we have much more power and we can create much more value for people.

John Furrier

>> Yeah, it's great to see because, you know, you don't see many standards in data. SQL has been around for a long time. Now you've got the graph query language, you're seeing a lot of momentum. What do you hope for, for new tooling? What's on your to -do list? What's on your wishlist, I guess, that's the question.

>> So if I think, you know, about tech, I'm very used to think about user first. So we have a very product mindset within the team. And I see this technology as something that can enable us to provide more and more value to these hundreds of thousands of people that rely on BonusX. So actually my wish is on one side that AI will keep evolving and be more and more accurate so that we can mitigate more and more the hallucination risk over the future. and secondly that this kind of technology can actually really help us to model reality because if we're able to model reality better and better then we can actually also create a digital context that is more accurate faster and then we can leverage this context this data again to provide value and support people across several things yeah i mean it's such an obvious great thing you're doing people can even get alerts hey you're missing out on some some benefits or

John Furrier

>> hey you left the the process here it's back. I mean, you could do amazing things for people. So congratulations and thanks for coming on.

>> Thank you.

John Furrier

>> Luminary series. Another great example of efficiency. Another great example of user benefit. Again, great work Giovanni. Thanks for coming on.

>> Thank you very much. Thank you for an invite.

John Furrier

>> All right, I'm John Furrier, host of theCUBE. This is AI Luminary Series with Neo4j and leaders who are making it happen as they discover the journey of putting in the AI systems. You got to feed the data, move that data to the AI, and then magic happens. The knowledge graph is becoming center of the conversation. Graphs work, graphs are compatible with AI, and it's all working great, and it's just another great surge in the data revolution. Thanks for watching.

Giovanni Pizza, BonusX | AI Luminaries with Neo4j

search

(INTRO)

John Furrier

>> I'm John Furrier, host of theCUBE here in our Palo Alto studios in California, in the heart of Silicon Valley. This is our AI Luminary series with Neo4j. We talk to experts and people in the field working on cool AI projects, platforms, who are building the next generation AI services, who are leveraging graphs, knowledge graphs, and actually putting together the platforms. Giovanni's here. Giovanni Pizzi, CEO of BonusX. Giovanni, thank you for coming all the way in from Italy. How are you doing?

>> All good. Thank you, John, for the invitation. I'm really happy and thrilled to be here.

John Furrier

>> Yeah, well, you're a luminary, so it's just self -identified by us. Congratulations. Let's get into it. Talk about what you guys do, because you guys have a platform, you have services, you have services that tie to relationships and context. Talk about what what you guys do?

>> Basically, at BonusX, we developed a platform that enabled people to discover the tax and welfare benefits they're entitled to and manage the application process end to end. Today, in Europe, four out of 10 people that are entitled to a specific benefit from government, so we're talking about public money, fail to access it because the system is so complex that they are not even aware of these benefits and often they are discouraged by the bureaucracy. At BonusX, basically, we enable people to impute data, create their own household profile, impute financial information within the platform. Based on this data, we match people with hundreds of eligible benefits that may be entitled to in Italy at the national, regional, and municipality level, and are able to manage this application process and bureaucracy end -to -end.

John Furrier

>> So you guys have a complicated, old, antiquated, bureaucratic system out there that people have to deal with, which we know anyone who's worked with any kind of government knows it's hard and certainly, you know, getting those services, there's a lot of old data, but you know, they're in old databases. So talk about what you guys are doing with graphs and knowledge, this knowledge layer, because if you look at the trends right now in AI, it's really turning out to be a core crown jewel in the architecture. Talk about how you guys are implementing.

>> I think it's a very interesting, actually, application of graph. Because if you think about it, when you have a regulation, like tax regulation or welfare regulation, you need to understand if someone is eligible for something, you need to have different kind of information data about the person, about the household. But actually, you also need to understand the connection, the relationship between these data. So, for example, if you have a family allowance, Family allowance may depend on the fact that Alice, a person, has or not a child, and what is the overall income situation of Alice. So a graph database actually is able to provide a much more direct access to this kind of relationship among entities, which in turn we can translate in eligibility rules to match people with the relevant benefits they entitled to. So it enables us actually to represent reality and tax regulation in a much more efficient way than an old -style relationship database.

John Furrier

>> Yeah, I mean, the data business has changed. I mean, you're seeing the search feature become great, but also when you have agents and knowledge layers built in, you see the benefits of accuracy, speed, tying in unstructured structured data. So, you know, Knowledge Graphs has got huge traction And you see in companies like Palantir in the US, entrepreneurs putting it into their AI systems, and it kind of works well when you can tie it together. So I have to ask you, it is gaining traction. How would you define what a knowledge layer is? And why is the concept important for organizations to understand as they look at the massive amounts of data they got coming in and how it relates to AI, not just search and finding something, but like making something actionable?

>> Well, if you think about it, it's often really matter to provide the right context to AI so that AI can actually enable people to reach a specific outcome. So if we think about regulation in a country, in our instance, a lot of regulation is actually based on things like income, financial assets, household, this kind of information. information. And actually the way we are able to connect together all these facts or these data and provide this kind of knowledge to an AI can enable the AI to answer important questions to a person like, what do I need to check to understand if I'm eligible to a scholarship? What kind of data am I missing? Why am I not actually eligible to this specific income, universal income, for example? And this is the kind of support we can provide to people when we have this kind of knowledge layer, because actually we make this information accessible better to AI, know that AI is much more context and much more understanding of the user situation to provide a reliable answer and really support the end user.

John Furrier

>> Yeah, the reliability is key. Talk about the alternatives. You didn't have knowledge layers. What would be your approach? Take us through what would have been the opportunity or what would be the solution? How would you tackle the data problem if you didn't have a knowledge layer?

>> Actually, it would be pretty difficult. For example, I can think about how our platform used to work two years ago. So before we were using Neo4j and having this kind of citizen relationships. So we had to use very complex joints, basically to understand the situation of specific user and then being able to match this user to relevant information, relevant benefits at the government level.

John Furrier

>> So basically, you basically do SQL queries, basically, right?

>> Exactly, so it was possibly SQL that we are still using, but just on some operational things within the platform, not to represent the full citizen profile.

John Furrier

>> Talk about the impact. What's been some of the results? Okay, so take us through the discovery phase. Okay, you have complex joins, they're error prone too, obviously in their complex rows and tables that are great for certain use cases. But what was some of the impact? What were some of the discoveries? Was it an oh, wow moment? Was it, hey, this is amazing? Did it move the needle for you?

>> Yeah, actually enabled us also to create a system by which we can create deterministic rules much faster. So when we were tackling this problem, we divided it into realms. So on one side, there was the legibility and joints where we decided to have a ai automatically translating regulation into deterministic rules but still human checking these deterministic rules in order you know to provide the clear outcomes without any risk for hallucination and then we have the eye that actually is enabling the user to understand why is eligible or not for something so basically having this kind of graph representation of real data enable us to to create also a system where through AI we can create this a very much faster. And for example, now integrating a new benefit within the platform takes 30 minutes. While in 2025 used to take four to five hours.

John Furrier

>> Wow, major time savings. Talk about the data iterations, because I imagine when you have bureaucracy and these old systems, you've got good data and it's laying around in a system of record. And what was some of the data challenges? You got companies, we see companies all the time, hey, I got Snowflake, I got Databricks, or I have some legacy data. And how do you stack that together? And how did you look at that integration? How did you partner and how does AI fit into that? Because I'd imagine when you start building these knowledge graphs, the AI really connects in once you do that. What's that transition? Because a lot of people are in the same boat right now. They have a vision, they go, okay, I got Databricks or whatever. You got a data lake, or they have a database, they've been doing SQL queries, but now the knowledge layer comes in. What's been some of the data ecosystem things you guys have done? How should people think about that?

>> Well, John, I will give you first a number. Today we have 880 ,000 users in Italy, but each of these users has own household on BonusX. So basically, we have data around 2 .5 million people. We collect data about healthcare condition, financial assets, income, real estate, earnings, because all this data are relevant to really match people with eligible tax benefits from government, even the European Union, and then, you know, all relevant services for them. So understanding the connection between these data is actually key to really provide user -relevant information. We are now moving one step further. So we are partnering with a major banking group in Italy to actually integrate financial data, especially financial transaction, to trigger automatic notification of new benefits. So, for example, if you're using your bank account in Italy to pay for kindergarten benefits, I can actually trigger an eligibility check that is testing if you're eligible or not for a kindergarten benefit, which is a national benefit that provides you to up €3 ,600. So putting together all this data is key, actually, to support people to handle bureaucracy. And today we are leveraging on one side Neo4j, so in terms of graph database to have all for this connection. But on the other side, we are also embedding more and more AI at different layers within the platform from document validation to data extraction, to really supporting user and understanding eligibility and how actually to track all this money and unlock this money for their own families.

John Furrier

>> Talk about the AI piece there, because I think the knowledge layer is clear, great benefits, time savings on your side, great user experience, it can get results faster. But when you think about the AI piece, you have ingestion of a lot of content, right? You are ingesting a pre -existing content. Benefits have policies. They have rules. How are you looking at that piece now that you have the knowledge graph? Because you've got the data connection. You've got the AI kicking in. I'm sure agents are on the horizon. Why not just eat up all that data? And then the reasoning starts kicking in. How are you seeing that play out, the reasoning piece?

>> case? Well, basically, first about the data point, there is something very important in our case, which is data access, no? So we need to make sure that the user can have AI only accessing his own information and not other user information, when basically they are interacting with our AI to get support. On the other side, we have, you know, memory, we have all the knowledge which pays about the regulation, national, local, municipality level. So everything provides AI the right context to really help the user tackle the issue. Bureaucracy is a major friction point in everyday people's life. So from a user perspective, a lot of people, especially from low -income families, today are cut out from the welfare services and financial opportunities because they do not understand which kind of services and opportunities they can access and how to get there. So AI here is providing a major help. But in order to really provide help, we need to not give a generalist answer to the end user. So we need to give very personalized support and move the user one step at a time over the application process to really enable them to access this kind of support.

John Furrier

>> So you're saying they give up because it's a pain in the butt to deal with. It's like, okay, they just abandoned. Like, that's what we see, right? You're solving that problem. So you just shortcut the speed to value.

>> Exactly. So we want to make, and we are making economic and social policies much more efficient because people that today are eligible to a benefit created by new policies are automatically notified by the system on how to access it. And then through this conversational system, they can input the data that are relevant to manage the application and having AI that support them step by step. So in understanding the data, in understanding the documents to upload, in validating these documents, in filling in the modules. So we cover different steps over the overall application process to get this public money.

John Furrier

>> Giovanni, it's great to chat with you. Talk about the situation where people now are learning about graphs and knowledge graphs and all the benefits. It's great to see Neo4j because now that the market wants to see this AI integration but a lot of people are leaning in and people have gone down this road and had to rebuild. They still had errors. The trust wasn't there. What's your message to leaders that are scratching the surface right now with knowledge graphs? What would be your advice? How should they think about what that does for them? Maybe potentially be luminary, but also their organizations. I mean, the power that comes out of this. What's your advice to people who are looking at this just from your personal perspective and what you've learned?

>> So thinking about the BonusX experience, we are really happy of the choice we took. We migrated all this data less than one year ago, so end of 2025. So this is something really fresh in our memory, in the product team memory. And this actually lent us to several advantages in terms of time in integrating new benefits, its reliability, about AI answer, and a lot of functionalities and value that now we can unlock for our user. I may also mention a bit about the migration process, because we still had to migrate data about half a million people at the time. So it was actually a process that we had to cover across different weeks and manage really carefully to avoid the data leak, to avoid data loss. But in the end, I think it was definitely worth it. and now, I don't know, we have much more power and we can create much more value for people.

John Furrier

>> Yeah, it's great to see because, you know, you don't see many standards in data. SQL has been around for a long time. Now you've got the graph query language, you're seeing a lot of momentum. What do you hope for, for new tooling? What's on your to -do list? What's on your wishlist, I guess, that's the question.

>> So if I think, you know, about tech, I'm very used to think about user first. So we have a very product mindset within the team. And I see this technology as something that can enable us to provide more and more value to these hundreds of thousands of people that rely on BonusX. So actually my wish is on one side that AI will keep evolving and be more and more accurate so that we can mitigate more and more the hallucination risk over the future. and secondly that this kind of technology can actually really help us to model reality because if we're able to model reality better and better then we can actually also create a digital context that is more accurate faster and then we can leverage this context this data again to provide value and support people across several things yeah i mean it's such an obvious great thing you're doing people can even get alerts hey you're missing out on some some benefits or

John Furrier

>> hey you left the the process here it's back. I mean, you could do amazing things for people. So congratulations and thanks for coming on.

>> Thank you.

John Furrier

>> Luminary series. Another great example of efficiency. Another great example of user benefit. Again, great work Giovanni. Thanks for coming on.

>> Thank you very much. Thank you for an invite.

John Furrier

>> All right, I'm John Furrier, host of theCUBE. This is AI Luminary Series with Neo4j and leaders who are making it happen as they discover the journey of putting in the AI systems. You got to feed the data, move that data to the AI, and then magic happens. The knowledge graph is becoming center of the conversation. Graphs work, graphs are compatible with AI, and it's all working great, and it's just another great surge in the data revolution. Thanks for watching.