Scaling Data and AI Trust: How Credicorp Built a Cross-Institution Data Product Marketplace

Summary

Credicorp, one of Latin America's largest financial holdings with 50,000 employees and seven financial subsidiaries, built a data product marketplace on Databricks Unity Catalog to enable governed data sharing across institutions that previously operated in complete silos.
The marketplace includes a reliability layer with data and AI observability, legal compliance workflows, hybrid encryption, PII hashing, and system table audit trails to govern cross-company data transfers securely.
Three business cases illustrate the marketplace's impact: a BCP credit score shared with Pacifico contributing $15M in earnings, a Mibanco energy-consumption risk model, and 100 data products delivered to BCP and Yape in days generating $8M in value.

Scaling Data and AI Trust: How Credicorp Built a Cross-Institution Data Product Marketplace

As financial groups scale across regions and subsidiaries, the challenge is no longer collecting data; it’s governing, sharing and trusting it at an enterprise level. Eduardo Torres, CDO at Credicorp and Alvaro Villanueva, Credicorp’s Data Marketplace Product Owner, will share their team’s journey to building a data product marketplace that spans financial institutions and countries using Databricks Unity Catalog. Torres & Villanueva will share how they built a reliability layer at the foundation of their data product marketplace with data and AI observability, enabling multiple subsidiaries to publish and discover trusted data products without costly duplication or security risk. They will dive into the team’s approach to key challenges, including operating models, incentives and change management when subsidiaries are also competitors. Attendees will learn how trusted data across a data product marketplace unlocks a unified customer view, and real business impact at enterprise scale.
Talk By: Alvaro Villanueva Ojeda, Product Owner, Banco de Crédito del Perú ; Eduardo Torres, CDO, Banco de Credito del Peru

Chapters

FAQs

What is Credicorp's data product marketplace and what problem does it solve?

Credicorp's data product marketplace is a governed platform built on Databricks Unity Catalog that allows its financial subsidiaries—including banking, insurance, microfinance, and investment units—to publish and discover trusted data products without costly duplication or security risk. It solves the historical problem of subsidiaries operating in complete data silos with no standardized way to share information across institutional boundaries.

How does Credicorp govern data sharing across competing subsidiaries?

Credicorp implements multiple governance layers: publishing checklists with minimum quality standards, legal and compliance workflows for cross-company data transfer approvals, hybrid encryption and PII hashing for sensitive data, and system table audit trails to track data access. These controls allow subsidiaries that are also competitors to share data with confidence in security and regulatory compliance.

What business results has Credicorp achieved with the data product marketplace?

Three concrete business cases are presented in this video: sharing a BCP credit score with Pacifico insurance contributed to $15M in earnings; a Mibanco energy-consumption risk model enabled better microfinance lending decisions; and delivering 100 data products to BCP and Yape in days generated $8M in value. These outcomes demonstrate that governed data sharing translates directly to financial impact.

What organizational challenges did Credicorp face building the marketplace?

Beyond technical implementation, Credicorp faced challenges around operating models, change management, and incentive structures when subsidiaries that compete with each other also need to cooperate on data sharing. The team addressed these through a carefully designed governance framework and by demonstrating concrete business value to secure executive sponsorship across institutional boundaries.

Full transcript

[00:07] Everyone, thanks for staying for us for one of the late sessions. We basically are going to tell you a little bit about our story and how we're scaling and data usage through Credicorp. Let me introduce ourselves.
[00:19] So, I'm Eduardo Torres, Chief Data Officer for Credicorp and Alberto Villanueva is the product owner for the data marketplace. So, let me tell you a little bit of about Credicorp and who we are. Credicorp is one of the largest
[00:31] financial holdings in Latin America. We have presence in different countries and we have several business units and all line of businesses, right? Universal banking, insurance and pensions, microfinance, investment
[00:45] advisory, asset management, banking basically. And then we are going into neo banking right now in Chile mostly with Tenpo our ventures. As well, we have gone into health care. So, that's something that that we are
[00:57] starting to to venture into. We have over 50,000 employees, over 30 years, listed in New York Stock Exchange, revenue closing on 7 billion USD.
[01:11] Um and basically all of this goes along with the purpose, right? We are in several countries in South America where we want we want to do is we want to contribute to improving lives and accelerate the changes that our
[01:24] countries do need. Um so, that's why we are really focused on financial inclusion. We are bringing more and more customers into the financial services. If you know a little bit about the countries that we are in, there's too
[01:36] much informality and people need to to grow through financial inclusions. But not only that, but also inclusive insurance. We are trying to go into small and micro insurance for people to
[01:49] have better health and better coverages. And basically our economies, and this is pretty important, are really imposed by migrant entrepreneurs. Like small and medium enterprises are the
[02:02] motors of the engines that make our countries grow. Um Okay, so what was the issue? Basically, as you said as you saw there um we were working and we were fairly
[02:15] successful. We have over 135 years, but we had been working in silos, right? So the pension fund didn't work didn't speak with the insurance company. And you will have certain partnerships like the usual, I don't know, auto loans
[02:29] with auto insurance or mortgages with home insurance, but it was really basic. So at some point we started working on standardizing our data
[02:41] architecture. Basically, most of our companies are on Azure and Databricks. And once we had that in place, we created a strategy where we said, "Okay, we want to build a data marketplace." This is um
[02:54] built on Delta Sharing. Alvaro is going to tell us more about our how we build and more technical details. But the idea was that having over Right now it's going to be 2,500 users.
[03:06] And And the good thing for us and and part of the pitch that we made to the to the holding was So it's it's really secure. It's it's a way different from what we used to work
[03:19] before. And also it's And you have seen the announcements these days, right? This is just a starting point because we have been sharing data like tables, data products. We are starting to share and
[03:31] try out Databricks apps. But that's just the beginning. This The next steps we open sharing are going to probably be models, skills, agents, and at some point multi-agents as well.
[03:44] Right? So now I I leave the floor to Alvaro. >> Okay, so first of all, sorry for my exotic Peruvian accent. I'm doing my best with with the English. So, I'm here to talk you about how
[03:57] Credicorp created a data marketplace for all the companies inside their our holding to share data just between them. Now, we of Credicorp, we have done data
[04:10] transfers before. But, we've done file transfers, which is is a technically tedious, time-consuming, not so scalable method of doing that.
[04:23] But, most of our companies already got Databricks. And with that, they have access to this data marketplace feature. Many of you might already know that feature.
[04:35] It's like a platform where users can access and browse for published data sets or data data products. And if they find something interesting for them, they
[04:47] request access. Well, data marketplace is more than that. Now, on the summit, they've been talking about sharing agents and Databricks app. We are at least just by now only sharing data products.
[05:01] So, that's something we wanted for our our holding. Because when you share data in inside data marketplace, you you do that with Delta sharing, which is a much more
[05:14] efficient way than file transferring. The idea was that any company inside each of our businesses could publish its own data, and the other companies could
[05:28] became aware of that. So, any company knew what's inside all of our companies. However, that could cause like a big bang on data publishing, and that is
[05:41] dangerous. Because we, as a corporation, we have to make sure we know what's happening inside the data marketplace. We have to apply data transfer governance, which it it
[05:56] was something new for us. So, how did we do that? First, let's talk about the common data sharing cycle. You know, first, you
[06:08] publish a data product, then people may browse that catalog and look for those data products, and then finally, you give the access to
[06:20] to those users. For publishing, we as Credicorp, we created the guidelines that other companies should follow in order to
[06:32] publish a data product, so they don't publish any any anything without order. For example, the minimum quality requirements, or that data domains shouldn't be mixed in
[06:44] the same data products, and so on. Only if that checklist was fulfilled, then that data product could be published.
[06:56] And when that is was done inside again, our own data marketplace, only for Credicorp to see. Right now, it's not a public data marketplace. Users inside our corporation could could browse and find something that maybe it
[07:10] could enable or enhance a business case. If they find something interesting, then they request for access. But we are different companies again.
[07:23] So, when the access is given, we have to sign a legal contract where we establish clear rules about the consent of the information being transferred, the
[07:36] accountability, the what are you using information on, and so on. There is a strong legal and compliance and regulatory aspect in all these data
[07:48] transfers because again, it's between companies. So, let's talk about more on that part.
[08:02] Each time a data product is going to be published, and each time someone wants to consume a data product, legal and compliance teams must check
[08:14] that request. I know that is tedious. It's really tedious, but we have dedicated legal and compliance teams working on that. These teams are our first checklist, and
[08:28] if they approve that data transfer, only then the data owner, you know, the owner of that data that is being transferred, has the final word if
[08:40] that data is transferred or not. And this is because a data owner must know what are other companies using his data for. And he may have a different opinion,
[08:52] like for example, a business rule or or a business opinion, like, "No, I don't want to share because I don't know, anything." But, okay, once the data owner gives the final approval approval, the access can
[09:06] be given. On the other side, we have security. So, data risk data sharing have some security in in in already implemented,
[09:18] but we took additional measures. First of all, we are different companies. We are not at the same maturity level. So, we as Credicorp, again, we created a
[09:30] checklist. So, each company um has like the minimal rule rules they should accomplish or that they should fulfill in order to
[09:42] be able to receive other companies' data. Because we are as a strong as our weakest link. So, if one company gets information of the other ones and there's something that happens and the
[09:54] information gets out, we're in problem. So, we have to make sure everyone can protect that data at rest. That is one. But in transit, when sensitive information is being transferred, we use
[10:08] hybrid encryption methods and if personal information is being transferred, we use hashing techniques that only are known to us inside Credicorp.
[10:20] Um at last or on the on the last um we think that Databricks has these system tables. We have to keep track of everything that is happening inside data marketplace.
[10:34] Who is sharing with whom, how much, when. So, we create KPIs in order to monitor everything inside data marketplace and we plan accordingly of our findings.
[10:52] Now, let's talk about money or three business cases that were the first ones that we got on on the data marketplace from Credicorp. We have many business cases. Um to date or at the end of last year, we
[11:05] have like 14 business cases more or less. I'm going to talk about three. Let me just drink some water or I'm going to cough.
[11:18] Okay. So, first one, BCP and Pacífico. BCP is the largest bank in Peru. It has a lot of
[11:30] of information, financial information of its customers. And on the other hand, we have Pacifico Seguros, which is an insurance company.
[11:42] BCP has had created its own scoring that uses a financial information of its customers in order to create a credit score. Pacifico, on the other hand, it they
[11:55] >>
[12:08] name that BCP put on that score. And again, legal, compliance, data owner. Do you approve? Yes, we do. Do. Okay. Give access to Pacifico.
[12:21] When that was done, Pacifico used the financial information BCP gave gave to to them to improve the knowledge of its customers
[12:34] with the financial information to enhance it. It enhanced their models of for the payment capabilities, the pricing models, and even included on fraud
[12:46] prevention models. The result was, economically, that Pacifico got with that information earnings like 15 million dollars in earnings before taxes last year thanks
[12:59] to only that information. And as a general, I mean, comment, today Pacifico today Pacifico uh 80% of Pacifico's analytical solutions
[13:11] use information from BCP. So, that's a that's an example of something that BCP already had. They didn't have to to process or to do more complicated things. We just needed
[13:23] to move the information from one company to the other, and Pacifico could do a a lot with with that. Again, we are not talking about areas of the same company. We are talking about
[13:35] different companies. That's why compliance and legal teams are really into this also. Another case, Mibanco. Mibanco it's a
[13:47] small businesses bank in inside Peru. And Mibanco believed that he could use the energy, water, and phone consumption
[14:00] payments in order to estimate the sizing of a small business. Because they thought that the more energy consumption that a company has,
[14:12] it must be because it's bigger. It was a a theory. And So, again, they requested access to BCP. BCP had this information because those payments are done inside BCP for most of
[14:25] of people there. And the information was was transferred to Mibanco. The result was that Mibanco enhanced or enriched its non-customer
[14:39] risk model. And it increased 1 and 1/2 Gini coefficient on discrimination power on his model. I don't have the monetary monetization yet, but it did get enhanced the lead generation for
[14:51] Mibanco. Here we have a different case. Yape it's the digital wallet, the biggest digital wallet inside Peru.
[15:05] BCP and Yape wanted to share information between them both ways, but massively. Well, for what at least we consider is a lot.
[15:18] They wanted to share 100 data products between them. To be honest, we didn't directly use data marketplace because we didn't publish like a 100 data sets. We wanted
[15:30] to do it faster, so we just use the technology the technology
[15:43] like you create a share, you put 100 tables there, there you go. You have access. Of course, again, the local compliance with the owner, that's mandatory. Without Delta Sharing, we could not have
[15:57] done this before. Because before data data marketplace and before Delta Sharing, as I told you, we used file transfers. So, exporting 1 100 file transfers, creating 100 folders,
[16:09] creating the pipelines to Never. No. We really needed like and it wasn't it wasn't an experiment. We Jape and BCP didn't know that that
[16:21] data was going to be useful. They just They were They were just like, "Let me try. Give me the information fast and let's see what I can do." Well, so thanks to Delta Sharing, we
[16:33] could do this. And the result was that Jape with the information that that got from BCP, they increased their loans in 54 million in a year and also in they got
[16:46] recoveries upgrade of 8 million dollars. So, it was not uh not a minor thing.
[17:00] Still, we have some challenges to address. Uh what's happening? The publisher, you know, the one that creates the data products and put them in inside the the data marketplace,
[17:12] has to put effort to do that, but they don't get that benefit from it. The benefit is is is to the consumer. So, why should a producer invest time
[17:24] and money to get that data product at the data marketplace? Well, that's a mindset that we as Credicorp must work on. We are creating some incentive models or
[17:36] KPIs. Uh we have some strategies that still in backlog to motivate these data owners to publish data products.
[17:51] Another thing is we as Credicorp, we are one company. We are another company inside the holding. And we are creating our own data lake. So, we can go get into our own marketplace. For what? We can get
[18:05] information from all the the other subsidiaries, all the companies, and we can create cross-level data products.
[18:17] I I told you before that not all our companies are at the at the same maturity level. That's also a challenge for us. I mean, in in semantic, in context, in data architecture, in data governance, we
[18:31] have different levels, and we have to work with them to reach at least a minimum level so we can work better inside data marketplace.
[18:45] Another thing is still with data marketplace, um we reduce a lot the SLAs, the time to get file transferred. But, still we have more
[18:57] margin to improve, to reduce time to monetization even more. Because on this workflow that I told you, that legal, compliance, data ownership approve, they take their time.
[19:09] So, again, mindset, let's reduce those times, because data can move faster. So, we there's chance to automate automatizate there.
[19:21] Next steps. Uh well, we are looking for other Peruvian partners that want to share share information for us. Is there anyone here? Let me know. Um we want to replicate the Peruvian
[19:35] model data marketplace model in other countries where we have company, like for example, Chile or Colombia. And uh even with our other international
[19:47] companies, we want to try other business cases, cross-border business cases, not to move information in all directions. And well, it's not here, but with what
[19:59] we have learned inside the summit, we can also use, for example, the the Genie share agent and other technologies to enhance that data the data and data sharing inside our
[20:11] marketplace. And that's it. >> Yeah. >> Yep. >> Thank you for your time. >>

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.