Modern data collaboration promises to help organisations work together without repeatedly copying, moving or exposing sensitive information. But while the technology has advanced quickly, trust, commercial incentives and internal processes have not always kept pace.

In Episode 6 of The First Mile, Dom Burch speaks with Adrian Bolosan, Head of ISV GTM Strategy for Enterprise and Emerging Business in the Americas at Databricks.

Adrian explains how Databricks is expanding its focus beyond the core platform to the wider ecosystem of data providers, connected technologies and software companies building on Databricks. The ambition is to make “Databricks plus partner” a force multiplier for customers, helping them solve specific business problems without assembling everything themselves.

The conversation explores how OpenSharing and clean rooms allow organisations to collaborate without surrendering control of their underlying data. Adrian shares examples from identity resolution, supply-chain data and retail media, including one retailer and CPG collaboration that took almost a year to complete. The technology worked, but lengthy approvals, misaligned incentives and excessive aggregation left the resulting insights too diluted to be useful.

Dom and Adrian also discuss the less glamorous middle of data collaboration: legal agreements, procurement, security reviews, data engineering and seller incentives. In many cases, the biggest obstacle is no longer technical. It is persuading people and commercial models to move away from established ways of working.

The final part of the episode turns to AI. Adrian argues that organisations cannot jump straight to sophisticated agents without first addressing the quality, identity, consent and governance of their underlying data. They also discuss real agentic media-buying pilots between publishers and agency holding companies, where agents are already buying media, receiving signals and deciding what to do next.

In this episode

• What data collaboration means in practical terms
• Why zero-copy sharing is replacing legacy files, APIs and duplicated pipelines
• How clean rooms support identity resolution without exposing raw customer data
• Why a technically successful retail-media collaboration still failed to create value
• Where Databricks chooses to build and where it relies on specialist partners
• What Databricks looks for in a Built-On partner
• How commercial incentives can slow adoption of better technology
• Whether clean rooms reduce costs or create another cost centre
• Why AI adoption depends on reliable and governed source data
• How MCP servers are connecting agents to external data and tools
• What is already happening in agentic media buying
• How businesses should identify the right place to begin

Chapters

00:00 Introduction
02:01 Adrian’s expanded role and the Databricks partner ecosystem
03:16 What has changed in data collaboration
04:47 Identity resolution and zero-copy sharing
06:37 A retail-media collaboration that underdelivered
09:28 Why trust still comes first
10:09 Where Databricks stops building and partners begin
12:35 The Partner Well-Architected Framework
15:44 Legal, procurement and the boring middle
18:22 Who pays for data collaboration?
20:21 Why poor data undermines AI
23:13 MCP adoption and AI-ready data
24:03 Agentic media buying across company boundaries
26:18 Governing what agents can access and do
27:50 Adrian’s advice for getting started

About Adrian Bolosan

Adrian Bolosan leads ISV go-to-market strategy for Databricks’ Enterprise and Emerging Business across the Americas. His work spans data collaboration, partner adoption and the ecosystem of specialist companies that connect to or build their products on Databricks.

Before taking on his expanded role, Adrian helped develop Databricks’ go-to-market approach around data collaboration, including OpenSharing and clean rooms.

Transcript:

Welcome And Adrian’s New Role
Speaker 0:07

Welcome back to the first Mile Podcast with me, Dom Burch. This is the podcast where we explore the beginning of the digital journey, the point where every click, conversation, and interaction first take shape, and where the quality, identity, and governance of customer data are determined before it flows down into that big black box of Martec. Later down the line, I'm delighted this week to welcome Adrian Bolosan from Databricks to the podcast. Now, Databricks is the governed data and AI environment where a company brings its information together and puts it to work. The timing is particularly good because Adrian has just announced an expanded role leading the ISB go-to-market strategy for Databricks enterprise and emerging businesses across the Americas. Now, data collaboration remains part of his remit, but his focus also now extends across the wider ecosystem of companies that connect and build on Databricks. So while his job is increasingly about the specialist software partners that help customers do more with that environment, the ambition is to turn that ecosystem into a force multiplier for customers. So today we're going to explore what that means in practice, whether it's already happening, and what gets in the way between two organizations agreeing to collaborate and any useful value appearing at the other end. Right, let's dive in. Adrian, welcome to the First Mile podcast.

Speaker 1 1:29

Hey Dom, thank you so much for having me.

Speaker 1:31

Do you know it's lovely that you're in sunny LA and I can tell you I'm in sunny Bradford, Yorkshire, which is not a phrase that often rolls off the end of my tongue. But anyway, great to see you. Um before we get into it, tell us a little bit about the new role. What's changed, and what does the expanded remit tell us about where Databricks is heading?

Speaker 1 1:50

So just on my two-year anniversary is actually where this role is changing. I was originally brought on to Databricks to help launch our kind of go-to-market around data collaboration, which I know we're going to touch on quite a bit. Data collaboration is a very broad term, but through the lens of Databricks, that's using patterns like delta sharing, which is now open sharing, cleaner rooms. But really, what it is at the end of the day, Dom, is just connectivity between our partners and our customers to power use cases, right? In two years, we've our teams have done an amazing job of kind of scaling our ecosystem. So talking about where things are going, I think of our partner ecosystem as healthy and still growing, but how do we get adoption in the field, right? We have two to three thousand sellers out there, Don, and their primary focus up to today has been like selling Databricks. And when we view the growth of the business, it's I think of it as Databricks Plus partner now solves for these use cases. So it's very important to get adoption in the field. And that really encompasses those three patterns you exactly said. It's not only like the data providers that are using open sharing, it's the connected partners and also partners that are building on Databricks.

Speaker 3:00

And what can organizations achieve today that maybe like if we sort of roll back, and five years is a long time in this world, right? But let before you'd even join Databricks, like there was such a time. Um, what can people genuinely do today that perhaps they weren't able to achieve a few years ago?

Speaker 1 3:16

Yeah, I think in the simplest form of just everyone can see my background online. I I come from uh my lineage is on the data side. I worked for one of the largest data providers in the world, Experium. And it was Experium Marketing Services specifically, it was a business unit. And like you said, five years ago, it wasn't that long ago. And some of these practices still exist. So apologies for those that are still dealing with the old ways and pains of that. But data movement is part of like if we just pick the ad tech bar tech ecosystem, data has to move all day long for multiple parties. And I think if you think of legacy ways of APIs, which are still leverage today in many use cases, but FTP data, let's just use that as an example. Copying and moving data to data place to data place, there's it's riddled with issues, right? And I think the biggest change today is just these modern ways of sharing data, zero copy sharing is what a lot of organizations say. As simple as that sounds today, Burse, what you just said five years ago, it's leaps and bounds as far as data security, freshness of data. I mean, all the all the governance benefits. And I think that's the single biggest change. And when I think about data collaboration.

Speaker 4:28

Now,

Zero Copy Sharing Replaces Data Movement
Speaker 4:29

I'm gonna put you on the spot a little bit, and obviously you don't have to mention names, but could you walk us or talk us through a collaboration that's worked particularly well? Like what was it that the parties were trying to achieve? What did they build? And then, you know, everybody wants to be able to measure the value, right? What was the bit at the end of it that made it, you know, sing and everyone went, yeah, that that worked, right?

Speaker 1 4:49

Yeah, I think, and I'm happy to share some names of those that are public that have done a good job with press releases of some of our partners, but collaboration within this space DOM has been predominantly driven by the industry leaders in identity resolution. And and the reason for that is there's always been friction of hey, I can resolve all these IDs and give you better visibility, you know, or clarity into who your customers are. But it requires you, the old ways five years ago, you have to send all your most sensitive data to me. And so I think really what's changed to the power of like this zero copy sharing, coupled with, you know, I'll say the word that Shallvin Havey said, but I'll say it anyway, is you know, clean rooms, regardless of what tech is out there, right? There's a lot of great options out there, that is really solved for the friction. Like I need the service and solution, but instead of having to go through uh kind of the red tape required to send all this sensitive personal information out of my four walls, what's really changed is your uh organizations are able to do that without trading off the governance of that data, meaning that PII no longer has to leave their walls. Partners like you know, Astra, Axiom, LiveRamp, uh Epsilon, all these folks have really adopted this modern way to alleviate that friction with our shared customers that need their services, but all that data is governed in Databricks.

Speaker 6:14

And in reverse, and again, this is perhaps one you go, Dom, I don't want to go there. It always works. Everything's brilliant all the time, right? But is there an opposite? Is there one where a collaboration it looked compelling on paper, but failed or underdelivered?

Speaker 1 6:28

Yeah.

Speaker 6:29

I mean,

Identity Clean Rooms That Actually Work
Speaker 6:30

we've only just met and I'm putting you on the spot. So I apologize, Adrian.

Speaker 1 6:33

No, no, no apologies needed. I mean, this is the truth, it's not everything. No one's ever batting a thousand, right? So I think there without say names, there was a particular retail media uh data collaboration. Um, very familiar. I think a lot of folks know retail media is having continues to have a lot of scale and a lot of growth. So we're very happy to see these collaborations. But with this one particular, it was a retailer, major retailer, major CPG brand. And the goal is pretty simple. We want to be able to measure our digital ad campaign return on ad spend by seeing the purchase data given from this uh retailer. So the technology existed. We've proven that. It's not like this was the first two parties to go through this motion. What actually went sideways was not necessarily the technology, but just misalignment and the incentives and the red tape I kind of talked about earlier. Um to get the data collaboration agreement, for those that are familiar, like you have the technology that's proven, but all parties have to essentially sign off on this data collaboration agreement, that took roughly like seven to eight months. That's before a single question or overlap was actually done. And then to make matters actually even worse from there, even though the technology proved that no PII was ever going to be exposed, um, that you know, differential privacy was used, and then very egregious aggregation was used. By the end, the insights were so diluted that it was really hard to like decipher the value of everything. So it was really time sunk, right? This whole project took about a year. Um I think both of these parties will come around. That was like, hey, let's see their first, their first go at it. But that was an example of like, hey, the technology's proven, but the motion as far as the people and the cross teams behind it wasn't modernized enough yet. And that caused to almost have like the same issues these companies have been having for a long time, and that was lack of insight.

Speaker 8:33

Yeah, and you know, building up those is trust, right? And at the end of the day, you aren't dealing with big corporations, and and you've obviously you've got a load of compliance and things and hoops you have to jump through, right? For all good reason. But also you're dealing with sets of humans who need to build up that level of trust that it's gonna work and that they can give something away and they're not actually gonna be breaking any rules and all that kind of stuff. That was the first iteration, right? And it can't always go to plan. So, as you say, maybe maybe version two might sing a little bit quicker and uh and not take eight months to get over the line.

Speaker 1 9:02

Well, I like what you said about trust. The reality is organizations that trust each other have been sharing PII in the clear in the raw for years. So, like, that's to your point, is probably the best trust. The point is to use technology where maybe trust isn't a hundred percent. But what you're basically pointing out, which I agree with, is like don't do business with people you don't trust with the start, and then that's your baseline. Then you basically use, you know, privacy enhancing tech as like a zero trust. Plus, not to mention there's the data collaboration agreement that protects both parties as well. So there's many layers. But I think to your point, the the key, which I actually never I always forget to note, is like it's all based on trust from the beginning.

When Collaboration Fails In Retail Media
Speaker 9:42

Now, uh there's an interesting line somewhere within your business, right? Where you have to choose what is core to Databricks. And obviously you've developed from where you started out, right? And now AI is such a fundamental part. But where do you actively choose not to build, right? And where does that then make sense in your role for partners, for the ecosystem to supplement, to sit on top, to enhance, you know, to plug in all those kind of adjectives that allow you to kind of go, right? Here's where we work, and here's where we want others to come and work with us.

Speaker 1 10:14

That's a great question. So, and this is ever evolving, right? Our our product teams across all of the different aspects of our platform, Dom, I'm sure, have different answers based on where they're eagerly looking to grow and expand the platform. I kind of start with it's not very sexy at all, but I like to start with the data governance point of view. So, like if it takes in very sensitive data and needs to be governed, and that could be not just data like PII, maybe there's particular models that need to be governed. I think a lot of that stuff gravitates towards the platform DOM. And like what Databricks is really good, you kind of said at the top of this was the governance layer. Like that is with you know, Databricks has proven we're really strong there. When you get more into these point systems, you know, that a non-technical user is using, that's traditionally where Databricks is really leaned on partners, right? Um, in the ad tech Martech space, that would be like built on partners in the customer development platforms, right? Um, so you have big partners um like an Epsilon or even like a high touch, uh Amparity, these are all publicly known as Databricks partners that have really owned that UI layer aimed at a marketer, right? That marketers weren't traditionally in Databricks. Is that changing? Yeah, it is, right? So it's it's known Databricks launched uh the first agency CDP in Databricks called Customer Lake, but it's early for us, right? Like we're that'll be our first foray into a UI that's really aimed for a business user. And arguably a lot of our partners, they've been there for years doing it. And so that's kind of the separation, Dom. I think about it to the lens of like data governance, where does the data need to live? And then like who's the end user? Is that a traditional Databricks user today? If it's not, then that's really where the ecosystem, I think, really shines.

Speaker 12:03

Now,

What Databricks Builds Versus Partners
Speaker 12:04

Databricks describes its partner, well-architected framework, right? That's the kind of bar for a validated integration. And I guess the result is then feeding directly into a partner's tier placement, as you just described.

Speaker 1 12:16

Yeah.

Speaker 12:16

What does a well-architected review actually examine? Like for those who are listening in and going, like, what does that look like? What's the practicalities of that for me as I come to Databricks with my thing?

Speaker 1 12:26

The simplest way for me to describe like what built on is, and it depends on obviously what your product is, or it's not a particulars, Don, but like at the end of the day, we're we're really looking at customer-facing B2B products. And then foundationally, underneath that, is that Databricks that's running either the compute or the AI engine behind it, behind their application or their API or their IP. So we don't necessarily need to know their IP, but foundationally, key components of that engine are actually built on Databricks. That is that is kind of differentiator than what we're looking for, Don. If you add, you can add a lot of our biggest built-on partners, they are also our biggest data collaboration partners. So you could have a built-on partner that's also leveraging open sharing to then distribute data outwards. Epsilon, for example, is a great example. They're they're a built-on partner. They have particular products that are literally built and running on the Databricks engine. They're also a big driver, again, going back to what you guys are saying earlier. Data needs to flow back and forth between our shared partners. They're they've really leaned in on originally on delta sharing, now known as uh open sharing.

Speaker 13:35

I guess a technically excellent partner could still not quite make the cut if their commercial or support model isn't mature enough. Because there's a lot of great innovation coming in, right? But you need to, you kind of need all of those elements to be in play, right?

Speaker 1 13:48

That's correct. And it's funny you bring that up. I'll I'll get brought into particular customer calls, usually by the AE saying, hey, like my partner's building a lot of stuff on Databricks. You know, we've had some conversation about partnership, but like can you come in with the partner engineering team and kind of scope out if this is a good fit? And to your point earlier, like they're maybe just you know 10, 20% uh shy of what we would view as well architected framework. And to be honest, they're totally fine with that. Like they need to, some of these customers aren't even aware of the partnership program, nor do they are they asking for Databricks to go to market with them. And so, like, I kind of view that as like that gray area. And maybe there comes a point where they're like, hey, now we really want to leverage on go to market together. Maybe we're able to like shore up those gaps in the well-architected framework. But I mean, the well-architected framework is wasn't designed for every customer to be a partner, right? And so it was really designed for those, again, B2B facing, that are looking for that additional um partnership component. Not everybody needs that. There's some very successful software companies that are building bits and pieces on Databricks, and they don't necessarily need to lead into the partnership.

Speaker 15:03

Yeah, no, I think that's helpful. That's that's really helpful for people listening to understand that. Um,

Well Architected Framework In Plain English
Speaker 15:08

now between the we've agreed to collaborate and the customers getting value sits, like this is the boring bit, but it's important. It's the legal procurement, security view, the data engineering, the operating model design, right? What happens during that period? Like, how long should a well-run collaboration take to move from agreement into production?

Speaker 1 15:25

That's a great question. I think the the legal components and things like that you said are are, I feel like those are smoothing out and moving much faster today, thankfully. Five years ago, three years ago, even two years ago, not so much. But I feel like that's become the norm, averaging some of this tech now, Don. I think the biggest thing that's holding up, you know, going to production is not even the production piece, it's actually the adoption of customer one to, let's say, to 15. And that's because sellers of these partners have sold the old way and now they have to go sell this new way. And that, you know, there's some friction there. I think sometimes also the commercial alignments isn't always hasn't caught up to the new modern technology, meaning I've seen patterns where certain sellers, Dom, are actually incentivized a certain way to sell it the old way versus the new way, right? And and if those aren't aligned, you know, in the best interest of that seller and and them taking care of their family, they're they're going to gravitate to what puts the most in their pockets. And I don't fault them for that at all. That I think is like technology moves at this rate, and then the human element and the commercial element, I always feel like it's a little bit of a lagging indicator. This too is also fixing itself right over time. But even two years ago, Don, that was the biggest thing. I would go, I'd be on these joint calls, and the reps like, hey, I agree. It seems like it's a better way, but like I'm incentivized actually more to do it this way. And that was kind of like a light bulb of like, oh, the commercial model hasn't picked up yes.

Speaker 17:05

Well, I'm glad that's shifting. I was gonna ask you like, what's the hardest part? Is it the data, the contract, or the people? But it sounds like that people element is like often overlooked, right? Because especially when it's smart people thinking about engineering and thinking about use cases. And as soon as you realize that some new bit of tech is gonna work, it's like you can't help but future vision, like, wow, the world's gonna be so much better as a result. And then you hit a roadblock and it's like, what was the roadblock? And it was like, that guy or that girl wasn't gonna make as much money this month. So they put the brakes on. Totally. Yeah, yeah, yeah.

The Real Bottleneck Is Sales Incentives
Speaker 17:37

Okay, so let's move on, right? Who normally pays for a clean room or data collaboration? Like, and and sometimes I guess it this is a bit provocative, but is collaboration sometimes a cost center dressed up as a strategy?

Speaker 1 17:50

That's a good question. So I think there's certain levels of collaboration, and your most basic is just replacing legacy pipelines like you and I talked about in the beginning, and using more modern ways. In those cases, Dom, it's it's actually neither party is actually spending a lot of money. It's actually saving them money. You're saving money from ingress costs, uh, transformation costs, um, et cetera. And basically, you know, a very, I'll give you a great example. A very large retail, um, another grocer provides all their, they package up and productize all their supply chain data. So, like, what are the items on on the shelf? And they provide it to their CPGs. They literally went from like zero to hundreds of shares because this was just the best way to kind of share that data out. These were already their existing customers. They're actually just providing a better way for them to get that data to get them almost faster value to insight and actually land the data where those customers are gonna use the data anyway, which in a lot of cases is Databricks because that's where data scientists like to play, right? So I think it's it's in that case, it's not a cost center. I have seen though, like to your point, there's always the other side of the coin, where organizations have used this as like, oh, let's now use this as a new commercial trigger, and we're gonna charge you, you know, percentage more than the traditional way. But that honestly has, I think that was early days, Dom, and and they proved that that didn't really work. They were getting pushback from their biggest customers that are they're smart. And so they're like, hey, I know the technology, I know this actually doesn't cost you more. Why, why are you charging me more, right? For the convenience. And so we see less of that today.

Who Pays For Collaboration And Clean Rooms
Speaker 19:32

I want to come on to AI in a minute, right? And we'll we'll sort of close off there. And we can't have a, you know, nobody can talk about tech without mentioning AI, right? The world is um, but what happens when the underlying data is wrong, right? What does bad data look like by the time it reaches the clean room? And how do you find out that it was bad, right? Because, you know, if we're gonna be building all these really smart AI systems, like, you know, not very good stuff in means not very good stuff out, right?

Speaker 1 19:58

You bring up such a good point. Like organizations that are trying to jump to being like super AI focused, it's all the foundational pieces, Dom, that are like I always say, like, you can't jump to like a rocket if you haven't figured out how to fly the plane uh or built the runway, so to speak. Because really what we're talking about is, you know, these foundational pieces and their data. Having all that almost AI ready, Dom is probably the most important part we've seen. And that quickly shows up in, you know, it's not even AI slob, I don't even know what to call, but it's like the old adage of like garbage in, garbage out, where they're just not getting the value out of these models because their foundational data estate isn't really there yet, right? And so, like, I think that's like a constant battle for organizations. No CTO or CIO is gonna tell you our data is a our data state's a hundred percent in line. Like it's a there's new data coming in all the time that could be first party data, second party, third party, joint party data. And so you it's an endless pursuit of maintaining this data state. And I think those organizations, Don, that have their data state in a better state are the they're the ones that are farthest in the AI game, right? Because they didn't have to like spend so much time on the foundational P because they've already invested there. But it's it's clear as day when you see who's adopted. Like I get a clear view of like which organizations are are adopting AI in their business, and and it's those two are too strongly correlated.

Speaker 21:31

Yeah, I mean, we see it a lot, right? If you're particularly if your data's missing that kind of consent, you've got duplicated events, maybe inconsistent identity, all of that stuff is kind of going into the big box. And then you're trying to like run sophisticated AI on top of that, you're sort of building on problem data, right? And it's not very sexy sorting out that stuff. And often people don't know how to sort that stuff out because they're blending so many different data sets, they're not able to do it in real time, but they're We're under pressure to solve things in real time. So they're like the temptation then to go apply a load of like clever AI on top of it because we've got to get to the answer quickly rather than going, we need to put our foot on the ball and we need to resolve some of those fundamental things. And if we don't do them now, we're kind of just pushing the problem into the future, right?

Speaker 1 22:20

Yeah. Actually, to your point, some of the biggest data providers out there started to surface in our marketplace their MCPs, right? They're and we are now an MCP marketplace. And if you look at the adoption DOM of those that are using their MCPs, it's not net new customers per se. They're actually those existing customers of those data providers that are the most advanced and ready to just plug in an MCP into their agents to, you know, combine with their internal data to do really cool things. And like that has been, it's actually funny you bring that up. Like when you start to look at the adoption of customers using our MCD marketplace, they're the they have their data state in place. They're far advanced in the AI game, and they're now leveraging their data providers in the most modern way through an MCP server. And it's night and day on like who's the most sophisticated there.

Speaker 23:09

Okay, we're

AI Readiness Starts With Data Quality
Speaker 23:10

allowed to go there then, right? We've resisted. We're 24 minutes into a podcast and we've not really talked about agents and all that kind of stuff. But has anyone actually run an AI agent across an organizational boundary yet? And what did the agent do? And and maybe what broke?

Speaker 1 23:26

Yeah. You mean by boundary, you mean externally, right? I want to make sure I understand the question. Okay. Yeah, I mean, there's yeah, so specifically within the ad tech bar tech space, there's this, you know, agentic is one thing. You say agentic and then add, but I think agentic media buying is probably where we see real activity happening. So these are real pilots between major publishers and holdcos, uh Dom. And it's happening, right? Like you hear, like we have ad week coming up and stuff like that. You're gonna hear a lot of that. I've actually seen that in production. Again, these are pilots though, like, but they're actually buying, media buying. Um, and then signals coming back and it's making decisions on what it should buy next. Um, but it's still early days. So, like for me to say that a lot of these things aren't breaking, like it is breaking because it's these are these folks are are paying the past of this new agency media buying. But I think six, eight months from now, we're gonna see a lot more production use cases of this. Um, and it's pretty exciting, right? Like, I think it's you know, a lot of this in a way, I'll always give credit to the walled gardens, right? Like, in my view, every major media company, especially big TV media companies, want to run their ad business very similar to what like a meta does, right? But that's uh much harder thing to do than than say, because those guys were digitally native, right? Like a meta, and they built, uh like we said, like this black box. They have arguably one of the blessed the best black boxes out there. You tell what KPI is gonna do, and it has all this uh great tech behind it to hit these KPIs. It's not really like that if you think of it like the traditional TV world, but it's going that direction. I think when we see these agency media buying, it's getting much closer to like the emergence of like uh the walled garden media buying and what traditional media buying and TV is.

Speaker 25:21

I mean, it's a little bit of a you know compliancey question, right? And I know we're getting towards the end, but you know, bear bear with me, lovely listener. When an agent can query, infer, and act across those boundaries, right? What permissions and safeguards become non-negotiable?

Speaker 1 25:36

It's a great question. So a lot of that is determined by like the architects on both sides, and a lot of that is it can be 100% governed, not to get techie, but Unity Catalog and Databricks can determine, regardless of what agents you're running DOM, like what data is always off limits no matter what, right? Or is there just aggregation limits based on like insights that you need to glean? But it actually goes to the people problem we said in the very beginning, is aligning on the incentives of what both parties are trying to get out of it, and then what data both folks are bringing to the table for that collaboration. And when I say that, that's basically then allowing, hey, what what can either agent read of each of each other? And I think part of that going back to what's breaking, that is what's breaking right now. It's just like, oh, it didn't have access to this, but it needed it. Okay, well, like at what level does it need that? For how long does it need that? And these are like the the growing pains of this uh agency era we're in. Um, again, a lot of this stuff was done by by people before, and now it's you know done by machines and and governance. To your point, everyone's being, I would say, safely cautious on what it has access to and what it doesn't have access to. And that's a little bit of a learning curve.

Agents Across Boundaries Need Guardrails
Speaker 26:51

Okay, this is my final question, he says, having already asked you kind of his final question, but this is the last one. If you were advising a business leader who wanted to begin a meaningful data collaboration like next week, right? What are the first maybe one, two, three things you'd tell them to do? And what's the one question they should ask a potential platform or partner that almost nobody asks today?

Speaker 1 27:14

Yeah, that's a great, that's a tough one.

Speaker 27:16

That is a tough one. It's quite a long question as well. So if you need me to go back to the first bit, that's fine.

Speaker 1 27:20

No, no, no, I'm I'm good. I what I would do is I would start with like looking at the current process today and how long it takes you from you know start to finish, right, on any particular task and see where all the friction points are at in there, and then start to identify in that friction point, Don, just to be clear, it could be internal friction for data access they need. It could be external friction because they're needing data from a third-party partner. And then you start to like quickly identify these friction points and see, hey, where can technology not replace people, help those same teams, right? Alleviate all that friction, right? And I think, you know, because the use cases aren't really changing, right? Like when I looked across AdTech and Martech, the use cases are kind of the same, right? It's it's basically how we're accomplishing that is different. A good way to think about it is as a media buyer, you know, submitting a ticket to the data science team to get a question, right? They literally have almost like a data scientist at their fingertips now, depending on what platform you're using. To be that data scientist, they need to know what questions to ask, right? And then how to use that information. They were always doing that, but that took like three or four steps, maybe days, maybe weeks. Now they can do it in minutes. And so, like that to me, Dom, is like the areas you need to identify. And that's like a game of inches. That then over time, I think, involves into like, oh, well, maybe this should be an agentic area of focus for us, right? So, like that's kind of how I see it. And it may sound basic, but I think that's that's where these organizations need to start start, right? It's a little bit basic where the biggest friction points are to move faster. That then starts to highlight other areas you can really invest in technology to make it a more of an agentic flow.

Speaker 29:03

Love it. Love it, love it, love it. Well, I we've covered loads of ground, by the way. It's been fantastic.

Speaker 1 29:07

Yeah. I'm excited, right? Like I think, you know, outside of the doom and gloom of these foundational models and all this stuff, you know, AdTech Mark Tech has always been at the tip of the sphere for leveraging technology. We've been very, I think, beneficial in seeing that. So I'm excited to see what's happening, right? Like, and it's it's moving pretty quickly. So I have nothing to add outside of the fact that I'm very excited to see where this new technology is going to take, you know, some of the biggest publishing companies, holding codes, brands. Brands are also like able to do a lot of their own stuff internally, which a lot of that you know technology wasn't available. So it's it's very exciting just to see the space changing, Dom.

Speaker 29:45

I think that's right. It feels like we're really going through an era of democratizing the accessibility to do really clever stuff that you have to be a scientist at before, right? And that be almost like your PhD. And now it's kind of like, you know, numpties like me get to play around with some of this stuff as well, right? Which is good fun. Adrian, it's been an absolute pleasure. Thank you so much for your time this afternoon coming on to onto the first mile podcast. It's been a delight catching up. And um and thank you for making it so interesting and easy for me to understand because I appreciate it.

Speaker 1 30:16

Cheers, thank you so much, Dan, for having me. This was this was really fun. Thank you so much.

First Steps And Closing Thoughts
Speaker 30:22

So there you go. You've been listening to the first mile podcast from MetaRouter, and that was Adrian Bonastan, who is from Databricks. If you're interested in finding out more, please go to the show notes where we will have all the links and all the access points that you need to find out more about Databricks, to find out more about Adrian. And if you're that way inclined, you can find out more about MetaRouter too. But for the time being, you've been listening to Middle Dombirch on the First Mile podcast.

‍