Spotify: Building the Core Foundation for Their Financial Data [Integreat 2025]

Transcript:

Introduce Ben Spencer from Spotify, Spotify R&D, specifically. The music the tracks you’ve been hearing are from Spotify playlists.

There are QR codes dotted around the place. So you can actually download the playlist for the day if you heard anything you liked, and that’s all enabled by Spotify technology.

But without further ado, Ben Spencer, welcome on stage. Thank you.

I have the checker done. Lovely.

Thank you. Afternoon, everybody.

Firstly, big thanks for being in this room. I know there’s another great conversation going on next door, so it is much appreciated that you are in this one today.

My name is Ben Spencer. I’m a senior engineer manager working in the financial engineering team within Spotify R&D.

So today, I’m going to be having a conversation with you about how we look to build our core foundations for our financial data. Starting off with some numbers.

Hopefully, you are all one of the ninety six million monthly active users that we have on our app. We serve 184 markets.

So from a financial engineering perspective, there’s a lot of regulations that we have to worry about from invoicing and taxes and other fun things like that. Last year, pounds 15,700,000,000.

0 annual revenue, so lots of 1s and 0s and payments in and out for us to think about. And from a product point of view, we’ve got north now of 100,000,000 songs coming up on 7,000,000 podcasts and upwards of 400,000 audiobooks.

So lots of things to keep you all entertained hopefully. And where do financial engineering fit in? Well, as you’d expect, we are a finance technology set of teams.

We look after fundamentally the money that comes in, all the calculations in the middle, how much we need to pay out from a royalties perspective and then getting those payments out the door and a lot of the other back end financial services like our expenses and procurement and things like that. So where do we start off on our journey with Snap? Well, when I joined the team about three years ago, we had a separate back end engineering team that built a bunch of Java microservices.

That was entirely separate team that we needed to essentially take one SaaS system and plug it into another SaaS system. That then meant that we had long timescales from development, a lot of handoffs and all that other fun stuff that I’m sure you all have experienced.

In addition to that, because these folks were more engineering focused rather than the applications experts that we had administering our SaaS systems, there was then a significant amount of knowledge anytime you wanted anything done from an integration point of view that we had to transfer over. So we went shopping for an iPad solution and ultimately went for Snap with the idea that we would simplify our development process.

We would remove that handoff into the other integration team and actually put the expertise in the hands of the application team, some of which are here today. And we wouldn’t then need to build and maintain our own platform and we would outsource that to an iPaaS provider leveraging AI technologies throughout.

So how have we got on? Pretty well. We have got to the point now where the application engineering team that looks after and administers our SaaS systems are the ones building the integrations.

So we can get something done completely within one of my teams, which is great. We have a bunch of pre built connectors, whether we bought some of those from Snap or we’ve pre built them ourselves and made those reusable and we can plug into our ERP and expense systems and things like that.

Or indeed, we look to reduce our operational overhead. So the team that I mentioned, the back end engineers, are no more.

They’ve sent them off to do other exciting things, and we’ve been able to reduce that team and that headcount entirely from my bottom line, which means that overall, we’ve had a real positive ROI on this. Also, it has increased our speed and agility.

We can move much, much quicker. Very transparently, we’ve also been on a journey about learning how to use an iPaaS.

There was, I think, one colleague within the team that had used it before, and it’s taken us either from a back end engineering perspective or from a SaaS administration perspective time to learn how best to use the tool. But we’re really starting to see some great momentum now to the point where we’re sort of really cranking out multiple of the integrations, rebuilding those in Snap within a single two week sprint.

So we’re really, really getting some momentum there. Really importantly for me and the conversations that I have with our stakeholders, it’s led to much better collaboration with our business stakeholders.

We can turn around a proof of concept really quickly. We can put something in the hands to talk about the art of the possible.

And as we’ve been hearing about sort of today, the exciting new things that we can put in front of them, also the art of whatnot is possible. And we can then demo using SnapLogic.

Integrations can be a little dry when you’re going through lines of code and talking about the logic statements. Actually, then being able to demo that to our business stakeholders sometimes isn’t the most exciting thing.

But with the SnapLogic UI, you’ve seen it on the screen or I’m sure many of you use it today. You can really get them in.

The colors change. It’s pretty interactive.

You can pop up one of the snaps and you can start to demo things much more quickly. So when you’re talking about how you actually do a proper agile process and get something in front of them, it means you don’t have to have the full working product if you can’t get there within your two weeks’ sprint.

A bunch of other benefits, but for us, again, we’re starting to see some emerging less technical colleagues really starting to get their hands on the tool. So in my world, we have three teams.

I’ve got the more technical team that look after the Snap platform, integrate with our GCP platform and other things like that. I’ve then got a functional team, sort of pseudo product managers with system administration skills as well.

And then because we serve all of Spotify for some of our tools, we’ve got an operations team as well. Those folks are more transactional, less technical minded.

But what we’re actually seeing now is, a couple of weeks ago, one of the operations colleagues got a ticket come through, found an issue in integration, went into Snap, looked up what was going on, made a fix himself, tested it and then just came and got the stamp from the technical team. So we’re starting to see people just proactively reducing that barrier to entry and come along and give it a go.

Just been great to see. So a couple of use cases for actually what we do with it.

So this is our exchange rate setup that we’ve got. So we pick up our FX rates from our Google Cloud Platform, GCS, because we obviously want to make sure that all of Spotify are working off those same rates, take through the daily and monthly through to our ERP, NetSuite in this example, and reconcile those back through BlackLine.

So obviously, all of the pink clients here are where we’re using SnapLogic. Another example for you is accruals.

So here we have our procurement system and we actually have got some homegrown AI functionality here where we go and parse out those invoices and propose some of the line structure and we then pass that through post human review to, again, our ERP. This time, it’s more of a technical integration, so we’re using GCS for that reconciliation process.

So where we’ve been, a couple of examples we’ve got, there are dozens and dozens more, but where do we want to go? And that, I guess, is why a lot of you are here today to start thinking about. So from our perspective, more reusable components.

As I said, having these reusable components that somebody that is maybe more technically minded or SnapLogic themselves can go and build actually means that you can lower those barriers to entries even further. You can have somebody that can go and open up Snap and doesn’t need to think about API specs and secrets and keys and all the rest of it, right? That’s often a big barrier to actually somebody with not a great deal of engineering experience coming and starting to use these things.

So super valuable. Let’s go and just take off a shelf the ability to plug into system A and then stick it all together.

So we’re going to keep going down that route. We’ve also actually had some entirely reusable pipelines that the folks have built.

So reconciliations is a common thing that you do. Let’s just go and build a reusable reconciliation pipeline.

So rather than just individual connectors, we’re expanding that out to be more pattern driven reuse as well. That is going to allow us to train more role types of less technical colleagues, open that up, increase the number of people that are able to build our integrations.

Obviously, AI is something that we’re heavily investing in learning and testing and platforms and solutions. So what we’re finding, I think, with some of the sort of changes in technology is the need to think about the use of platforms and solutions a little bit differently.

So with my background, I’m used to thinking of a platform as a technology platform, an integration platform as a service. It’s in the title.

It’s a specific technology that is reused for multiple functions but does the same sort of thing. What we’re finding actually is that we’re needing to spend a little bit of time or opportunity maybe to think about whether we can have more business focused reusability that might use a different range of technologies under the hood.

And as we go into more sort of agentic AI solutions with agents able to pick and choose and think what the right solution might be to solve a particular problem with sort of identifying that there might be some benefit to go after there. So it wouldn’t be a talk today and in 2025 if there wasn’t a title with the word AI in, so I will put that up here.

We’re obviously in Spotify thinking about this as you would imagine. If you use the app, you’ll come across it every day.

There’s a bunch of stuff we’re thinking about and doing, and it’s all super exciting. We’re also very lucky in that as a company, they are investing massively in us as individuals with learning time.

We have some dedicated hack weeks where it’s all about just get your hands on it. You don’t necessarily have to go and do anything specific for the business, but that’s meant that we can all really start to understand the tooling.

There’s a lot of buzzwords. There’s a lot of hype.

But as you’ve heard in the keynotes as well, fundamentally, the building blocks are all things that we know and love having been around the industry for some time. We obviously though are in a finance world, and we need to think about things a little bit more cautiously maybe.

We’re floating on the New York Stock Exchange, full regulations and all of that kind of stuff. That’s obviously super critical.

So we have to have lineage and be able to identify and evidence exactly where all of our financial decisions were made. So it doesn’t really fly to an auditor to say, I asked CHAC EBT and it told me to do this.

So we really need to be able to have that traceability through. The companies that we are obviously using are sort of all the big names you’d imagine and they’re starting to go down that.

You saw in the Claude example earlier, there’s the thinking now that’s popping up in the UI and you’re able to get confidence scores from them. So they’re going sort of on a path where they are realizing that we need to understand a bit more about what’s going on behind the scenes.

But just by their nature, they’re never going to be a logic tree where you’re going to have the full traceability. So something that you really have to consider if you’re in more of a regulated industry and something that I think that from a regulator perspective, they’re going to have to meet us in the middle with what the technology is capable of at some point.

So these are our three phases about what we’re thinking within our world. First one there is simple augmentation.

So making your job a little bit easier by using some kind of AI, Gen AI or an LLM through agents, chatbots and direct integration. So I’ve got an example of some expense report automation that I’ll talk you through in a bit more detail.

But a really quite nice one that we’ve done recently is just some advertising insertion order extraction and we just took a chatbot. We just used OpenAI’s GPTs, but I mean they’ve all got a flavor of it.

And insertion orders are simply a PDF order form with a bunch of different lines that we need to make sure we are serving as adverts within the app. So these PDFs come in subtly different formats.

And what we found actually in looking at the agents that went and reviewed this to ensure that the lines tallied because they were primarily checking from a compliance perspective, they spent the majority of their time just finding the information in the PDFs. So what we did is just really simply go and create our prompt that interrogated the PDF, extracted into a table in a nice, consistent format so that they could actually do the thinking themselves.

We weren’t asking the LLM to really think and do their job for them. It was just a really simple way to structure that data that took us, I mean, not a very long time to build.

Phase two then is human in the loop. But crucially, before taking the action, we get that human review.

So this is the next step in our world from an automation perspective where the LLM would come up with some suggestions. It would infer something, and it would come back like our invoice parsing and say, we think that we should do this, have these line items that we’re updating.

But critically, before any decisions are made, you’ve got that human review that can say, yes, I am comfortable with that. I am fully accountable for that decision and I’m going to click the button before it takes action.

And then Phase three, autonomy with review. In our world, we expect there always will need to be a review post fact, whether that’s full or sampled or periodically, but something to make sure that everything does check and balance.

And I’ve put there, as I mentioned, will we ever get there? And we’re just waiting and seeing how the technology is progressing. Hope so, but let’s see.

As I say, from an AI perspective, I think we all are experiencing quite a fast pace of change. So we aren’t trying to focus on reuse from many of our AI solutions right now.

We’re embracing the fact that there’s exciting new stuff coming out on a daily basis. So we are focusing on the business problems we can solve with the technology of today.

And I think we’re going to wait just that little bit longer before we think about how we’re going to really make all of the AI things reusable. So talking to you about a specific example.

So expense processing, as I say, it’s in the financial engineering space. So it’s some really riveting use cases, but it’s been quite a useful one for us.

So we get sort of around 4,000 expenses a month internally. They go through a manager review, employee submits, manager approves and then it goes to a central compliance team to make sure that it adheres to the policies and things like that.

Challenges with this are that, as we said, we serve many countries. We’ve got people that work all over the world and travel to many places.

Therefore, receipts come in different languages and there’s a translation element to it. Each of the different countries might have nuances to the policies of what is allowed to get paid out in what circumstances.

So again, big overhead if I, as a manager, am expected to understand the local expense policies of all of my team. And then, obviously, this needs to get balanced with our company ethos, which is Spotify, tech company.

It’s exciting place to work and it’s all about trust and speed. Fundamentally, we are a back office team within financial engineering.

I’m very proud of this, but we are a back office team. Sometimes we just need to get out of the way and let people do their jobs right.

So something like expenses just needs to be something without a great deal of friction for our employees. So we built an AI agent that extracts and analyzes the receipts, takes those and does all of the language translation, also infers is this the right kind of thing that this person is buying, does it sound like a restaurant that they’re buying a meal from and provides a contextual recommendation to their manager, who as I say, is the first reviewer, removing hopefully in the future the need for the secondary compliance check.

So we built this using SnapLogic. This was an early version of it.

It essentially takes the expense report data, the business policies and the standard operating procedures and does some Snap magic to get them in the right format and join them together so that we can have a really super simple payload that we ultimately send through to the LLM. So that is all very standard snap technology.

I think as we’re hearing, the fundamentals don’t go away. It is important that you get it in the right format to make it as easy as possible for the LLM so it doesn’t get confused and you get a bit more consistency with the results.

So don’t forget the standard stuff that’s just as critically important. We actually got the support of Snap to do this.

We got some of their professional services folks into the office just for a day and we built this super quick. And it was great just to be able to take away that upskilling, the reading, the trying to understand how it all works just to accelerate, right, let’s get on to the exciting stuff.

Let’s figure out how we can write this prompt, actually get some value out of it. So that was a really good day for us and it sort of buoyed the team and got them excited into using some of this new technology as well.

Ultimately, what it gives us is you probably can’t read this too well, but it’s a Slack message that pops up to the manager and it does an assessment of the, the expense. And in this case, there were some key findings and a violation that wasn’t the attendee list that was submitted along with the team meal that the expense was.

So from there, it took that from the SOP, which is the standard operating procedure of what the compliance team does. It took the policies and it went through and assessed that and inferred it, which was great.

Really simple and within a very short space of time, we did manage to do this. As I say, we’re focusing in this world on building specific chatbots and agents for this kind of process rather than try and necessarily share all of that functionality because the context is very relevant to something like expenses.

But we are obviously using all of the buzzwords you’d expect from a reusable platform perspective as well. So key takeaways for me.

Firstly, is dedicate time to learning. So learn yourself.

I think sort of one of the questions in the other room was around, sort of the world of removing jobs and things like that. To me, my answer to that is similar that was on stage.

It’s you need to get on the bus and you need to learn about it because the jobs will evolve. There will be different jobs in the future.

So you need to go and understand it and learn it yourself. I think to me the barrier to entry has been all of the buzzwords and the hype.

So just getting your hands on it is something that I would really encourage all of you to do and to promote your teams to do, to not get left behind. Leveraging SnapLogic Professional Services, for example, just to accelerate some of the learnings.

They obviously work with a lot of different customers. So what can you glean and share from them and get some advice to get you started? Anticipate the limitations and remember the basics.

And I think as we’ve heard in some of the keynotes, the same problems are still there. For us in our expense example, we needed to make multiple API calls rather than one to get images from one and expense lines from the other.

All of that kind of stuff doesn’t go away. Data still needs to be curated and available in the right kind of format, right? Whether you’re feeding it into an LLM for an MCP in a rag or you’re feeding it to something else.

Fundamentally, it needs to be good and that is the same problem that we have always faced. So those kind of problems aren’t going to go away.

Those kind of roles aren’t going to go away. And also sometimes it’s not the right thing to use AI.

Obviously, that is the hype at the minute, right? But the process you’ve just seen back here, isn’t choosing an agent. It’s just a standard logic tree that takes the data and packages it and puts it with a prompt to an LLM.

We built an agent since then. More to play around and have a little bit of fun.

Right? But fundamentally, we are using in this example the pipelines as the tools that is now your LLM equivalent and packaging that up and send it to the LLM. So again, sometimes you will need an agent and there’s lots of opportunity and it’s very exciting, But sometimes just bolting on some of the new Gen AI technology to what exists today is equally as valuable.

Don’t worry if you can’t get to full automation right now. As I said with our Phase one, two and three, we’re not aiming for full automation right now.

We don’t expect to wave a magic wand and to have all of our processes completely automated with no people because the technology is not there yet, certainly not from a financial engineering perspective. And we’re just not ready to get there.

So from our perspective is don’t be afraid. I don’t know if anybody was in the workshop this morning and there’s lots of talk about ROI and how much benefit we’re getting from it.

We’ve seen benefit from keeping it simple like how can you take a simple chatbot and take part of a business process that is the most cumbersome and just automate that bit. Don’t try and do the whole end to end process.

Don’t try and take somebody’s entire job and try and automate that. Just pick a thing and go and automate where the most time is spent.

And to do that, as we’ve heard this morning, you need to understand your business processes in-depth so that you can actually assess where the right time is to invest. But as I say, don’t worry if you can’t get to full automation right now.

The tools that are coming out will make it easier in the future when you want to and when you’re ready to get there. And then also consider when to strive for reuse.

Think about reuse from a business perspective as well as a technology perspective. We’re getting to the point where I think we’ll be able to do this.

So actually I think this will open up the world for more reuse, but not necessarily more reuse just using AI, maybe having the intersections, we’re sort of seeing some of the things that Snapp are talking about with the MCP servers. That’s essentially using AI to leverage a bunch of stuff that already exists in a non AI world.

So sort of have a think about when it is that is the right time to get that reuse and when you might already have it rather than spending money just reengineering something for the sake of it because it’s a new exciting buzzy term. And don’t platformatize too soon.

May have made that word up. But fundamentally, embrace the learning yourselves.

And with the pace that the tools are moving, don’t worry about trying to make something just yet into too much of a reusable pattern because we expect everything to sort of change in the very short term horizon. Final plug from me is a promotion for Dominic’s podcast on the Spotify platform.

So for any of us that have our app, then please do go and check this out, and you can hear some great things from Dom as the enterprise alchemist. But I’m going to finish up there with a decent amount of time for questions.

So if there’s any thoughts or comments, I’d love to hear from you all. If you disagree with any of it, I’d love to hear that too.

But thank you very much for that and happy to take any questions.

There’s a microphone in the back of the room, so if anyone has questions, please do raise a hand. Oh, there’s a question right there at the back.

Fantastic.

Hi. Thank you, Ben.

I was just curious. I come from working with the financial processes, just pre I just joined SnapLogic, but I was curious about the sort of the trust and risk element.

You’ve mentioned that the regulators are yet to catch up with how fast the technology is going. Right? So and I think for any when you’re trying to automate a process or with Agentic integration do that completely with us.

How do you go about thinking about the trust and risk versus business benefits? Just expand a little bit about your way you’re working.

Yes, absolutely. I was hoping I was going to be out from finance because you’re going to know much more than me about where I work.

But no, so we’re thinking about it maybe from a couple of different elements. There’s the security of the data element and then there’s the auditability of what the process is functionally doing.

So from a security perspective, we need to make sure that sort of we are very careful and aware about how our data is being used to train on. We have very sort of tight contracts we put in place with both our AI providers that we use and also our SaaS providers such that we don’t allow people to train on our data and things like that.

So I think there’s an element now that you just sort of is a bit black and white. You just need to take care of it, make sure that you’ve got contractually you are covered, that you understand what your data and information is being used from.

And then sort of more functionally, which is where we sort of think about day to day, it really is about being able to trace through an evidence to our auditors how a decision was made From a SOX control perspective, which is where we sort of operate most from a regulation point of view, we need to have made sure that we have the approvals in place, which we can still do today. Like I said, with our three phases, the human in the loop one is then key because then you’ve immediately got a bunch of stuff that’s been automated with some clever technology, then you’ve got a human turning around to say, do I agree with it? So immediately, you’re able to then put in your steps that could be approved.

You could have multiple approval steps if you need multiple from a peer review or whatever perspective. So I think using those sort of more tactical take the process, break it apart and just automate a piece of it allows you to be able to put a human in front of an auditor to say, I did this and I understand what was happening before an action was taken that affected our financial data or whatever it is that you’re doing.

So I think just I would encourage you to think about the process smaller, work out where your points are that you do need to make sure you are evidencing control or auditability and wait a little bit for those, very honestly, because I think you’re going to spend a lot of time trying to have conversations to convince the auditors. They’re going to have to get there.

They’re going to have to meet the new technology at some point, but we’re not finding they’re there today. So I would just say that, yes, find the parts in your process that you can automate and leave some of the decision making and the thinking where it is really needed to a human based on where we are right now.

Does that help? Yes. Thank you.

Next question. There’s a microphone coming so that everyone can hear you.

Not sure it’s linked to the control aspect, but just curious as to why you didn’t look to remove approvals from expense expenses which were relatively low risk? It’s an excellent process. And Raff, who’s my engineering manager says you’re going to get asked this because it’s a ridiculous process that we do.

I’m not getting recorded. So I don’t know why we don’t so our ethos of trust and speed is not one we’ve yet brought into the fore for all of our expenses.

It’s something that we’re trying to influence our finance partners into. And we’re the same.

So I mean, for our example, like I’ve got two answers to that question. One, it is madness that we review all of our expenses and we fully appreciate that.

We’re just not there yet. I mean, I’ve worked at a UK bank and we at least had a tolerance on it, right? It’s crazy.

But anyway, so I think definitely there’s something to be considered there. And also, as you say, managers don’t necessarily ever approve, right, ever sort of not approve.

So we might tailor the Slack message to go to the user or it might go to the compliance team or something like that. So it’s just that’s, I guess, a configuration choice when we get to it.

Right now, we’re testing the agent to see if it’s any good. But at the point in time, it would probably make more sense to put it back to the person that submitted it so they can correct it real time because that’s where the value is because like you say, it’s probably just going to go straight through from that stage.

So completely agree. And I knew that catch was coming, and Raff is very happy that he that someone picked up on it.

So thank you. Do you want to ask it and I’ll repeat? Probably easier.

Sorry, I sort of thought of half a good question and then thought I’d make it up because I got it. So in terms of, obviously, you within your team, you’re implementing AI. How are how do you go about getting your data together and making sure that your data is clean, it’s validated, it’s working? Does that sort of fall within your remit as a team? Do you have data engineering team who goes through that? What’s it look like?

A bit of all of that. And yes, we need to think about it.

So right now, honestly, we haven’t got a silver bullet answer. So we do have data engineering teams within financial engineering that take some of our data out.

So our ERP, for example, we’ve got a core data set where we take out of the SaaS system that we use and we put that somewhere that’s then available internally where we don’t have to rely on going out to the third party SaaS solution. That’s all contained internally.

We’ve got that for our procurement team and we’re potentially expanding that. So I think the avenue is we will look to have more curated data within our own control so that we can do more internal sort of leveraging and putting it together.

I think in my team, I’ve got 15 plus SaaS systems, so there’s a benefit in having it internally. If we were less of a sort of a broad landscape of systems, maybe we would be happy relying on getting it from the system.

So I think on one part, yes, we’re putting the data that we need to join from various systems internally, and we’ve got data engineers that look after feed and water and curate that. From an integration perspective with our third party, sometimes that might not make sense.

So we’re exploring where some of our providers have their own MCP offerings. We’re sort of assessing that at the minute.

So you’re finding some SaaS providers will be all open to it and they’re like have an MCP server, bring your own agent, do all of this, that’s great. Some others are saying we’re going to control it ourselves, we’re going to use our own LLMs, you can’t necessarily plug your own in.

So it also depends on what the view is of the systems we use, where it is open and then we can plug into one of those systems and they make it easy for us to do so. Maybe we’ll leave the data there.

But if they don’t and that is harder, then maybe we need to think about a different approach. So right now, honestly, we haven’t got one silver bullet, but we are moving in a direction to have more internally where it makes sense to do so and as the technologies evolve, leverage the systems that give us easy access to it.

Okay, perfect. Thank you.

One last question. There we go.

So you talked about you were successful or at least partially successful in shifting the development to business SMEs. I was wondering, was there anything specifically you did or data that built that confidence that allowed that shift to happen?

Yes. So I probably said the wrong words there.

Not necessarily business SMEs. It was the application administration SMEs.

So in our world, they are people that look after the system. They are still technologists, but they just are technologists for SaaS systems.

So they have the skills that you would expect from an engineer, but they write then the application engineer the application code rather than sort of work in the back end. So they are technologists and therefore do understand technology, which made the jump a little bit easier.

And it was we find that in my world, sometimes looking after a SaaS team, we can be thought of as second class citizens. But it is an art form to be able to use a SaaS tool, right? But we find that very transferable.

As I said, I’ve got about 15 systems that we look after and our engineers can go and pop into one or many and make it work, right? That’s the beauty of them. But it is also an art.

If you go and take a back end engineer, which we’ve tried, they look at these things and they’re like, I don’t know what to do. It doesn’t make sense to me.

So for us, it is an art form. So I don’t think we’re at the point yet of going into the full business SMEs and the citizen developers.

Very honestly, we’ve had some poor experiences. We’ve tried to do that with other tooling.

We went a little bit crazy with some of our RPA tools a few years ago where we opened them up and allowed people to go crazy with building their own bots. And what that was, was we didn’t necessarily put the right wrappers and controls around it and that meant that a lot of work ended up coming back to the central technology team.

So we are being a little bit more cautious when it comes to opening up something like Snap to true business colleagues, trying to contain it right now and becoming less technical. Where if we get there in the future, maybe.

But I quite liked the keynote there where they were talking about more sort of citizen data users rather than citizen developers of the integration. So I think that’s a choice you need to make.

Sometimes it will make sense where you can put the right wrappers and controls, but I think ours is a little bit it would be a little bit too much of a step right now for us to try and put some of these things in the hands of the accountant. So keeping it with the people that do understand the wrapper of technology for now.

All right. That’s a fantastic answer. So thank you once again.

It’s been an excellent session. I hope everyone’s learned from it, something that they can take away.