The Hidden Costs of AI, Cloud and Kubernetes

Episode 6 • November 19, 2026 • 00:18:48
The Hidden Costs of AI, Cloud and Kubernetes
Her Head's in the Cloud
The Hidden Costs of AI, Cloud and Kubernetes

Nov 19 2026 | 00:18:48

/

Show Notes

Who really owns cloud cost optimisation? And as AI adoption accelerates, how can organisations keep track of what they're actually spending?

Cloud environments are becoming increasingly complex, and managing costs is no longer just a job for the FinOps team. So how can organisations build a culture of shared responsibility while balancing innovation, performance and efficiency?

In the Season 2 finale of Her Head’s in the Cloud, host Egle Scerbaviciute is joined by Donna-Jane Acornley, Senior Platform Manager at Coles, to explore the realities of cloud cost optimisation, FinOps maturity and the challenges emerging as organisations embrace AI.

Drawing on her experience spanning gaming, product management and platform engineering, Donna-Jane shares why effective cloud optimisation requires more than just the right tools. It demands collaboration between engineering teams, finance, architecture and leadership.

Together, they explore:

Whether you're working in DevOps, Cloud, Platform Engineering or technology leadership, this conversation offers practical perspectives on managing cloud costs while continuing to innovate.

Tune in to the final episode of Season 2 of Her Head’s in the Cloud, with more conversations coming in Season 3.

View Full Transcript

Episode Transcript

[00:00:00] Speaker A: Foreign. Welcome back to her Heads in the Cloud. This is our final episode of season two and we've already got some fantastic guests lined up for season three. But today I'm super excited to have Donna Jane join me. She's currently a senior Platform manager at Kohl's, but I'll let you do a bit more of a deep dive and introduce yourself and DJ [00:00:28] Speaker B: hi everyone, I'm Donna Jane. I'm currently the Senior Platform Manager at Coles, looking after the cloud space. Within our domain sits an internal developer platform, a bespoke landing zone that we've got created all the DevOps tooling, Kubernetes, SME Cloud Ops space and finally our FinOps function also came out of the space too. I've been at Coles for five years now. I have a very strong background in gaming, but that that old life seems to be very far away now as I'm older and older every year. Not native to Australia and I'm South African, but have lived here for a very long period of time of my life [00:01:08] Speaker A: and have managed to keep the South African accent [00:01:11] Speaker B: a little bit. Comes out better with wine, but yes. [00:01:15] Speaker A: Awesome. And what's your, what's your story, dj? How did you end up being a senior Platform manager at Kohl's? What's the story behind that? [00:01:25] Speaker B: So I've had a very interesting career over my life. I actually started off in gaming as I alluded to earlier, came straight out of university in Europe with that brought me over to Australia in the gaming industry, done B2B, B2C being on the supplier side as well, done hardware and software. So I got to a point in my life through my gaming industry that I thought I'm a mom now, I should probably do something in a more suitable sort of industry than gaming for my little girl and decided to pivot out of that about seven years ago. That took me into working with Census for a little bit, went out and commoditized their data as they were looking to sell themselves and from there decided to take a pretty daring step, I thought, and moving out of sort of that higher leadership level back into product management, going into the core of what I really loved to do. And that's how I landed here at Coles. I took up a role as a technical product manager to come in and help with the platform engineering product capability setup for infrastructure teams. That role very quickly moved from the technical product manager role into the role that I have now, which is a senior platform manager. We obviously expanded the domain and portfolio to incorporate dev tooling Kubernetes and FinOps over that time as well and I'm just loving it. I love being so close to hardcore engineering teams. I love the work that the team are doing from, you know, really building out those platforms to enable all of our internal consumers. So we don't work with any of the actual customers that purchase the product offering in store, but we service all of the application teams, all the services, the CFC that have to bring the product from the farm into the supply chain and out to the supermarkets. [00:03:09] Speaker A: Yeah, no, absolutely. That's an amazing journey and it sounds like there's no looking back. Despite having left gaming, it's onwards and upwards. [00:03:18] Speaker B: It is still today though. It's interesting. The gaming industry is so far ahead of many other industries, you just don't realize it at the time when you're working in there. It's when you move into large corporates like Coles that you really start to see, oh, hang on, moving at significant pace and using really cool tooling all the time. But that's sort of what we've been trying to do here at Coles as well. [00:03:40] Speaker A: Yeah, absolutely. And I guess that's why your experience is so valuable that you're able to bring and share that expertise as well. [00:03:47] Speaker B: I'd like to think so, yes. Bringing about the customer centricity and the product mindset to hardcore engineering teams is the thing that's really fun to do here, Coles. [00:03:56] Speaker A: Yeah, no, absolutely. And the relationship between FinOps, platform engineering, cloud and application teams, who should be responsible for cloud optimization? [00:04:09] Speaker B: A lot of people would like to say the finops function is, but the approach that we've sort of taken here over the last couple of years is how we embed cloud optimization as part of the day to day BAU work that people undertake, no matter which area of the business you sit in. So whether you're a delivery manager, whether you're a platform manager or whether you're the cloud engineer, it's all part of your responsibility to understand how we are best to use the cloud in a way that is optimal for the organization. It can be very expensive very quickly. Things get turned on and never get turned off. And so sharing that responsibility over a large base of internal consumers is really what helps make sure that we're getting value for money. [00:04:49] Speaker A: Yeah, no, absolutely. How long has Spin Ops been in the organization for now? [00:04:54] Speaker B: We sort of set up the functionality about three years, maybe four years ago. We are reliant on a cloud cost management tool to help us manage the cloud costs we don't work just with the hyperscaler vendors on that. And so we set up that function about three years ago, working very closely with IT Finance in relation to IT to get an understanding of what they're trying to look at and what optimization activities they sort of thought we needed to undertake. And with this culture of co responsibility, what we've done is we've taken both the top down and a bottom up approach in relation to working with the engineers, understanding the things that they see and what they work with and how they themselves can take a bit of and responsibility in making decisions to influence, you know, what kind of skew that they're taking. The ability to shut down things on their own or, or not and just helping them bring a bit of control into their own lives around what they can or can't do. [00:05:51] Speaker A: Yeah, no, absolutely. And what would you say a mature finops function looks like today? And how can it be improved even more so? [00:06:01] Speaker B: A mature one is one that is using architecture as a group of people that helps you influence decisions right up front. So when people are making, starting to build new things, having architecture have an understanding of what are the right tooling suites to use, what SKUs to apply, making sure that we're building the application so that they can be shut down, whether that's a forced shutdown or per schedule, that's really important. Having that relationship with your executives so that they understand sort of how the cloud consumption works and what activities situation with the application teams to do versus what we can do at the FinOps enterprise level, that's really important. So if you've got, you know, a cloud cost management tool in place that's helping multiple levels of leadership make decisions on what we need to do to optimize that cost, that's when you really are making a big change. We've even taken it a little bit further. So I'm very proud to say that we've even moved into a space now where we are at an enterprise where level doing low to no engineering activities that takes the responsibility off of our application and engineering teams. Because as you know, everybody's busy, the backlog's ever growing, no one has time. So one of the sort of broad enterprise activities that we've done as FinOps is implementing rules that transverse across the system to give people advice to say, hey, there's a miscellaneous storage account out there. If you, if we deleted that for you, that would save you $100 a week. And we've built enough trust up with our internal consumers that we can actually action that deletion on their behalf. So that's quite exciting because it's really going into the weeds and finding all, you know, the hundred dollars here, the thousand dollars there, because those are the really hard things to go and clean up in an environment as large as calls [00:07:49] Speaker A: 100%. And that's incredible to create such a huge change in Coles and to have the trust and the backing of everyone, engineers included, to go down that approach. [00:07:59] Speaker B: Not everyone's a believer yet, but we, we're slowly chipping away. It's definitely taken about three years here to get here. [00:08:05] Speaker A: Yeah, no, absolutely. And with. I see this, I see this quite often every day. But you know, you've got so many different job titles in the world of like DevOps Cloud platform engineering and something that has come up sometimes is a FinOps engineer. What are your thoughts on that? Do you think FinOps engineers is a thing that should exist or do you think that's something that it's just a bit of a buzzword at the moment and it will die down or it encompasses so much more than just one person. What's your take your stance on it? [00:08:36] Speaker B: Oh, great question. I don't have an answer in the sense of whether I think it's right or wrong. Every business has got to do what they believe is right for, for their operating model and what they're trying to do. The concept though, that someone is wholly and solely responsible for ensuring that your tech stack is running at optimal from a cost perspective. So not only performance or reliability, all of those things. I like that idea. Again, trying to get the priority in someone's backlog to say, hey, you need to go and clean up all these storage accounts, you need to update your scripts so that you're not reinstating things that kind of work. Having someone dedicated in a squad to do that and have them float around essentially the organization would be highly beneficial and probably would make you quite successful in realizing those optimizations too. [00:09:26] Speaker A: Yeah, absolutely. No, thank you for that insight. Moving a little bit onto bleeding edge cost optimization, where do you see organizations are investing in newer tools and automation to optimize cloud environment. [00:09:43] Speaker B: So definitely getting a cloud cost management tool specifically. That would be the new tooling suite that I think people should invest in. So that's looking and listening specifically across your entire state, agnostic of who your hyperscaler is and what other SaaS platforms you're using. Having a tool that consolidates that view for you so that you can really get into a subscription and have a look at what is the services that they're using? Why are they using these particular services? What skew other skus could you apply to it alongside? You know, looking at what opportunities do you have from a savings plan perspective or is there reservations you could be taking? I think that kind of tooling is underutilized at the moment within the organization. And then a lot of the new sort of partnerships and SaaS tooling that people are taking on, a lot of the partners today are quite keen for you to save money. I know we all need to make money, right? But helping your customer being able to realize savings means they're going to take that savings and reinvest it likely back into your tool to expand it into other features. So that sort of functionality within a tool I think is really helpful for people today. [00:10:57] Speaker A: Yeah, absolutely. And we'll probably see more adoption of that as time goes on. Which also brings me onto the topic of one that I, I just can't ignore. I don't think I'd be able to do a podcast episode without mentioning AI as it is slowly but surely taking over everyone's world. But as AI adoption accelerates, how should businesses think about the infrastructure and cloud costs behind AI workloads? [00:11:24] Speaker B: AI has. Look, it's really tough. We're struggling ourselves at the moment to really get a view of AI cost. There's a lot of associated services that enable the AI functionalities as well that people just aren't sure about, don't know about. And we're all sort of, you know, paving the path as we're moving along it. As much as we're all using the same sort of tools, there's considerations like commercial constructs that are different between, from the, for the same vendor for different suppliers. So trying to take an approach where you do have a very strong governance on, when you're bringing, when you, when you're on boarding AI services, what are the associated services and costs with that? So having that traceability is highly important. Having that raised at a governance level, that people beyond the engineers actually know that and understand it makes it sure. Or tries to help everybody with that cost visibility piece. You do need to get on top of it asap. Having your system of authority allowing you to manage budgets, if you don't have that from the outset, you're in trouble because your costs will grow exponentially and you traceability in that is quite difficult to find from a cloud cost management tooling perspective. You know, we are looking, exploring that so that we can ingest any AI tooling source agnostic of what it actually is and implement it in such a way that we're visualizing it whether it's a DataBricks or a GitHub or an OpenAI. The look and the feel of the cost for the executive and the stakeholders is the same look and feel every time. Helps remove some of the barriers of their understanding and the cost management piece around that. But again, very early days for everyone, I think, and quite difficult to integrate all these toolings into your finance systems and your cloud cost management systems. That makes it easy to understand and trace. [00:13:15] Speaker A: Absolutely. What would you say your team has done, been able to utilize most successfully within colds [00:13:22] Speaker B: from an AI tooling perspective? [00:13:24] Speaker A: Yeah, [00:13:26] Speaker B: probably sharing too many secrets here. Here we go. Anyway, GitHub Copilot has been a really easy tool for our developer community because that's the community I face into. The tooling itself as the sort of authoritative system had a lot of budgeting, functionalities, visibility, ability to assign it to users. That helped us really nut out what that spend profile looks for, looks like for a user, what the value use cases are that we could attribute to help that facilitate that conversation around optimization. But also productivity. So that's probably been a little bit of a blessing in disguise having that tooling suite integrated with the IDEs, engineers using it, understanding the technical aspects of usage, that kind of thing. That's really helped us a lot when it comes to the citizen developer. So I personally am a citizen developer. I love that we're introducing a tooling, a technical tool that is helping us better do our jobs day to day by making our lives easier. But that's the really, really difficult one. How do we govern that citizen developer? What are they allowed to or not allowed to do? Because now we've opened a door to a whole bunch of things that we don't actually govern. And that's the one trying to pool all of the AI tooling, cost visualization to get that understanding to help manage that Citizen developer is the one we really need to take control of. So that's a whole new thing for most organizations, I think. [00:14:52] Speaker A: Yeah, it's exciting that you're already talking about it and thinking about it. It's. Yeah. Worlds apart to businesses that haven't even had that thought yet. [00:15:02] Speaker B: Yeah, one of the other little, you know, things that we've learned so far is definitely finding a way to pool all of your tokens. And the reason for that is a lot of tools give you a particular budget per user and the ability to pool all of those tokens together. And then issue it out per user based on, you know, just. Let's talk about you and I. You and I are on the same pool. I'm a heavy user of AI. You're not such a strong user of AI. But both of us have a thousand credits a month. So when it gets to sort of halfway three quarters of the month, they're like DJs definitely going to need more than a thousand tokens. Well, let's just take what's not utilized in the pool that we've already paid for and give that to her because we're at, you know, three quarters of the month. Echo's not going to use all of her tokens this month. So that way you don't have any wastage on your pre purchase tokens but it also allows you then to control the spend. So you've used your thousand. I've used my thousand. You up my budget because I needed to get an extra 500 tokens because I'm working on a particular project. And we know that we needed that extra 500 tokens but we can clearly see and manage that extra spin so that we're not overspending and it's just not going. Going crazy. [00:16:16] Speaker A: Yeah. Oh yeah, that's, that's awesome. And that also brings us to our final question. DJ Kubernetes can make cloud cost management particularly interesting. What challenges do you see organizations facing when trying to understand and optimize Kubernetes costs? [00:16:35] Speaker B: We're still not there. That's the hard. So I Kubernetes for me is like the black box on a plane. So annoying because you know it's there and you know it's got, it's doing something really critical. But what is actually happening in it is just so difficult to unpick from an engineering perspective. It's just so complex. So we've done, we went down the path for a while where we were assigning all of the Kubernetes costs specifically to users and we were doing all these complex calculations about the infra apps that are associated it and apportioning that out as well. We did that for about a year. And the journaling overhead versus the value of being able to say supply chain versus cloud engineering used this much versus that much wasn't really giving us anything. There was no benefit to that visualization. So we've flipped a little bit in that we do using our cloud cost management tool, we do showback the core services around the Kubernetes tech stack. And then from that infra apps perspective we just bundle that up as part of my cost to myself as the custodian and owner of the Kubernetes offering to Coles. We have this last couple of months taken on a new tagging standard in relation to Kubernetes to try and get a little bit more granular granularity in that showback and portraying a bit more of that traceability so people can understand what their posture looks like from a Kubernetes usage. But you know, from one day to the next people are doing things with their applications or their services that just changes that immensely and we don't, we don't really take any action from that too much. [00:18:04] Speaker A: Yeah, one, one to stay on top of. Basically [00:18:07] Speaker B: the other thing that is really beneficial for the Kubernetes tech stack is the shutdown. So we have automated shutdowns every evening. We have a schedule that runs throughout the night and as soon as we can we turn things off. So that has been significantly beneficial. We've seen lots of cost savings in relation to that. We've had that now for two years and we'll continue to explore what other sort of auto stopping functionality is out there for excerpts [00:18:30] Speaker A: like [00:18:31] Speaker B: yeah, [00:18:32] Speaker A: well thank you so much for sharing your insights DJ and for joining me today. And everyone, thank you for tuning in to our final episode of season two. It's been a pleasure having you and I hope everyone's enjoyed the episode.

Other Episodes

Episode 1

May 21, 2026 • 00:19:13
Episode Cover

Leadership Lessons Behind High-Performing SRE Teams

Season 2 of Her Head’s in the Cloud kicks off with a powerful conversation on leadership, empathy, and the human side of engineering. Egle...

Listen

Episode 4

July 09, 2026 • 00:27:21
Episode Cover

Why Persistence Matters More Than a Perfect Career Plan

In this episode of Her Head’s in the Cloud, Egle sits down with Marjan Zia, Head of Cloud Engineering and Technology Operations at AFG,...

Listen

Episode 3

June 26, 2026 • 00:13:31
Episode Cover

Confidence, Cloud & the Future of Networking

In this episode of Her Head's in the Cloud, Egle sits down with Mini Mol, infrastructure leader at AMP, to discuss a career spanning...

Listen