← AWS Podcast
AWS Podcast
#718: From cost center to value creator: How IT automation creates business value
00:30:52
Learn how organizations are transforming IT from a cost center to a value generator through automati
Loading summary
Transcript43 lines
- [00:00]
Simon
This is episode 718 of the AWS podcast, released on April 28, 2025. Hello everyone. Welcome back to the AWS Podcast. Simlish here with you. Great to have you back. Joined by not one, but two very special guests. Today I'm firstly, I'm joined by Adrian San Miguel, who's a principal enterprise architect here at Adabisca. Adrian, how you going?
- [00:21]
Adrian San Miguel
Hey, doing well, Simon, good to chat with you again. It's been too long and kind of feel guilty about that.
- [00:27]
Simon
Well, let's assuage that guilt. We're also joined by Greg Harris, who's a senior solutions architect and is located in the great state of Texas. Although let's see if we can guess from his accent where he's from. Welcome to the podcast, Greg.
- [00:40]
Greg Harris
Yeah, thanks Simon. It's great to be here. It's first time here with you and haven't talked to Adrian for a little while as well, so it'll be good to catch up with him too.
- [00:48]
Simon
Exactly. We're all going to have a good chat. And for those playing at home, yes, Greg is a Kiwi, so he's my cousin. Good natured ribbon.
- [00:59]
Greg Harris
Deep, deep, deep South Texas.
- [01:04]
Simon
So the topic today is performance, but it's in a different lens. So if we often think about performance from a technology standpoint, we spend an age discussing speeds and feeds, the latest hardware, which chipset is right, which AI model works faster, how many teraflops we can squeeze out of things. There's like lots of detail that you need to go into and you do need to go into that. But it's interesting from a business standpoint. I'll do the big reveal. Most business people don't care about that stuff. They sort of wave the arms and go, yeah, my IT folks worry about that stuff. So today we want to talk a bit about performance, both from a technology standpoint, but from a business outcome perspective. So maybe Adrian, let me start with you. Give us some context of what you've been seeing at the macro level and why we're talking about this today.
- [01:55]
Adrian San Miguel
Sure. So more and more, given the economic headwinds and what have you that the current climate has kind of brought to us, and I'll leave it at that. There has been a bit of a renewed focus and kind of the age old adage of wanting to treat it as a cost center that's never really gone away. And even though in the pre Covid times we were starting to finally get it and tech and business folks in the room to actually talk together and work together, Kind of taken a bit of a step back there, especially with the expend additional expenditures of folks getting to cloud and rushing to get these projects done in a timely fashion. And as we continue to normalize, it's kind of one of those things, bit of a finger pointing exercise saying hey, you didn't realize the value that you told us we did this or we may be actually spending a little more money. But what I'm seeing is for the first time in quite some time it is transforming itself from a cost center to, to a revenue generator. And this has an interesting focus on performance because you know, in the old days you, you would take a, a knock for example, where I lost most of my hair and the color in my hair due to being up at unforsaken hours of the night a while back in, in about 2019 and I did a reinvent talk on it, GPS bus 202 where we talk about the cost of not doing stuff. And that's still very true where if you take a individual that is chasing gremlins, for example high cpu that has a real dollar value to it. This is what the business people always look at. It costs money to staff a knock, it costs money to do X, Y and Z. And eventually you have to change the perspective in the conversation from this is also what we're able to do for you because we're not doing that. What I'm starting to see now in the partner space and interestingly enough, and I know Genai is the soup du jour and that's all anybody wants to talk to but I promise this is actually like a real thing that isn't something hokey or that can be put together in about two minutes. A partner of mine that I, excuse me, work with quite deeply innovative solutions. They were able to eat their own dog food and prove back to the business. Yes, we can help you by not having to run our systems and focus on actually reducing the footprint that we consume. They were able to cut a business process for month end invoices for about 84 hours by using AI and using AWS services in about an hour and a half, two hours tops.
- [04:20]
Simon
So 84 hours to two hours to do a business process.
- [04:23]
Adrian San Miguel
84 to three individuals, full time employees dedicated to rerating, recoding, invoicing, triage. No longer doing that, they hit the button. Magic think box goes and spits out a thing. A quick QA and QC is done. Off to the races. Customers are singing praises about more accurate invoices. Less need for a follow up, less need for getting on the phone and nobody wants to call and dispute a charge. And it has for the first time in a long time, this partner is actually feeling valued at the table by its business counterparts simply because of the performance that they were able to bake into their systems and just let the bots, let the automation, the self healing value, self silicon over carbon. Let the machines do their thing so we can help you, the business start realizing that revenue of no longer having to chase that pie in the sky. And of course I've got plenty of different stories here, but it would be interesting to hear if I'm kind of scratching that itch that you're talking about now.
- [05:26]
Simon
It's interesting that the perspective there is very different. As you mentioned, these are folks who adept at performance analysis and diving deep on technology domains, et cetera, yet they've been forced or wanted to pivot into a business process and identifying often what to the outside observer is just oh well that's obvious or it's low hanging fruit. But it may not be because often the quote unquote business side, and I dislike using that term in as much as we're all part of the same business, but if we have to categorize tech and non tech, the non tech folks don't know what to ask the tech folks for. And so the tech folks kind of need to ask the non tech folks, hey what, what takes the longest? And if someone says, well it takes us 80 hours to do month, then there's three folks doing it and they have a clear eyed view and go, actually we, we could change that. There's a mental model shift here that I want to, want to dive into. Adrian, how, how do you, how did they sort of come to that agreement? Like they didn't just trip over 80 hours of processing. Oh, we could fix that. Like what, what had to happen.
- [06:27]
Adrian San Miguel
So what had to happen is as so far as many of us that may have children or know folks that have children like to be held as a bonus uncle. It has to hurt a little bit sometimes. And in that regard there's so many things that you can do to warn people and help them see you around the short corners or as you mentioned, bump your head with the low hanging fruit. But it's until that there's a level of discomfort that is commensurate with understanding that you have to go and do something. That's really what it took here and in this partner's case it took the realization that they were falling behind on invoicing, that they were missing their SLAs to customers that it was a difficult decision to have to really look at things where everybody across the industry is battening down the hatches and reducing the amount of expenditures that they have and to have to be confronted with. If we continue at this trajectory as we take on customers and expand our scope of support, we're going to have to hire 2, 3, 5, potentially 10, a full staff of entire, you know.
- [07:27]
Simon
It was only going to get worse.
- [07:29]
Adrian San Miguel
Exactly. And it the sprawl becomes exponentially more difficult as this partner begins doing more things like exploring, going up upward into the enterprise market as opposed to the SMB greenfield space where they're playing today. Which again incidentally because of their hyper focus on performance, they can actually help the business go out and tackle these things. But as they're doing that exactly to your point, they came back to the business and very much so as a white glove consultancy to come back and say, hey look, what are the top three things that are bothering you folks? In very much the same way that they would approach a customer, they are approaching the internal business stakeholders. And here's key, they're approaching them just like a customer of IT ops of the IT team where they're saying don't worry about the machines, we'll sort the bits and bobs. What we care about is helping you achieve your, your business outcome in leveraging of tech to do so.
- [08:24]
Simon
Now I'm guessing part of that comes from a degree of, I guess trust or awareness or connection where you know, if as an IT function you're struggling to keep the systems up and running. Typically the business stakeholders are not that interested in having these cool conversations with you. They're like, hey, just I don't even trust you to run the stuff. Tell us a bit about some of the technology underpinnings that allow this to happen because you touched a bit on self healing and automation. But let's dive deep on that. What's in place for this customer to help them get to this point.
- [08:53]
Adrian San Miguel
So the partner is making use of just about every single one of our AWS managed services, but also is a strong proponent of automation by default. So much so that their support team is no more than a handful of individuals for their external and their internal customers. Like the different business units and auxiliary companies that they work with, they, they're a very power whatever the next upper echelon is beyond power user of EKS. They are a very strong user of SageMaker. They believe heavily in distributed systems as well as NoSQL so their power users of DynamoDB where they have fully bought in as a managed service provider to allow the managed services to do the thing so that you can use your humans to actually better your business processes, to actually work on iterating and improving your P90s, your P99s of your systems and the underpinning. So for example, when I first started working with these folks back in about 2018, 2019, it on average took them about four to five weeks to onboard their customers because everything was still very manual and tedious things like, hey, we need to go launch four EC2 instances. And it was somebody sitting at the console potentially and just manually doing these things. Yeah, maybe using a launch template here and there if it was similar enough. And now everything is automation by default. They're a terraform shop. Everything is defined as code checked into a centralized repository, various levels of QA and QC before anything gets done. They have a base operation template that they know is validated and good. They have automation built in to go out and pull the latest CIS hardened images as applicable. If there's a specific compliance framework that needs to be complied to, they pull that out of their repository. And much like many of their customers and other partners in the space, they're leveraging things like EC2 Image Builder into their pipelines to act off an SNS notification to say, all right, cool, we got a new update of this server 2019AMI with SQL baked into it that solves for these handful of KBs, go through, pull it, run it through their near prod environments, make sure it's not going to blow something up that way at the next patch Tuesday, offer the customer the opt in or the opt out to go and do the thing automatically through use of SSM and SSM patch manager to effectively orchestrate the whole thing from start to finish so that all the customer has to do is either respond to an email or respond to a ticket and say yes, I want to do this, or potentially postpone a couple clicks or a CLI command. And that too is automated so that the only time that they ever touch a customer is if it's something that is both high value and something that is relevant to the customer's business problems that they're facing. Their goal is to never have to reach out to a customer and say, hey, look, your EC2 instance is down, or hey, I need to restart Lamp or Apache. Is that cool if I go through and do that?
- [11:46]
Simon
Yeah, all the transactional stuff this has always been the challenge is that I think roughly 80% of it time is spent on just maintenance, keeping the lights on and 20% on the cool stuff. And the goal has always been to flip that script. And automation gets you a long way there.
- [12:03]
Adrian San Miguel
Yeah, it does.
- [12:05]
Simon
Adrian, tell us a bit more. We've touched on Gen AI because you know, that's what a lot of folks are talking about now. And I think what I'm noticing is people saying, well that's great, I know you sort of can do things, but help me understand the deployment pattern that was used to solve a problem, you know, so what do they actually do? So in this case, you talked about the fact that this technology helped shrink that business process. Can you unpeel without giving away any sort of secret sauce what it looked like from an architectural perspective.
- [12:31]
Adrian San Miguel
So I'd be happy to talk about it from an architectural perspective. The underpinnings, there was AWS bedrock at the absolute core of it, leveraging both the Claude Haiku model as well as I want to say it was Sonet 3 I think is when they first started dorking around with it. I may be incorrect on that, but interestingly enough it is as simple of a model as you can think of. If it's they are eating their own dog food, it's the exact same solution that they put out to customers. And I can definitely leave you a link there to, to the success story for this. So what they do is upon purchase, a customer is going out and launching a EKS cluster that is comprised of a couple pods and worker nodes that are going to spawn the processes and effectively the underpinnings of the chatbot that is going to be developed and dedicated to this customer for this specific instance. Behind that is going to be sitting a weaviate database for Vector Store with the customer having the full ability to either swap in, hey, I'm a post pre shop, we prefer to use Elastic, we're doing everything through DynamoDB. Whatever the individual case is, it's configurable enough to allow for the core source code to go through and leverage this. So the really interesting thing here is at this point is where it starts getting kind of, kind of complex. What we're starting to see and what they saw interestingly enough, was just pointing to S3 for object storage to hold the documents, to hold the xml, to hold for this particular instance the raw invoicing or the customer usage reports of the curs. It was good enough. But after over time, as the bot became more and more intelligent and more performant, they realized object storage is good for the overall repository. But now we're having to start tiering storage. So they introduced a EFS layer in front of it to effectively serve as that intermediary to help determine and judge what type of usage, which customers, which accounts, which segments. Where are we seeing the most churn and burn? Where are we starting to see the most interaction to help further on the back end, reduce the cost of having to do the gets, retrieves and pulls from archive. If we have a customer that's asking for the same handful of things over and over again and in the middle of that transaction, if a customer pulls up a browser and says I need to understand what the billing is for container services on this account, in this region, as the query goes out of Bedrock and it goes through the architecture, VectorDB is gonna plot out exactly where within the repository all of this information list comes and gets spits out. At that point, what really tripped them up was the inability to go out and ask additional questions that we hear a lot about multimodal, we hear a lot about multip engagement. Complex queries is exactly what it's talking about. Okay, great, tell me this, but I also need to know specifically EKs for the last 30 days and that's where typically the rails would start falling off. So what they were able to do was leverage built on aws of course, IBM's WatsonX agent platform. Using that in the way that it was set up finally gave them the ability to install that intuitive nature really for the first time, let them get as complex or hyper specific as they want to be able to further drill down its say okay, but I need to know only on GPU backed instances. I need to know G4DNs I need to understand in this region, in this Availability zone, in this particular data center or whatever the macro is that they're specifically trying to look for that gets spit back out, presented back to the user on average in about under a second and a half in most instances, as opposed to having to open those in some instances, 100 gig CSV files either sitting on somebody's laptop or an S3 waiting for that to munch, running the query, getting the output and then continuing to have to massage it. So just the simple laying that down and showing that back to the business was kind of revolutionary to them because they got used to okay, third Friday of the month we're going to start running invoicing and this kind of going to run through the weekend. Maybe it'll work, maybe it won't, maybe we have to restart a run. So when they started it at about a noon on that Friday. And when it was done by like 130 for every one of their least 150 different customers that were leveraging that they were in scope for this first run, they're just kind of besides themselves. Like, I can't believe I don't know.
- [16:55]
Simon
What to do for the. It's almost like, did it work? Like, did it even run?
- [16:59]
Adrian San Miguel
Yeah, sure, it worked. They go check and they, they go, hey, this is different. This is liberating. This is not something that we were really thinking was within the realm of possibility. And as you made mention before, the trust was there, the buy in, like, all right, cool, we don't have to worry. This stuff we know you got, you folks are running it, but for us, it gives us that level of confidence to not only expand the scope, but you gave us back all of this time that we can either work to iterate on invoicing or do other things that are far more valuable than just kind of sitting here and babysitting a batch file or just some long running process and hope it doesn't time out.
- [17:37]
Simon
No, that's a great example of that direct benefit from a business standpoint. And Greg, let's come to you. So you're located, despite the New Zealand Zealand accent. You're in the great state of Texas in one of the world's most famous universities. Tell us the story of what your customer has done and how they changed organizational performance.
- [17:59]
Greg Harris
Yeah, so Texas A and M is the one that I spend most of my time at these days, and they've gone through a whole lot of change over the past few years. It all really kicked off around about COVID coincidentally just happened to be the same time with a lot of organizations.
- [18:13]
Simon
Funny how many places that's happened to.
- [18:15]
Greg Harris
Yeah, I mean, I think it helped speed things up in a lot of ways, but it was stemmed by a lot of leadership changes around consolidating their IT organization as a whole and sort of trying to get their arms around shadow IT and improving efficiencies and economies of scale and optimizing all their spend and all that kind of thing, but also driven by a cloud first division. The cto, who wasn't CTO when I started here, but has now moved into that role. He is our, he's our champion here. He's the one that helps us really drive everything. So all the tech organizations under him and really that allows us to have that level of trust that we've built at that layer and sort of, you Know, it filters down, down through the organization and we take that trust out to the business as well and bring them in to start talking to it about those business problems that. That you and Adrian have been just discussing. Right. And help build that trust within that organization and overcome some of their previous baggage because it's public sector and there's always that kind of thing. But we did a couple of things and one of the things that happened on, I think it was my third day in the field here is we got an escalation. It's like, hey, we have to build a Covid dashboard for the university. Great. Okay, we're going to do that on quicksight. We like quicksight. Turns out quicksight can't do. Couldn't at that time do public shared dashboards.
- [19:33]
Simon
Certainly can now.
- [19:35]
Greg Harris
Certainly can now. And that's part of the reason why.
- [19:37]
Simon
There were some of that customers needed it. Yeah.
- [19:40]
Greg Harris
Some of that customer feedback made away. Yeah. So sat down with them and over the course of less than a week, there was one developer at the time on that team. We helped or I sat down and helped him build a serverless dashboard to take all of that data. It was still hosted in Athena and S3 and pulled from a Redcap database, which is sort of a server. It used to be a survey tool, came out of Vanderbilt. It's grown and to a lot of other capabilities now. And we built that over the course of just a few days sitting down, doing some buddy programming, put it fully serverless. It was up and running within the timeframe that he was given, which was very, very short and has actually, I think it's still live, still serving the university and the county. And then because that was successful over a couple of days, fully serverless, he was given the task of building the code vaccination scheduling system and reporting system for the Brazos county around Texas AM University. Now, we didn't hear about that until after the first failure, unfortunately. And you never want to be on the front page. Right. So they came to us and we took a look at that with them. And it was a pretty standard deployment backend on rds and it was running in elastic beanstalk, actually. And we sat down with them and went through, overhauled that architecture. We brought in our specialists, we brought in a temporary TAM and we just overhauled the entire architecture. We optimized all of the database queries, we optimized the RDS solution. We provided some feedback at the time because observability into RDS was not the greatest then for looking at slow running Queries has significantly improved due to customer feedback like that. And they went from crashing instantly to being able to serve on their relaunch, something along the lines of 48,000 unique customers in the first five minutes of that connectivity. And it did not go down again. And then that ran through Covid, several waves of vaccination releases and appointment scheduling and that kind of thing. And that was on the front end and on the back end of that same application they had the field side of it where they're actually doing the administration and the tracking and all the nurses in the field were logging in and administering the vaccinations and then tracking the vaccination administration there as well. So we went from a failure and front page news to absolutely perfect. And no one really knew it existed in the back end over the course of just a couple of weeks of that very tight time frame where they had to relaunch. And that was something that helped us build a huge, huge amount of trust as a new team coming into the customer with them and allowed us to sort of branch out and talk to a lot of the other IT folks. But then that's been and gone. That's Covid, as we all know. Covid's gone now. We won't go there. But something that we've recently talked about and worked with the new team here as they've consolidated is around their website deployment hosting service and they used to host that on premise and VMware on VMs. And every website had its own VM stack and all the rest of was all click ops and it used to take them if they were really, really lucky. And it didn't, it weren't a lot of, wasn't a lot of work to be done. Maybe it's taken some days, some days to deploy a website. Usually it was in the order of a couple of weeks from hey, I need this website to it's going live. That results in some backlogs and frustration and a lot more extra shadow it. So working with their architecture team we managed to get a fully serverless, almost fully serverless architecture. They still have their DNS on prem, but at least that has an API so they can talk back to that. So they're using Lambda lambda and Edge S3 for hosting and Cloudfront. And from the time that the marketing teams or the website development team approve the pull request and get it's just, it's now just a few minutes before that site is live. So we've gone from a couple of weeks to just a few minutes and it is fully automated now and all they need to do is train the teams how to commit to git because that's a learning curve for sure. And it goes through and it validates the style sheets, it validates the specific templates they need to use. It runs a bunch of security checks in the pipeline. It uses GitHub and GitHub Actions primarily for that side of things before it pushes it out here. And then it deploys all the AWS infrastructure again using Terraform. So they are a very, very big terraform shop here. So they're doing a lot.
- [24:11]
Simon
It sounds like they focused on not clicking faster, but actually sort of, you got to step back and think about the problem again from that outcome standpoint. This was that we need websites up to service students and researchers in the university and we can't do it quickly. Can't do it quickly.
- [24:32]
Greg Harris
Yeah. And once they shifted over their main university website, it was very hard for departments and other folks who say, oh well, I don't really believe that's going to work because like, well, we're running our primary websites on this and it works fine. You guys didn't even know, right? So it's there and that be able to prove again, like you were saying, eat your own dog food. Like, we built this, we run this, we do this, we recommend you do this as well. And it will save you time and add value. So what the team here is really trying to do is given public sector, it's not revenue generating, but it is value generating. They're trying to generate the value for the organization and multiply the value for the organization. So it's a little bit of a different spin on that revenue generation standpoint. Well, they are still technically a cost center. They are trying to escape that moniker as well.
- [25:23]
Simon
Well, I guess they're trying to help the institution be more appealing to its potential attendees. As any higher education wants to attract the best researchers, wants to attract the best students. And part of that is, well, what's my life like? Is the campus nice? Are the services good? And technology is now kind of front and center in that because certainly the current generation of folks coming through, they expect a full digital wonderful experience. They don't want clunkiness.
- [25:52]
Greg Harris
They do. And they're also trying to get over some of the restrictions of obviously the state pay. Right. I mean they can't go out and pay like the Amazons and the other big techs. So they have to really optimize their human resource time as well. So the investment into these technologies really allows them to go and innovate more and do more of these really cool optimization projects that they've been working on.
- [26:18]
Simon
So it sounds like if we think about, thematically about what we've talked about today, really what we're saying is a automate as much as you can because that actually frees your time from firefighting to actually having these conversations. Because the secret here is you have to talk to people. And as IT folks, and I speak personally, we don't always like talking to people. Like we've got into it because we have these beautiful machines and we can issue commands to and stuff happens. But you got to literally go and talk to people and probably people you don't normally speak to. And those people are likely to go, why are you talking to me? My computer's working. Why am I talking to you? And so you've got to change the conversation. You got to earn some trust by showing stability in the environment to both demonstrate capability, but also to give you that mind space to do this more interesting stuff. Then you've got to ask really great open questions of what's the top three problems you have? From a business process perspective, it can be as simple as what takes you the longest.
- [27:18]
Greg Harris
That's key there because a lot of IT folks are used to defining a solution and then going out and dictating that solution to their constituents. And that does not build trust. So switching up their mindset and going out and asking them the actual business problems and the challenges they're trying to solve and then working backwards from there and building that solution, that changes things and people don't expect that from it.
- [27:42]
Adrian San Miguel
Most of the time it is disarming. I'll vouch for that. It's almost confusing where they come out of that and say, did I just talk to an IT person? They don't expect us to approach with the business hat approach it in that lens. And I haven't been, you know, fly on the wall in a couple of Greg's conversations with, with Texas A and M, despite the fact they may hold it against me that I'm a die hard Longhorn fan, but we won't go there. They appreciate profoundly approaching it in a business sense. What can we do for you? What are the type of things that bother you? What are the type of things that are on your plate? Changing the narrative and having that trust built with credibly and reliably having those systems up and perform it, that's key. I hate to say it's table stakes, but to change the perception of tech and Folks in tech, over time, that's a skill set that just has to be picked up.
- [28:30]
Greg Harris
Yeah. And I will say that the way that we bring security to the table as well, and that just underpins everything that we do, that eases a lot of those concerns as we're talking to both IT and business, because they're always concerned about, hey, I don't know, is the cloud secure enough? It's like, is it secure enough to what we do? It's not how we've always done it. Will it give us the visibility that we need? And we can turn around and point to all of our large customers, our federal customers, our government customers and say, well, these guys are doing it, it's going to work for you. We've got all of these use cases for public sector and enterprise that do similar things to you. And again, that helps build that trust and it speeds up those conversations. From the performance side of things, you can build it, but you still have to get it out there. You still have to work your way through those, those discussions.
- [29:16]
Simon
And it's interesting you touch on that because certainly, you know, particularly in public sector organizations, risk management is naturally front of mind huge. And what I've found, almost, almost without exception, is that when you discuss the automation in the cloud and the ability to set guardrails and the ability to perform real time auditing, et cetera, they become the biggest advocates. Like, yes, this is the way we need to be doing things. Yes, I want to better click here and get this set of criteria applied to my environment. I want to know that people can't do the wrong thing accidentally because they just can't. Like, that's, that's a big shift. And again, it moves you away from, from click ops, individual heroics, you know, bailing wire and sticky tape to make things work. You know, it comes back to it and it's, and it's, you know, to some degree it's, it's quite unquote boring. It's like, well, go automate this stuff. It's like, oh, it's like that's, that's what gets you out of this trap. So if you're in the trap, automation is your friend.
- [30:13]
Greg Harris
Well, automation also helps you to understand the business processes and whether you actually even need to have that process anymore because there's no point automating something that you don't need at all. Right. And then, because if you can drop that, then that improves that. It frees up resources and it reduces cost. Straight off the bat.
- [30:30]
Simon
Absolutely. Greg, thanks so much. For coming on the show and telling us all about it. Thank you. And Adam, thank you for sharing some context on the way we can look at performance from a not speed and speeds perspective.
- [30:41]
Adrian San Miguel
Yeah, happy to have been here. Thank you so much for the invite.
- [30:43]
Simon
And thanks everyone for listening. We do love to get your feedback. AWSpartcast at Amazon.com is the place to do it and until next time, keep on building.