In 2017, Dunelm came up against a challenge familiar to many retailers: how to meet the demands of the modern tech-savvy consumer with technology that was designed a decade ago.
Join us as we learn from Tom Hayman, Dunelm’s Head of Platform Engineering, how the UK’s leading home furnishing retailer completely reimagined their digital platform to prepare it for the decade ahead. He’ll also cover the challenges they faced and the lessons learned from this ambitious overhaul, including how they worked with partners like Fastly and Datadog to help them deliver on their ambition.
Chloe Rigby (00:03):
Hello, I'm Chloe Rigby, the editor of internetretailing.net. Welcome to today's webinar: Out with the Old, How Dunelm Transformed Itself by Rethinking its Technology. And today's webinar is held in association with Fastly. Today, I'm going to be talking to Tom Hayman, Dunelm's head of platform engineering, about how back in 2017, the home furnishing retailer, now ranked elite in RetailX Top 500 research proposed the task of re-imagining its digital platform in order to make sure it's ready for the decade ahead, which meant it was in a good position to handle the unexpected, the Coronavirus pandemic.
Chloe Rigby (00:41):
So we've got a three part structure for today's webinar. First, Tom's going to be talking about the transformation program in a short presentation on how Dunelm has worked with partners, including Fastly and Datadog. And then Tom and I are going to broaden that out into a conversation. Finally, we'll move into a Q and A session so that you can put your questions to Tom. And you can ask those questions through the chat box on the screen, and we'll get to as many questions as we can today. Just to let you know as well that the session is recorded and we'll be sending out an email with the link so you can watch it again, whenever is convenient for you.
Chloe Rigby (01:19):
I'm going to introduce Tom now, and I'm going to turn off my camera, but I'll come back for the conversation section after. So without further ado, it's my pleasure to introduce our speaker today. Tom Hayman, head of platform engineering at Dunelm. Tom, welcome to the webinar.
Tom Hayman (01:37):
Thank you. And welcome everyone. Cool, I'll get started. So why did we replatform dunelm.com? We basically had a vision from the top to the bottom of the company where we wanted to create a home that everybody loves, to become the number one homewares destination for our customers, and an even better business that grows sustainable. Our ambitions, as I just said, is to bring our service and experience, as well as product, be a company that operates responsibly in all our communities, leading on sustainability in homewares and a great place to work for everyone. We want to grow our business and be basically the best homewares company there is in the UK.
Tom Hayman (02:21):
So we had some challenges with our existing platform, and it definitely helped us get to a place in e-commerce, where we're very, very successful and we're selling lots of products on our website, but it was a very old and monolithic architecture. It was very difficult to scale. It was in a data center on typical hardware with lots and lots of dependencies. And it was very, very difficult to manage and to take volume and capacity when we had different periods of time, for instance, in peak and so forth. It was really, really tough to make change on the platform as well. To make one change in a basket or checkout experience, it would take 10 to 15 people many, many hours to make that change. So we summed it up. We were making about one or two changes a month and that wasn't enough to go fast and accelerate the business.
Tom Hayman (03:19):
It was also not resilient to failure. So if one part failed, everything failed. So for instance, it was tightly coupled with our order management system at the time. And that order management system, for instance, was linked to the database that the e-commerce platform used. So if the database failed over to a secondary location or just went down, or something went wrong with it, not just the website would go down, but the contact center wouldn't be able to make, or help customers with orders. We also had a massive third party reliance. We couldn't really make much change quickly without engaging with our partners that helped build the e-comm platform. So this also meant that if we had an MI, or a problem with the site or something was not working, we couldn't make the change ourselves. And it was very, very slow and tiresome. For those reasons, we knew that we had to rebuild and replatform.
Tom Hayman (04:16):
So some of the things that we thought about when we were at the start of this journey, some of the principles and enablers for our future strategy included microservices, cloud and API driven architecture. So with microservices, we wanted the teams to focus on core business domain capability. We wanted to be independently deployable with cloud information so we can go faster. We want the teams to scale horizontally as requirements came in, adopt new technologies and innovation for shorter refresh cycles. We wanted to focus on automation so we could be consistent across environments so we could ensure security and IT practices were consistently applied.
Tom Hayman (05:01):
We looked at different routes of how to engineer and build this new platform. And at the time, we decided to adopt serverless technologies. And that basically at the time, was very, very difficult to do that because it was kind of, as it is now, it's still awfully quite, as I mentioned, quite new. So we did struggle a little bit at the beginning to understand how to deploy, and maintain and configure our new serverless technologies. But it gave us the ability to be highly available to scale automatically.
Tom Hayman (05:32):
We could also see that there was less overhead of managing servers and infrastructure, whereas before on the old e-commerce platform, we would have to do patching and maintenance out of hours. Teams would have to be up late at night and so forth. We wanted to provide the teams with the ability to self-serve so they could adopt serverless technologies, but also do the things that they wanted to do to build their microservice. And with APIs, we wanted them to be loosely coupled, extensible and drive collaboration and reuse. This meant that services and products are opened up for the people that use them for the business and for our IT partners
Tom Hayman (06:14):
Just a second. Cool, so I'll move on. So our technology stack consists of the following providers. We wanted to build on the latest technology. We want to be cloud centric and we wanted to choose the best of the breed. AWS definitely helped us build a service capability, Lambdas and API gateways were heavily used to build our new platform. A lot of the site is built in React, TypeScript. We use Prismic for our content for our CDA. And we also used Fastly after we replatformed the site. And we also use Datadog to understand where monitoring and failure occur.
Tom Hayman (07:11):
So what does this mean? We built a new digital platform. It gave us the ability to deploy and make change over 200 times a month. We could scale pretty much automatically, it was much more highly available. The key thing that enabled the teams, and the business and the stakeholders was that we built it, we run it, we own it. So if we made a mistake or a failure, it was up to us to fix it. We could also build new features quickly. We could release them very, very quickly for our customers. We can sell more now. Website page load times improved significantly. So we took seconds and seconds off the homepage and the other parts of the website.
Tom Hayman (07:57):
During peak periods, which are generally between December and January, for our sell and Christmas period, we can take over 150% more traffic now, which is really, really good. And when we configured and decided to use Fastly and deployed it to 100% because the platform, we could see an average speed improvement of 473% across the entire platform: product pages, search and basket. The homepage had over 900% speed improvement, pretty much across the board. And very subjectively you can see, just by going to the site, after we started using Fastly, that you could see the load time was massively increased. Also gave us the ability to feature flag new improvements to a subset of our customers.
Tom Hayman (08:41):
So for instance, we've got live chat, for instance, to help our contact center. So our contact center colleagues can talk directly to our customers if there's a problem with an order. We feature flagged that, so we sent that to a sub set, I think it was one or 2% of our customer base for a short period of time. So we could work out the training for the contact center and get them up to speed, as well as it's getting feedback from customers about how successful it is. Also provided us with better security in terms of the web application firewall. And we have a lot better view and understanding of that now since we moved to Fastly. I think that's it. Any questions?
Chloe Rigby (09:34):
Sorry about the microphone. That was really interesting. And now Tom and I are going to build this out into a conversation about the Dunelm experience. Tom, if you put your presentation back into presentation mode, we can get to the slides. Brilliant. And when we get to the end of our conversation, we're going to broaden that into some questions. So audience, if you have any questions, do start to ask them via the chat bot. So Tom, obviously in the light of COVID-19, the timing of your platform transformation has been really quite excellent. But back at the beginning, one, you didn't know there was going to be a pandemic. How then did you and the digital team get broad buy-in for this kind of change?
Tom Hayman (10:22):
Yeah, we were having problems with the platform because it was kind of coming to the end of its life. We were having problems, as I said, at the beginning of the presentation, having problems with scaling. It was very, very expensive to buy the licensing for instance, to be able to scale. So that was definitely a factor. As I said, it was very old and end of life. And we definitely thought that we could perpetuate that problem and upgrade that platform or take it into the control of ourselves or in hand. So we decided to do the latter, and we had to bring on obviously stakeholders within the company, which included obviously the exec as a massive support in that area.
Tom Hayman (11:03):
And it took us a lot of hard work. It took us around 18 months, two years for us to replatform that. It was roughly about 150, possibly 200 people, at some point or another, working on that. Lots of development, lots of dev ops engineers and so forth. So it was definitely not an easy task, and probably one of the biggest projects I've been involved with, if not the biggest one and possibly the most successful one. But it was very much a very big step change for the company in tech at Dunelm. And very big success story. And now when we look back, we'd 100% do that all over again with all the late nights and stuff.
Chloe Rigby (11:46):
So how important was it when COVID did then arrive, to have done this? I think you built on this platform to provide a new level of service online. Can you tell us about that?
Tom Hayman (11:59):
Yeah, so we had lots to do. We had reserve and collect before, but now we've got click and collect where we take the payment upfront and the customers go and collect it from store. So that was definitely a great enabler for the business as a whole. But when COVID came as well, we had the ability to basically take the platform offline. I think it was March the 23rd or something like when lockdown came. I think the following week or thereabouts, we decided to bring the site down into browsable mode because we had to do a lot of work internally to make sure our colleagues could be safe and work in a safe way. Senior staff in-store, and distribution centers and so forth.
Tom Hayman (12:43):
So we took it down for around a few days, a week or something like that. So he brought it back up with different offerings from different suppliers and so forth. We wouldn't be able to do that really quickly if we hadn't have replatformed. And we wouldn't have obviously been able to make the change subsequently to different parts of the platform to enable us to go faster during COVID. And also the demand has changed, probably for a lot of people on this call.
Tom Hayman (13:10):
The demand has definitely changed in terms of traffic patterns and stuff like that in terms of people shopping. Shopping behaviors have changed. People obviously are at home a lot more now. We've seen, and as other retailers have seen a growth in online shopping. And without those things, without replatforming, we definitely wouldn't be able to do all those things.
Chloe Rigby (13:30):
No, absolutely. And I think you've also enabled some in-store digital services, possibly as a result of COVID as well during COVID.
Tom Hayman (13:41):
Yeah. So a lot of that stuff is to come. So we're going to look at how are we going to enable the platform. We look at it as a platform, not just a website, that can help us move. So it's already helping us with inventory, and stock and things like this, but also possibly helping us in the future of different ways of selling to our customers and all the different hooks the platform has that will enable us to build the business to go so fast from a technological point of view.
Chloe Rigby (14:09):
That's great. So there must be an enormous team behind all this. How do you go about practically organizing the work and making sure everyone's working in the right direction?
Tom Hayman (14:19):
Yeah, so we divided the teams up into two pizza sized teams. So two pizzas can usually feed around, depending on how much pizza you eat, eight to 10 people, or something like that. So we sliced teams up into what we call squads. They'd have between six and eight or nine developers, something like that, as well as a product owner and delivery manager. And we organized teams in that way. At that time, it was more customer journey teams.
Tom Hayman (14:49):
So teams like checkout, and basket and browse, which would be like the front end team. And that helped us understand and drive delivery in those areas. It was also not easy to do, and there's lots of organization from different levels going on, but I think having the teams, giving the teams the ability to self-serve and be semi-autonomous, helped us to go faster and deliver the platform in each of those different customer journeys to IT.
Chloe Rigby (15:25):
Great. You've mentioned a blameless culture. How important is that?
Tom Hayman (15:32):
One second, sorry. Can you still see my presentation?
Chloe Rigby (15:37):
Tom Hayman (15:39):
Cool. Sorry, one second. What did you say?
Chloe Rigby (15:43):
Okay. You mentioned a blameless culture. How important has that been?
Tom Hayman (15:50):
So that's been really good. I mean we definitely believe that people, teams, systems fail. I think if you blame them, it's just generally quite unhealthy and demoralizing for people in teams. So we took the position and the idea that you shouldn't go after teams and blame people. You should understand that failure so you can improve and get better in the future. So we do things like a post-incident or post-mortem where we summarize everything that's happened, all the calls that have happened or things that happened around the event.
Tom Hayman (16:26):
So if it's a major incident, for instance, capture all that stuff and try and think about how we, as a whole team, as a whole, we're all accountable for failure. How we can improve as a business and as a tech team to make sure that we can learn from it and improve and get better. And that's gone really well with the teams. And it's basically given us all, definitely in my opinion, some more trust with the engineers that you can try things that are new, and if you fail, we can learn from it and get better.
Chloe Rigby (17:02):
That's great. So where do you go from here? How are you planning to build a new platform in the future?
Tom Hayman (17:08):
Yeah, so I think as you touched on the questions in this, so store technology would definitely be improved from the platform. We've aligned things like stock control and things like that. And product control through the platform. Also gives us the ability to go after, to release more. We're doing 200 releases at the moment a month, or thereabouts. We're quite ambitious. We want to double that, triple that, quadruple that into the thousands. We want make this website or platform even faster than it is now, which I think since we started using Fastly. Potentially we'd want to, basically every part of the business basically links into the platform. So there's everything's up to ourselves now, which is the great thing.
Chloe Rigby (18:00):
That's great. Now let's see if we've got any questions from the floor. This one's from Rosie who says, "How's your mobile site being affected?"
Tom Hayman (18:09):
Say it again.
Chloe Rigby (18:13):
How is your mobile site being affected? I'm not quite sure by what, Rosie, but has it been affected?
Tom Hayman (18:21):
So with our traffic, we definitely see a lot of people obviously shopping on their phones and stuff, especially in the evenings. So we're definitely seeing, generally speaking, or previously to COVID, we'd see a lot people shop generally from their servers or whatever at night time, on their mobile phones, stuff like that. That's continued, but we do see a bit of a mixture of all of that now. But essentially the mobile site, when you go to the mobile version of the site, is a lot more accessible, easier to use, a lot more faster. Yeah, so generally speaking, the mobile experience generally for our customers is faster, better to use, easier and more reliable than it was on the previous platform.
Chloe Rigby (19:13):
No, that's great. What are the core things that you've helped to start or scale the digital platform?
Tom Hayman (19:26):
I guess it was ambition and buy-in from the exec. We had to obviously be sponsored by an executive to go off and do this. There was a lot of planning to start the project, and we broke it down. We did a big storyboarding session for instance, where lots of Post-it notes and things were put onto a huge board in London and we have pictures of it. It looks very scary now. So yeah, that was the starter when we first started this, which seems a lifetime away now.
Tom Hayman (19:55):
It was a lot of work and a lot of collaboration from people between not just ... I guess this is quite an important point. Not just us or techies of the business, but engaging other people and other business partners within the business to make sure we understand and understood what they need. And we could take that back and work with them to deliver different parts of the platform for them.
Chloe Rigby (20:22):
That's great. Thanks Tom. So now that you've built this tech stack to support your core business in the new digital world, do you now envisage adding a lot more capabilities? For example, maybe visualization experiences in the absence of store traffic?
Tom Hayman (20:40):
Yes. So we'll definitely look at things like answer the questions. We're looking at image search and visualization, augmenting reality, for instance. So you can put some furniture in your house and see what a red chair might look like against your white walls, or whatever it is. So definitely looking at about ... The platform definitely enables us to do that stuff a lot faster than it would have been on the old site. Probably wouldn't actually wouldn't be possible in the old platform, to be honest. So we can definitely go after all that leading edge tech, including the visualize search stuff and other things.
Chloe Rigby (21:16):
So next question. "Would you mind expanding on how you achieved an increased add to basket performance by 23%, while optimizing 100,000 plus product images?"
Tom Hayman (21:30):
Good question. So we naturally saw an increase in basket conversion. So we replatformed last year in October, and that's mostly come up to peak. So it's awfully hard to understand, for instance, whether some of the traffic we're seeing, for instance, was definitely peak orientated or the new platform. But pretty much, we did quite a bit of analysis at the time, pretty much it was a massive step change across the whole platform. You could literally see it go from X to Y overnight of the amount of traffic and amount of users and customers that we can then take on board, not just the basket.
Tom Hayman (22:17):
So pretty much overnight we could see that change. I don't remember the numbers exactly, it's quite a long time already. And then obviously this year, since Fastly, that's just gone even faster in terms of speed improvement for the site. We've seen a massive uplift on that presentation, including all the images and stuff. Generally speaking, they're all backed off now onto Fastly, they're all served by Fastly. They're all optimized. They all get presented onto the website very, very quickly. All of that stuff, every little bit you take off in terms of improvement in the performance, it all helps conversion, and people will seek completing their journey on the site.
Chloe Rigby (23:02):
That's great. Next question. "Can you tell us how you enabled the developers to move quickly in AWS? How did you enable them to deploy rapidly?"
Tom Hayman (23:14):
So that was probably the slowest wheels at the beginning. It took us quite a few months to build pipelines, to educate ourselves, as well as obviously the development teams, how serverless works in AWS. At the time we got a lot of support from AWS because for them, it was awfully crucial to help one of their customers, but also to make sure that this could actually be done on this scale. So it did take probably around about three months where we had to build to the tooling, the automation in order to go fast. So I hope I've answered the question. We worked very closely with the development teams to get to a place where we could make change from a dev environment, all the way to production in a seamless way, and a quick and safe, automated way.
Chloe Rigby (24:12):
Sure, no that sounds good. Are PWA part of your future development plans?
Tom Hayman (24:20):
Chloe Rigby (24:20):
Tom Hayman (24:22):
Chloe Rigby (24:26):
Anything more on that?
Tom Hayman (24:29):
So that's actually coming ... That's actually been delivered in the last week. So yeah, that's done, sort of next, I think.
Chloe Rigby (24:39):
Brilliant. Okay. And how do you manage the collaboration between platform engineering and the development teams?
Tom Hayman (24:49):
Well, so a lot of us techies and stuff, we pretty much use Slack, pretty much every day, every minute of the day. So Slack is definitely a conduit for messaging. We also use all the other things as well, like Zoom and stuff. We use a lot of things, the agile framework sprints. So every two weeks, we have a wash up all the different teams and they show what they've done to everybody else. So that's a good way of demonstrating what you've done, and to get some feedback and stuff like that.
Tom Hayman (25:22):
Platform engineering also have a similar thing every month. There's also something called, like a specifically called a [Super Sprint 00:25:30] highlights, where pretty much anybody can do a sort of demo of something that they've discovered or something that they found, and tell the rest of the tech community or outside of the tech community. So we try and have a lot of sessions like that. It's obviously difficult at the moment because a lot of people are obviously suffering Zoom fatigue and stuff, so you kind of have to manage it a little bit, the team manage it themselves. But yeah, we do a lot of stuff like that. And I think beforehand, pre-COVID, we used to have a lot more meetups and things like that in person, and that might be at the park and stuff or whatever. And that would obviously help with collaboration across different teams, including platform engineering.
Chloe Rigby (26:10):
That's great. So before COVID, you had third party sellers on your platform. Was this successful that you're going to do that again?
Tom Hayman (26:22):
I don't think we did it to a massive extent, but I think it's being talked about. I don't really know, to be honest.
Chloe Rigby (26:30):
Okay. So by doubling or quadrupling releases per month, do you mean much more of the same kind of releases or the ability to release smaller changes more frequently?
Tom Hayman (26:46):
Exactly. So having that smaller releases, have a shorter feedback loop, so you can make a tiny release. And if there's a problem, obviously the blast radius is often minimalized as well. But that'll give the teams the ability of the business to go a lot quicker, obviously with things like the feature flag and stuff like that. You can have that blast radius, for instance, even smaller. You're only showing that change to a subset of a people. Obviously a lot of e-commerce and other social media companies and stuff do the same sort of engineering.
Chloe Rigby (27:18):
Okay. And Brad asks, "Can you explain the stack a little bit more and also you chose the stack?"
Tom Hayman (27:27):
What's the last bit, sorry?
Chloe Rigby (27:29):
And how you chose it.
Tom Hayman (27:32):
So the stack is made up from pretty much everything serverless within AWS. So that could be S3, so that is, strictly speaking, serverless. DynamoDB, API Gateway, Lambda, obviously. We use other services, SQS and SNS, as well as things like RDS and the serverless versions of audience as well. And we do have some technical debt where some stuff is still on ECS, so container services, but that's pretty much going to go. Been working on getting rid of all that stuff to be 100% serverless. We're nearly there. We're not quite there yet, but we're nearly there. So that's pretty much the stack. And what was the last part of the question, sorry?
Chloe Rigby (28:17):
And also how you went about choosing it.
Tom Hayman (28:21):
So we're a little bit prescriptive, outgoing service. So we basically said to the development teams, "We're going to serverless. Let's just give it a go." And pretty much the feedback was like, "Yeah, let's try it." There was obviously some challenges back about ... Some people in the tech community can be, or in the tech world, as we obviously call it, can be opinionated about stuff, which is good. But some people wanted to go after cubes, or ECS or whatever, so some of that.
Tom Hayman (28:55):
But we basically said, "No, we want to go serverless." And also it did take a couple of months, around about three months, I think, for people to start winning hearts and minds and stuff like that, about the benefits of serverless. You don't have to manage any of this stuff. It's all within AWS. It gives us the ability to go faster and to work on things that are important to us. So that was for those reasons really.
Chloe Rigby (29:27):
Okay. And with so many releases in a month, was it Agile methodology that you used or another?
Tom Hayman (29:36):
Yeah, so agile, completely agile. Yeah, some people will and some people scrum pretty much. Yeah, agile was across the whole board.
Chloe Rigby (29:49):
And do your digital teams within your IT team, are they separately and how closely do they work together?
Tom Hayman (30:00):
So it used to be something called, at the time it was digital and core technologies. And as I said, we used to have sort of customer journey based teams. I said checkout, and basket teams and stuff like that. We collapsed that. We went through a different route now where it's domain driven teams. So we have a selling domain, for instance, a platform engineering domain. We've changed that recently. And I think that's a natural progression to accelerate business outcomes.
Chloe Rigby (30:36):
Okay. A technical question here. Is Fastly as CDN and do you host via AWS? And did you look at other vendors before you chose Fastly?
Tom Hayman (30:53):
So the answer to the questions are yes and yes. So with Fastly, we used it as the CDN. We used the verification one, a feature flag tooling and other sort of dev ops tools that they've got. So that fronts the sign, accelerates our code and images and stuff. And then beneath that, it's in AWS. We did look at other vendors at the time when we first started this journey, but we were already using another CDM, but they were bought out. And we basically found, through a few PAC's that we could essentially accelerate our code and go faster via Fastly.
Chloe Rigby (31:43):
And what impact did the front-end media design have on the performance of the site?
Tom Hayman (31:48):
So you went for SPA, so a single page application. And that generally means it can go faster in lots of ways. For instance, if you've got some internet connection or you've lost your connection momentarily, the front end basically enabled the rest of the systems they're linked to, to go ... Some of these are, again, too technical. To go faster. So yeah, I think hopefully I answered the question. Sorry if I didn't.
Chloe Rigby (32:24):
Yeah, that was fine. That's great. So another question about the platform. "Which e-comm platform did you go with and what was the RFP process from start to finish?" Maybe briefly.
Tom Hayman (32:43):
So what with the previous one?
Chloe Rigby (32:47):
What e-comm platform did you go with? And what was the RFP process from start to finish?
Tom Hayman (32:53):
So for the old e-comm platform, that was chosen before my time. So I don't know the answer to that one, I'm afraid. The new one, was obviously built from the ground up, which we did ourselves. So it's our own version of an e-commerce platform based on service technologies.
Chloe Rigby (33:14):
How has this affected your total cost of ownership?
Tom Hayman (33:20):
We actually looked at this quite a lot recently. It's this hard one to try and analyze in an analysis paralysis. Excuse me. So it's helped us to go to deliver quicker, to move faster, to have less management overhead in terms of infrastructure, release quicker. How we don't have to do many of these massive patching cycles or upgrades and things like that. So we can work on the things that we'd like to do and that's to make change within the platform as quick as possible. So to gauge that stuff, I'm not 100% sure how much more it would've been if we used different technologies for instance, because we don't use those technologies.
Tom Hayman (34:08):
But compared to our old e-commerce platform, as I said, we can definitely see that we spent ... It would take 15 people that were non-technical, for instance, and 10 technical people to make one change per month, whereas we can do, per squad or team, per eight or 10 people, and make that change themselves with some governance, all the time. And we don't have these massive production cycles where we've got to take systems down, and patch them and upgrade them and swap out hard drives, whatever. We really have the freedom to go as fast as we can possibly go.
Chloe Rigby (34:43):
Chloe Rigby (34:47):
And final question for the moment. "How did you manage the support for this and how quickly did the team land serverless?"
Tom Hayman (34:58):
Yeah, so I was just saying, it was quite slow to begin with, and we definitely set back the rest of the time. So I don't be too disparaging to them, but at the time, and a couple of years ago, serverless documentation within AWS was quite weak. So we needed a little hand holding from their enterprise level solutions architects and serverless specialists. So they would come and visit the offices, they'd do chats, they'd walk us through examples. We asked and we'd have an instant feedback loop because they were there. So we had a lot of these sessions to begin with, and that really helped the teams understand the tech.
Tom Hayman (35:36):
Because obviously when any tech team takes on new tooling, or new software or new technologies, even if you're doing it at home or whatever it is, it does take time to understand troubleshoot failure and do stuff. So yeah, it was definitely slowly to begin with in summary, but as we've been using it for the last year or so, we feedback now to AWS about improvements to that, so roadmap. We have a quite good, very healthy relationship with them about where we think there are deficiencies in serverless or places that we would like to see improvements, and they're very, very understanding. Some of those things have come to life recently.
Chloe Rigby (36:23):
Brilliant. Well, that brings us to the end of the Q and A session, Tom. So I just would like to wrap up by saying thank you very much for being with us today. So I'd like to thank our speaker, Tom Hayman, head of platform engineering Dunelm. And I'd like to thank you as well, our listeners, for tuning in today and for your questions. And as always, we'll follow up by email with the link to the on-demand webinar so you can watch it at your convenience. So thanks all for taking part in this webinar and goodbye.