How Washmen built a self-running quality and claims system

Welcome everyone to Work Innovation Summit customer showcase. Today, I'm joined by twenty twenty six AI breakthrough award winner, Washman.

And my name is Norbert Durko. I lead the global Asana practice at CloudFresh as the EMEA solutions partner of the year and a platinum solutions partner of Asana.

So we started working together with Washman last year in March when we basically teamed up to solve the and translate complex facility needs into practical AI workflows.

During our discovery workshops, we actually noticed that Watchman reached the moment where operational volume started to outpace human capacity.

So we worked together with the team to actually completely redesign their architecture.

And our core conviction, real core to success, was actually that agentic enterprise must be built intentionally.

We really need to be aware of the moment when we actually hand over the work to AI, and then how do we use the human judgment at the right time.

And the goal of the session today is actually to give you replicable blueprints that will help you eliminate the manual overhead as much as possible.

And I'm joined today with Amir Ali from Washman to show you how we actually build this. So Amir, over to you.

Thank you Norbert for the very warm introduction. Thank you to our friends at Asana and CloudFresh for giving us this opportunity. We're very proud to be here. My name is Amir. I'm part of the operations team that maintains and builds Asana AI workflows at Washburn. We're a tech enabled textile care company that offers various different services ranging from everyday laundry to shoe care, bag care, and luxury couture garments as well. Our workforce is over five fifty employees and we've all been using Asana since twenty sixteen, but it was only in the last few years where we really realized the potential and capabilities that Asana has, and thank you to CloudFresh for bringing that to light as well.

At this volume, the volume at which we process about tens of thousands of items every single month, At this volume, consistency can either become a liability or a competitive advantage, and Asana has certainly made it our advantage.

Alright. Thank you for sharing that, Amir. And before we dive deep into workflows, I want to talk first about the operational reality, right? Because in the case of Washman, decisions changed based on who was in shift. And this was really the discovery that we had, thanks to the work that we've done in terms of really understanding how the team has been working before.

And we actually realized that the real bottleneck wasn't really the physical labor, but it was actually the cognitive load of checking histories, of checking risks, of actually understanding all the context around how they worked with different items and pieces of clothing.

All of this was actually done manually.

So, Amir, I'd love to dive deeper into how did this pain actually manifest for Washman and why was building an autonomous system the only way out of it?

Yeah, absolutely.

Two of our largest service lines, Definery and Shukar and Bad Care, the entire infrastructure and ecosystem is entirely built on Asana. So each service line has its own Asana project board. Now at the volumes at which we operate, a lot of knowledge is always trapped inside individuals heads. People who are experienced in the laundry are technicians, they have a lot of expertise and knowledge which remains stored inside their heads and this was proving to be a bottleneck and something of a challenge for us, particularly concerning claims because we would revert back to a senior person, an expert to give us advice whenever something would go wrong and if this person was not available we would find ourselves kind of wondering what to do next.

Asana and CloudFresh helped us really solve this problem and gave us an idea on how to make this knowledge available easily and accessible via Asana and the AI teammates.

Previously claims would take two, three days to resolve and we've managed to bring this down to about six hours on average.

An moment came when we realized that Asana needed to be a safety net, a database for us and something which warns operators before they make a decision and go live with it.

Yeah. No, that that sounds really interesting. And you know, one of the things I remember initially was just realizing how much information people actually had to hold in their heads in terms of the volume, in terms of you know, the value of different items that you're handling.

So the stakes for Watchman have been always very high to make this right, to make this work.

So in terms of how we actually were able to build this, it wasn't really just one standalone tool or one standalone solution.

Throughout our discovery, we understood that we had to develop a system that is actually working with a handshake automation.

And what it means in practice is basically a system that is very highly connected, a multiplayer ecosystem that works autonomously. Once one task is completed, then the work is actually handed over to the next one.

So, Amir, can you walk us through this multiplayer network that we've built?

Yeah, I'd love to. We're very excited to go into these blueprints. Now we have four AI agents who are working in complete harmony with each other and giving us the optimal result that we are looking for. There is no human intervention, there is no human oversight unless it's absolutely needed. The first one is the item intelligence expert.

This expert goes in and sets and verifies the approximate retail value of each item. What that does is allows us to have an idea of what kind of item we're handling. It could be a thousand pounds, it could be ten thousand pounds, and this will also dictate the way and the level of degree of care we go into in handling and processing this item. The second Genetic AI is the quality guard, which offers thresholds against risk and allows us to ensure that risk is managed in the most efficient way. The third one is a damage and risk assessor where we have a history presented to us. If you've sent us a Loropiana jacket in the past and it got damaged, the next time we receive the same jacket from you or from any other customer we have a history that a damage happened with this item in the past and how to avoid repeating that same mistake again.

The fourth one is one of my favorites which is the resolution analyst. This is saving our customer experience time customer experience team a lot of time which they previously spent on drafting claim histories, case studies, going through all the details that took place from A to Z, reviewing photos and CCTV, it's taking away all of that effort and doing the work now and simply drafting a response to the customer experience team. So it's helping us save a lot of time and effort there. The impact is that we have a soft running quality control loop which is replacing manual decision making.

Yeah. Wow. That sounds really very sophisticated, perhaps even advanced, but today of course we want to break this down into three actionable blueprints. So the first one is going to focus on the facility sync, right? The second one is the risk screening, and last but not least, the claims resolution.

So let's talk about the first blueprint that we have to share today about the intake. So when it comes to obviously handling a very high volume of information, very high volume of items being processed through the laundry, when we started working with Washman, we realized the first thing that it was really crucial to actually standardize the entry process.

So how do we actually put this whole data together so that it really builds a very solid foundation for the AI?

And we also realized from the other angle that we cannot actually automate chaos, yeah, because the results would really not actually help you, but will probably bring you down, you know. So the way how we use this blueprint in terms of Asana technology, that was leveraging the IP API integration, the rules, custom fields, and of course the AI teammates.

And because at CloudFresh, because at Washman, every operation has an intake moment, we really knew that it had to work properly. Otherwise, like I said, AI just wouldn't really deliver those results. And then we also had another challenge, because Washman have had developed the in house solutions to actually process and properly manage all the items that are coming into the facility.

So we, of course, didn't want to disrupt the habits and how the floor stuff used to work. So what we had done, we actually leveraged the API connection. We built a real time bridge so that people working there wouldn't even have to really think about Asana. We had this real time connection, and here I want to again ask Amir, in terms of walking us through, what actually happens when the item enters the washman system, and what does the item intelligence expert do for you?

Yeah, absolutely. So as you mentioned, we're processing tens of thousands of items a month in the finery itself.

This is maybe thousands of items each week and at this scale when we're tracking every item from intake all the way to dispatch, it's very important for us to have that transparent visibility. We have our own internal tech, our entire facility runs on that tech infrastructure. What we did is connect an API to our own software and at the sorting stage where the bag is open and someone physically removes an item and takes photos and looks at preexisting damages and stains. All of this information is logged and an automatic Asana task is born on the project with pre populated information.

So you sent us the Iloropiana jacket we talked about. Now I can see the brand name. I can see the color, the item type, and even the measurements recorded before we process them. What this does is it creates all this information in a perfectly readable format for the AI, which will help us to get the information that we're looking for.

A smart rule is then triggered, which allows the item intelligence expert to go and search online on physical retailers like Unas, Farfetch, Bloomingdale's. It will find that same Laura Piana jacket and fill out the approximate retail value field, letting me, the team, the laundry team, and the customer experience team, everyone know what is the value of this item. Which again is very important for us because we need to know are we handling an item which is a regular everyday item of laundry or is it something which is very delicate, high value, expensive and creates more liability and exposure for us.

Yeah. No, that's that's super interesting. And I think this really leads me to, you know, trying to maybe inspire a little bit in terms of how do we actually, you know, build something like this. Right? So the first step, just like we've done with Washman, is to really identifying the highest stakes data point. So for you guys, it was the value of an item. And then make a custom field in Asana, which is going to be auto populated with the help of the AI studio or with AI teammate, just to make sure that all the subsequent automation is basically able to gather the right information, and of course, feed the right insights into every stakeholder in the organization.

So once we have the intake process done, the next question is, of course, how do we actually proactively prevent risks? Because that's another thing. So yes, we have a good understanding what sort of product we're handling, how do we have to take care of it, and so on, but at Washman, we realized that every operation actually had a pre work gate.

So there is this culture, or there is this, I would say, way of working that every time we want to ask, could this go wrong? Like, what could go wrong?

And we actually wanted to embed the same kind of logic with the agentic AI with the help of the teammates. So when it comes to the blueprint number two, the technology used from Asana was the AI Studio. We're actually two teammates working in parallel, as well as the Slack integration.

And as CloudFresh, we of course, you know, had a look at the process, how you guys worked, and given advice that perhaps using two different agents instead of just over indexing on one would be a better solution because of the complexity. And I'm sure you'll talk more about it. But it was really the goal for us to build a very solid bridge between the frontline assessors actually looking at the items and their managers responsible for all the decisions.

So, Amir, walk us through about the human handoff process and why did you actually split this work into multiple different teammates?

Yep. Thank you Norbert.

Here is where I'd like to give a special mention to Asana who have an amazing product as it is, but also CloudFresh because since our partnership began a year ago, you guys have always given us really great advice and this is a prime example because our idea here was to have one agent kind of try and do everything all at once and you advised us to maybe have two because in our experience, and I'm sure everyone in the room has experience with AI at this stage, has used Claude and OpenAI's chat GPT, and we find that these agents and language models have the ability to hallucinate and break. So

to give ourselves the maximum chance of success, we decided to go with two agents running in parallel with each other in complete harmony. So on the Asana board, after sorting is complete and an item moves over to assessment, the assessor will log here how do we clean this item, how do we press this item, how do we fold, all the special instructions provided by the customer are logged into a form and then they have to set a custom field called assessment as done. So once that trigger is made, this then starts these two agents, the assessor and the risk and quality guard. Both agents are now looking for has this item or a similar like for like item come into the laundry facility before?

If it has, how did we process it? Did we follow the brand's care label instructions? Was anything damaged? Did anything go wrong?

Did we lose the item? Were there pulled threads? Everything. So both these agents are now focusing on whether a claim has occurred in the past, identifying risks and making the laundry team and us aware of these risks.

Now how do they make the laundry team aware? As soon as something is caught by either of these two agentic AI agents, they will shoot a Slack message directly to a dedicated channel. This is built through a Slack integration with Asana. They will tag the laundry team, the most senior person in the laundry who is one of our managers gets directly notified so that they can intervene before a smaller problem becomes a larger problem.

And it's all like warning triangles and you know, it's prompting them to make a decision fast and act on it.

Right. That's that's really interesting. And of course since you have so many different people, so many different teams and stakeholders involved, is there anything else that I can share about the Slack integration and like what are the other benefits that you're seeing?

Regarding the Slack integration, I think it's an immediate message that gets shooted across over to Slack. All the relevant departments and teams are tagged and it requires an operator to take an action. I believe in, so far this year alone, we have prevented fifteen high risk incidents from happening as a result of these agents being actioned and put into motion.

Wow, this is a real life example of how you guys have saved probably thousands of pounds?

I imagine so, yeah.

Thousands, yeah. No, that's really great.

So let's talk about like how can we implement something like this, right? So just like we've done with Washman, the first step is certainly trying to find where failure history lives in your organization.

So whether those are customer support tickets, spreadsheets, some other notes, perhaps old emails. So we have to take that information, properly structure it, and pull it into Asana.

The goal here is basically to make this information machine readable because, well, in the case of Watchman, it was pretty simple. AI can only protect what it knows about. That's why it was really very critical for us to have AI trained on all those previous cases that you guys have had in in the past.

And, yeah, I think now that we've covered, obviously, the the intake process, the the risk management, let's talk about the third aspect, which is claims resolution.

Because obviously, it sounds like we have a really solid foundation for the intake for proactive risk screening, but sometimes real life is different, and things do go wrong.

Whether that's like an SLA dispute, whether the order is late, or there is any sort of other dispute that you have with the customer, This is something that we have observed at Washman when we were doing our initial discovery, that sometimes the claim resolution process would actually take days. And most importantly, there was literally no real time visibility for the leadership.

And by implementing the blueprint number two, Washman was actually able to solve both of these issues. So when it comes to technology, again, we use the AI teammates, the custom fields, universal reporting, as well as the portfolios, which is something that we're mentioning for the first time so far.

Amir, can you walk me through what actually happens when a claim arrives today?

Yeah, absolutely. And as you mentioned, in the eleven years of us doing business in this industry, something we realized on day one was that things will go wrong. We're dealing with fabrics and these fabrics are being handled by human operators and you can put in several layers of defense but things will always go wrong. This particular AI teammate is one of my favorites because it has saved me a lot of time and our team as well a lot of time because previously we would spend hours and days trying to compile a full case study of what's gone wrong and what's happened. Looking at how we picked up the item, how was it received, the pre existing condition, any manufacturing defects.

All of this information, a lot of words, lot of photos, lot of CCTV and by the end of it you start despising garments.

But this particular AI agent has helped us to eliminate and alleviate all of that workload because now it goes back to the information that was collected in blueprint one. It looks at the information that was collected in blueprint two and it then closes the loop by creating a case study and a response which is generated for our customer experience team in line with the customer's lifetime value with us, their average spend, their order frequency, the facts of the case, so we make an impartial decision. All of this work is being done and triggered through a smart rule that we have in place, allowing the AI teammate to create a response which our customer experience team can send over to the customer. As a result, our claims have gone down, the claim resolution time has gone down from three days to six hours, which is a very significant time and cost saving for us.

Wow.

That's really impressive number and I did some math, it's literally twelve times improvement. Yep.

But one thing really stood out in terms of what you were talking about, the leadership visibility.

Maybe you can tell me a little bit more, how did Asana and portfolios and universal reporting in particular help you to improve that part of the org as well?

Yeah. So because ShoeCare, Bagcare and the Finery, they all operate independently on this Asana board, we have complete visibility as a result of Asana portfolios and universal reporting to really monitor and track all of our KPIs on a daily and weekly basis. In the past I remember we would need to ask humans for a manual update on what is our damage rate or missing rate, overall claim rate to make commercial or operational decisions, and now that information is so easily visible on Asana portfolios and universal reporting that leadership can just go in, look at the metric they want to see, understand the data, make a data driven decision, and it's again saving us a heap of time because we don't have to ask anyone for a manual update anymore.

Yeah. No, that that definitely sounds like the setup made a big difference. Of course, I mean the metric that you mentioned itself is really impressive.

Twelve times improvement, I think it's really something we probably didn't even expect, right, when we just started working on that. So how do we actually implement something like this for any other organization?

I think the key takeaways for us was really about understanding and documenting the resolution logic that you currently have within your organization.

We need to know what is a valid resolution, what is a partial, what is a rejected resolution, and really, after we properly map that, we're able to encode this logic into AI Studio rules, as well as the AI teammate behavior.

And after, obviously all of this is done, after this is set up, we're able to build a similar portfolio view and universal reporting for the leadership as well to understand the sheer volume and the quality of work that's being done in the organization.

And the setup we're talking here today, of course, is specific to Washman, to the laundry business, but this setup is extremely transferable. Right? So no matter whether you're working with some warranty claims, service recovery, escalated tickets, I mean anything. Once you map that process out, you can properly embed that, and then AI will be able to build you the case where a human being basically makes your final decision in terms of what is the ultimate resolution path for that sort of claim.

And that basically brought us also to, I would say the last point in terms of understanding that with the agentic enterprise that we built within Washburn, it's a real life example that it's no longer just a future concept, right? That it's really something that we can build now, that we can build even within an industry which we probably would think is not a right fit for Genetic AI if we don't dive deeper. But in this case, it's such a good example that it can actually be done already.

And just to summarize, I think, three blueprints that we covered, the agentic enterprise basically has to start from a process that drains time and really relies on human memory. So we're not starting from something super complex and just trying to build it and fix it with AI, but we're actually starting from something very simple, very repetitive, that can be fixed very easily. And then we can, of course, scale the success just like we've done together with Washburn.

Amir, any final words from your side?

Yeah, thank you Norbert.

At Washman we believe in constantly innovating, continuously improving, and trying new ways of working. I believe by adopting AI, we are reshaping the way the business is run and how we operate. And like you said, in our history of trying to like find new ways of working and with this wave of AI, our learning is that if you start with the more repetitive problems rather than the more complex ones, you have a better chance of success. We were trying to solve too many complex problems and that wasn't quite working out. But when we looked at things which were consuming additional time or which were mundane tasks that needed to be done and we tried to tackle these, we saw a much better result. Compiling case histories for example, like I mentioned the case studies we used to create, two days for an agent to create a case study just for someone to read and understand what's going on was too much time.

So we started with this repetitive kind of mundane boring task and we you know offloaded that to AI and the result is time saving, it's cost saving and it's given us a lot of more space to operate and think. For people it's very important to think, I think AI being leveraged in the right way does not kind of suppress judgment but it creates way for better judgment and more consistent judgment.

That's our philosophy and just a final thought from my side is to let systems do the tracking, the drafting, and let people do the thinking because that's where we're really good at.

Yeah. No, thank you Amir. I I couldn't agree anymore. I think I really like everything that you said really resonates with with us.

And of course I want to really inspire the audience today as well, to actually start building something similar. And the goal here is obviously not just replicating one by one, you know, what we've built with Washburn, but actually starting from answering a very simple question, like where does this manual pattern exist in my operation, in my organization?

And once you understand that, then you're basically leveraging these blueprints, you can very easily start fixing those problems, just like we're done with with Washman. Thanks everyone for listening to our part.