How do you prepare for the Welsh HHSRS hazard rollout?

What are the 29 HHSRS hazards Welsh housing associations must monitor from April 2026?

The Housing Health and Safety Rating System (HHSRS) defines 29 specific hazards in residential property — including excess cold, excess heat, damp and mould growth, falls, fire, food safety, asbestos, lead, radon, electrical hazards, structural collapse, noise and crowding, among others. From 1 April 2026, under the Renting Homes (Wales) Act 2016 provisions on fitness for human habitation, Welsh registered social landlords are expected to monitor for all 29 from the first day the duty applies — not the phased rollout England received under Awaab's Law, where damp and mould landed in 2024 and the rest follows through 2027. The poll opening this session showed only two of 50+ attendees felt fully prepared to monitor all 29; 43% felt confident on damp and mould but not the broader list. Wales is starting where England is finishing.

How do you categorise all 29 HHSRS hazards consistently across surveys, calls and complaints?

Tenants never describe a hazard the way the regulation does. They say "no heating, my boiler's broken again, my kids are sleeping in their coats" — none of which contain the word "cold" — and the same hazard appears worded a different way in the next channel. Keyword tagging misses these completely. Out-of-the-box generative AI tags them inconsistently, returning one classification on Tuesday and a different one for the same comment on Wednesday, which is not defensible to a regulator. The Guinness Partnership approach trains transparent, definition-led theme models inside Wordnerds, one per HHSRS hazard, so the categorisation rules are visible and the same comment lands in the same category every time. All channels — TSM survey, transactional surveys, CRM entries, emails, complaints — flow through the same model, so a damp-and-mould mention in a repair call is counted alongside one in a complaint form, not lost in a separate spreadsheet.

What is the inner-loop / outer-loop approach to closing the hazard-response loop?

The inner loop and outer loop come from Bain & Company's Net Promoter System. When a tenant mentions a hazard, two loops start: the inner loop makes that specific tenant whole (route to the right person, resolve the hazard, follow up to confirm the tenant agrees it's resolved); the outer loop is the helicopter view — identifying structural patterns across the inner-loop cases that need policy or service change. The clock matters in both — the inner loop because the regulator's response-time obligations begin the moment the tenant raises the issue, the outer loop because preventing the next inner loop is how you stop the queue building. Guinness operationalised the inner loop with a follow-up damp-and-mould survey that auto-reopens the case in their CRM if the tenant indicates the issue isn't truly resolved at root cause — not just whether the repair job was completed.

How do you identify HHSRS hazard hot-spots across your housing stock?

Once feedback is consistently categorised, segmentation does the work. The Power BI cross-table demo in this session — built on synthesised data, not real Guinness data — surfaced that 23% of all comments from the West region related to damp and mould growth: proportionally higher than every other ward in the dataset. That's an outer-loop signal worth a targeted intervention. The dashboard's "Key Dates" view then tracks whether the intervention moves the proportion down over time, comparing pre and post the intervention date both in absolute volume and as a percentage of feedback from the affected segment. Cross-tabulation works on any dimension you have metadata for — ward, property age, tenure type, vulnerability flag — and each cell drills through to the underlying customer verbatim so executives can read the actual tenant words behind the number, not just the count.

How do you handle Welsh-language tenant feedback in HHSRS hazard monitoring?

Bilingual feedback is the Welsh-specific wrinkle. Wordnerds' approach for bilingual datasets is to translate Welsh comments into English first using a translation step, then run the translated feedback through the same theme-based categorisation that handles English-only data. Steve flagged this as a pragmatic compromise: the categorisation models are deeper and more nuanced when working in English than when working multilingually, so translating first keeps the categorisation quality high. The original Welsh verbatim is preserved alongside the translation so the customer voice in the original language is still available to operational teams who need it. For a Welsh association with a substantial bilingual feedback flow, this is a configuration step rather than a separate product.

Full Webinar Transcript

Sarah Wilson: Hey, everybody. Good afternoon. I can see some people are joining, so sorry we're a minute late there. We had a couple of technical problems, but I think we've ironed them out. I'm your host for today, I'm Sarah Wilson, and I'm delighted to be joined by Hayley and Jonathan from CHC. Hope you can hear us all okay.

We've got people dropping in now, so I think it's all good. Loving the background, Jonathan, you're going to have to come off mute now to make a comment.

Jonathan: Well, I didn't think you needed to see my kitchen and dining room behind me because it's in an absent state. So I thought it best to pop something up to hide the shame.

Sarah Wilson: Yeah, we've had a variety of funny backgrounds, haven't we, Steve, over our time? Well, hopefully if our marketing lady's got a WiFi back, we're hoping to pop a poll up just to kick things off as people are joining. So do feel free to introduce yourself in the chat. We'd love to use this space to sort of create a bit of a community. Feel free to introduce yourself — whereabouts in Wales you're from, we'd love to know. I'm in a very grey and rainy Gateshead. Although tomorrow I'm going to Paris, so I'm looking forward to that.

Oh, the polls worked. Our marketing lady's back. So we're asking how prepared you feel, how do you feel about the sort of hazard response, regulation changes. Just be completely honest, there's no bad answers. I think everyone's at various stages, and that's what we always find. So do click and answer the poll and Steve will pick up on the answers a little later.

Oh, hey, everybody. Hi, Phoebe. Hey, Paul. Hi, John, Paul. Hey, Silke. Silke used to work with us at Transport for Wales. So yeah, she's a Wales Word Nerd.

I'll make a start because I'm aware we've got a lot to cover. I know that once Steve starts talking, it's really hard to get him to stop. So I'm going to make a start. Do continue to introduce yourself. We do ask that all the way through, you're obviously on mute, but feel free to just pop your questions in the chat as we're going along. We'll have a dedicated spot at the end for Q and A. But feel free to ask whatever's on your mind in the chat. We do commit to answer all of the questions, not always in the live session, but afterwards we get back to you with the answers if we don't get to it in the session. So ask away.

The number one question we always get asked is whether it's been recorded, and it is being recorded. So if you've got to jump off — you know, you get a parcel, the dog needs a wee — just jump off and you'll get the recording straight after the session.

I'd love to hand over to Hayley, if that's okay. Hayley's just going to tell you a little bit about why we arranged this webinar, how it came about, and how she hopes it'll be valuable. So over to you, Hayley.

Hayley: Great. Thank you so much, Sarah, and good afternoon, everyone. It's great to see some of you here this afternoon. Today we're going to hear about how Word Nerds have worked with one of the largest UK social housing providers to tackle all 29 HHSRS hazards and have built a framework that hopefully others can adapt and use here in Wales.

Following the Welsh government's announcement that new hazard requirements will be in place from April this year, we reached out to Word Nerds to organise this session. We know many of you are now deep in operational planning, making sure that you're ready to meet the new requirements from the 1st of April. So we felt that this was a really practical case study that hopefully can support you as part of those preparations.

So today's all about learning, thinking about how that could be transferable into your organisations. And I'm really pleased to hand over now to Sarah, Steve and Zoe, who are going to walk us through the session.

Sarah Wilson: Perfect. Thanks so much and thank you for setting this up. We're absolutely delighted to be here and thank you to so many of you for joining us and giving it your afternoon — we really appreciate it. We're just going to start with a very quick introduction, just so you know who we are.

You've got three varieties of us today. That's me — I'm the vegan with the obsession with leopard prints. So I'm Sarah Wilson. I work in the social housing space, and do everything sort of account management related. We're also joined by Zoe. Zoe works in the housing space within customer success land. So she's all about working with our housing associations, taking their suggestions and feedback and bringing those into our model. So she's perfectly placed to talk through some of the work that she has done so far with Guinness.

And then the man that needs absolutely no introduction, one of our co-founders, you might recognise him off the telly because he's just been on BBC's Only Connect. We were all totally in awe of his ability to find links between the most unrelated things — you know, like 17th-century poets and types of pasta. But it's his knack of really surfacing critical connections that you need. So things like "my walls are black" needs to be flagged as a mould hazard, or a cold-room complaint is an excess-cold categorisation, for example, and that's what he's going to be talking about today. So Steve's the head of our insight team. I think the Chief Scientific Officer title is probably a while ago. Steve, you're all about insights and innovation, aren't you, and everything to do with analysis of tenant feedback. So he's a great person to ask questions to.

Now, who are we? Just a really quick introduction. I'm assuming that most of you won't have heard of us. We are a customer feedback analytics platform and we really specialise in that space. That's a lovely picture from Asima Barbecue — very informal. So we have got some great names in the social housing space as you can see below. Brands like Raven Housing Trust, Sovereign, qro, bpha, and we also work outside of that space as well in retail, transport and finance. Transport for Wales is actually one of our oldest customers. So we've done a lot of work in the kind of Welsh space previously.

Now what you'll get for today. We've got a very short agenda — I'm going to keep my talking to an absolute minimum. So I'm going to set the scene on what we've heard that Welsh housing associations are facing right now, and why April is so important. Then I'm going to hand over to Steve and Zoe and they're going to lift the hood on how Guinness Partnership built a framework that covers all of the hazards, not just the ones that make the headlines, and what that actually looks like in day-to-day operations. Then I'll come back at the end, and I'll chat to you about what you can do next to prepare. Then we've got a good chunk of time, about 15 minutes of questions at the end. So get your questions in the chat as we're talking.

Right, so let's get into it. So where are we at now? From what I've heard, it's a lot. There's a lot to be dealt with. The Renting Homes Act is already in force. We know that hazard response duties are looming. And unlike England, who got that phased rollout for damp and mould — which started a couple of years ago and doesn't end until next year — the expectation in Wales is that you will be monitoring all of the hazards from day one. So that's all of them, from the start of April.

Now, we've got a real mix of job functions joining us today. And you're all coming at this from a slightly different perspective, which I just wanted to touch on. So if you're in asset and maintenance, you might be thinking what properties are at risk before the Section 91 kicks in. If you're in housing ops and customer experience, you're thinking, I've got hazard mentions all over the place in every different channel — not just the form of complaints forms. How do I bring those all together? And compliance really needs the audit trail that your ombudsman would actually accept, not just a folder of spreadsheets.

But here's the most important thing I'll say to you today. The good thing is that your tenants are already telling you about these hazards right now. And that's in repair calls, in survey verbatim, in emails, in WhatsApp messages. All of the signals are there, but the question is whether your systems are actually wired to pick that up.

Now, let me give you a specific example of how this goes wrong. I suspect a few of you will recognise this. So a tenant calls about a cold room in week one, they send a survey response in mentioning a damp patch in week two, they send you an email about a mouldy wall in week three. That's three channels, three different inboxes, and nobody's joining the dots. That's a real systems problem. And it's made worse because tenants just don't speak in the categories. They don't say things like "a Category 1 excess cold". They say, "no heating. My boiler's broken again. My kids are sleeping in their coats." And it's exactly the same hazard, but just described differently every time in three different systems and spreadsheets and maybe not even flagged as a hazard at all. And that is what's really keeping people up at night. It's the hazard that's been sitting quietly in your data for weeks and nobody's noticed it until it becomes a real issue.

Now, we're going to put forward the way that we find to handle this best. We're going to walk you through the system that we developed to help solve this problem. The people that are really getting ahead of this have stopped relying on manual tagging and they've looked at the processes and systems. So by the end of the session, when somebody asks what's our plan for April, you're going to have a real answer. And not just that we're looking into it — a specific defensible plan. And the best working example that we've seen is from Guinness Partnership, as Hayley mentioned. And it's a completely closed loop. Steve will tell you what that means in practice in more detail. But from the moment that a tenant mentions an issue, we go all the way from that contact right the way through to your compliance audit trail. So to show you how this works in practice, it's over to you, Steve.

Steve Erdal: Brilliant. Thank you so much, Sarah. And thank you all for joining us today. Lovely to see so many people here. Lovely to be here with my two favourite Wilsons. That includes the volleyball in Castaway.

Thank you all also for filling out the poll, which I think makes interesting reading. You should hopefully be able to see that on the polls tab in the chat. I think it's really interesting the split there. I think the first thing that jumps out is that only one person has said you feel well prepared — or two people now, so well done to the other person — that they feel fully prepared to understand all the different 29 aspects. I think it's also interesting to note that the highest proportion, the 43%, are people who've said we feel like we can handle damp and mould really well, but not the full list of 29 hazards.

And as Sarah said, that is the big difference between the rollout in England and the rollout in Wales. The fact that in England they were given this staggered start, they were given start-with-damp-and-mould and slowly over the course of three years get to the full list of 29. In Wales the starting point is you should be ready to categorise and understand all 29 of these hazards from April. So it is a big challenge and I think hopefully you can see from that poll that if you are feeling like this is a real challenge, a real area that you may need support in, you're not alone. This is something that very few people are feeling fully confident on.

We are taking our best practice from Guinness Partnership in England, who I think are really at that kind of far end of best practice. We're not presenting this as where you should be right now. This is very much best in class at the moment, but hopefully it will give you a sense of how you can get there too.

So what do you need in order to be fully prepared for these hazards coming in in just over a month's time? You need CPR. There's three key areas that we think are particularly useful for you to think about as you're looking at your response to the legal requirements that are about to hit.

You need to categorise — gathering the data together from all the different sources where a customer gets in touch. Bear in mind that it doesn't matter whether a customer talks to you in a complaint or in an email or on a phone call. As soon as they have mentioned this hazard to you, the clock will start. So capturing that feedback and then having a system where you know that a particular customer in a particular piece of correspondence has mentioned one of these 29 hazards. So you need a way of doing that.

You then need to be able to prioritise — based on that list of customers that are having this issue, how do you then go about understanding which ones you should be putting to the top of the list? Where are the really urgent things? And start to identify patterns.

At that point you then need to take a step back. So you need to deal with those individual customers that are having these issues — you then need to take a step back and think what are the patterns in this? What are the correlations, what are the things that we can do in order to ultimately improve customers' lives at a broader scale? So not just looking at the individuals — looking at that wider conversation. So those three points, the C, the P and the R, we're going to take you through today, and Zoe is going to show you how this looks in real life, how this looks in practice for how Guinness are presenting this information.

And we're so grateful to Guinness to show — you won't be able to see any of their actual live data, but all of the steps and all the processes that they go through, we're going to show you them as live. So let's get started with that first area of categorisation. When that data comes in, how can you go about categorising it?

We're going to start with the key ways in which people tend to do that. The first and most obvious is getting someone to do it — getting a customer-experienced professional, when they see something that mentions one of these 29 hazards (or might mention one of these 29 hazards), to say something. They've got a system where they can pass that information on to someone who can actually help that situation. That is your first line of defence. Anybody who suggests that that is not a necessary part of any kind of response to these hazards should not be trusted. This is a pivotal part and always will be. It is worth saying that that puts a lot of pressure on them. And if there is a slip, if something does slip through, what then? So thinking about this as the key part of the work you can do, and giving them an easy way of doing that, it's really important — but it's not the whole story. What else can you do to support them?

One solution that a lot of people will be doubtless presenting to you in any number of different ways at the moment is through tagging the data using generative AI. So getting generative AI to look at all the comments as they come in and say, there's an issue here with excessive cold or noise or damp and mould or whatever the issue is. This is obviously easier to scale, this won't get tired. But you will doubtless be aware of the challenges that generative AI has in this kind of situation.

The information it gives will be different every time. It is famously unreliable to the point where actually we're hearing now that in classrooms these days, kids are using — if a kid says an absolute outlandish falsehood, their classmates will say to them, "that's AI." It's become like the standard thing that they say. It's like — I don't know what it would have been in your day. In my day it was like you'd stroke your chin and go "yeah", and it just meant that's nonsense. These days kids will go, "Oh, I've got a girlfriend but she goes to a different school," and the other kids will go, "Oh, that's total AI." So not a great PR moment for this new technology when it's being used essentially as a synonym for nonsense.

You can get more robust versions of this, but it becomes eye-wateringly expensive very quickly to get it to check in a way that will ensure that you're getting the same data from it every time. And obviously if you're an analyst, if you're somebody who is having to present customer information — particularly presenting it back to the regulator — that idea of "if the data comes through on a Tuesday the AI catches it, if it comes through on a Wednesday the AI doesn't" — that's not acceptable for you when you're looking at trying to catch these really important hazards.

So what Guinness are doing with Wordnerds technology is to be smarter about the way that they're using this undoubtedly powerful tool. We're bringing together generative AI — so the ChatGPT, all that sort of stuff, a similar version of that as the front end, as the way in which we communicate with the person who's saying this is what I'm looking for. And we are pulling that data through and then using more robust machine learning algorithms in order to then categorise it in a much more secure and concrete way.

Now, this is something that I think Sarah alluded to at the start — that I do have a tendency to get overexcited and start talking for longer than anybody wants about machine learning. So really trying not to do that. But the basics here are that we are bringing together a much more numerical robust way of categorising the data, which means that you get the same result each time, but you still have the GenAI interface to allow you to say "this is what I'm looking for".

And that's what Guinness Partnership have done. They've done this exclusively in English. We know that you guys will probably be dealing with bilingual data sets more than your counterparts in England. What we would generally do there is to translate the data into English first, just because the stuff that we're talking about here is more in depth, and I think there's further to go with multilingual than the translation software. So we would generally use one of these tools to translate the data from Welsh to English first and then move it through this process. But once we've done that, the process then allows you to structure your data in such a way that you have a much more robust categorisation process to then play back to your seniors, your team, your customers, the regulator.

Zoe's now going to show you what that process looks like in practice in the real world.

Sarah Wilson: Great, thank you, Steve. Just for the record, before I jump in — we love hearing you talk about machine learning. Never stop.

Steve Erdal: In that case, allow me to do the longer version of the "not right now". Let's all meet Steve.

Zoe Wilson: Great. So, yes, I'm going to be talking a bit about the approach that Guinness have taken using the Wordnerds platform to overcome this challenge. But just before we do, to bring it to life a bit more, I just wanted to zoom in on what that categorisation challenge that Steve has started talking about really looks like in practice in relation to the 29 HHSRS hazards.

We're going to look at what this challenge looks like in relation to one of those hazards in particular, which we've chosen excess cold for today. We've got a few examples on the screen there of how this actually might materialise in real human verbatim. So we've got "My child sits shivering", "It's impossible to keep the flat warm", "I can see my own breath like refrigerator", "drafts, always freezing". A few different examples there of how people could refer to this excess-cold problem. And you'll notice that actually not one of them uses the word "cold". I always find this quite interesting that one of them actually uses the word "warm", but in a completely different context to how we might usually say "I'm really warm".

This is really just to illustrate the example of how the approach of keyword tagging really just doesn't go far enough. If we were just looking for keywords around cold and cool here, none of these examples would actually be categorised. In fact, if we were looking to categorise excess warmth, this example would be incorrectly categorised in that category around "it's impossible to keep the flat warm". Hopefully this just helps to illustrate that challenge a bit more, in relation to real pieces of feedback might actually materialise from your customers.

Furthermore, doing this manually across all of those 29 hazards just becomes near impossible. And that is why you need a specialist tool which can do this job for you, which can understand the nuance of human language and also categorise it robustly at scale. This is where Wordnerds comes into it. And using the approach which Steve has described, we allow you to train individual categorisation models, which we call themes, in order to group subjects together when people are talking about the same thing, but could be using different words and different language.

How that actually works in practice on Wordnerds is you simply describe what you actually want to categorise in your theme. On the screen there, I've just popped in a screenshot taken directly from a theme which was created on the Wordnerds platform for excess cold, where we've described what we'd want to categorise within that theme. This is really to make it super transparent. It's not a black-box AI tool. Everyone is on the same page in terms of what you set up the rules for each of your themes to do.

Sometimes that real challenge isn't actually spotting the individual comments. It's really having a system that can then track them at scale, measure them every time in the same way across all of your different data sources, and be consistent. That is where we aim to support with the Wordnerds platform and the themes that you can create.

So looking at the Guinness approach in a bit more detail, this is exactly what Guinness set out to do on Wordnerds. They set out to train one of our themes to categorise each of the 29 HHSRS hazards. This means that when they bring all of their feedback together, they have confidence that it will be categorised consistently in the same way across all of their different data sources.

So they have on one side all of their messy feedback across different data sources. They might have their TSM survey, transactional surveys, CRM entries, emails, complaints — it all probably lives in different places and different formats. So what they do to standardise that approach is put all of that verbatim feedback into the Wordnerds platform. Then because they have trained those themes, they've set their own categorisation rules in terms of what they want to define as each of the hazards. All of that feedback is processed and categorised consistently. And you can put an accurate number on how many people have mentioned each of their individual hazards.

So that really just centralises all the feedback and gives that consistent approach.

That being said, that step that Steve mentioned in terms of closing individual cases is obviously super important. But what that doesn't allow you to do is really understand the scale of those issues across your whole customer base in a way that you can actually rely on to use and understand trends and hopefully put those actions in place to improve things at the higher level rather than just for each of the individual cases.

So with this approach, Guinness have created that theme framework which has each of the 29 hazards and they categorise and track that each month. At the really high level, this is now integrated in their monthly board reporting. So they can get this straight in front of the executives who it matters to. And what I really like about their approach as well, is that they always include direct customer verbatim examples around each of the hazards. Just so to ensure that real voice of customer is always visible to executives and they can kind of see in direct words how customers are actually expressing how they're experiencing these issues in real life.

Taking this a step further, I'm just going to start sharing my screen briefly. What this actually looks like in best practice — and again, in terms of the Guinness approach — is moving beyond just those static reports which you might put out and produce at a regular cadence, and instead also having this live in a BI dashboard.

Those reports definitely serve a purpose, but the aim of this is to create a living, breathing resource that really puts that voice of customer in the room with the people who need it, so they can interrogate the reports themselves. And different stakeholders who have different needs can get those directly. We take all the data from Wordnerds, we categorise it consistently so that you have confidence that everyone is looking at data that has been categorised according to the same rules, and then they can investigate it themselves.

So what I'm sharing on my screen now, just to be super clear, this is a demo data set that we have synthesised. This is not real Guinness data, but it follows the same approach and it's a Power BI dashboard which we have created to present data from across the housing industry.

What we're showing at the moment on this screen is just a really high-level view which shows each of your individual 29 hazards. And we are simply showing the volume and the sentiment breakdown of comments about each of those hazards. So we get the volume number here at the high level — we can see, perhaps unsurprisingly, that damp and mould growth is our most mentioned hazard. But also moving through some of the more niche hazards we can start to see how often they are surfacing in the feedback as well.

So this is useful to a point. It's a good starting point to get that high-level overview of what the actual numbers are. But the real power comes in diving into this and using it as a starting point to view trends and particular issues that are coming in particular segments of your data. So we're going to move into that more shortly. But before we do I'll just hand back to Steve to talk a bit more about the approach in terms of inner and outer loop feedback.

Steve Erdal: Brilliant. Thank you so much, Zoe — that was fascinating and I think hopefully it gives you a really good sense of how we approach that first issue, that issue of categorising the data and ensuring that you have a system for figuring out which of these issues is coming up most regularly, how you go about establishing the relative sides of these different issues.

We now get to the point of what we now do with that customer complaint. And for a lot of the theory on this we turn to Bain and Company. You might also have heard of the NPS scoring system for feedback analysis. This is part of their broader NPS system. And they talk about when a customer complains — or in this case mentions a particular hazard — it should then set off two loops.

You have the inner loop, which is about making that specific customer whole. So getting that customer, solving the hazard, getting them to a solution that they are happy with.

You then have the outer loop, which is more about the helicopter view. You're dealing with all these inner loops, all these individual customer problems. What are the structural improvements around service, around process, around policies that will allow you to improve customers' lives going forward, and preventing as many of those inner loops from happening as possible.

So with each customer these two loops begin and Zoe's going to show you in a moment how Guinness have gone about closing both of those loops in practice.

But first a little bit more information about them. That inner loop we talk about sometimes being the prioritisation level of the CPR thing. What do you need to do in order to close that inner loop? Firstly timing. Now that would be true of any customer feedback loop. It's particularly true here because obviously a clock does start ticking when that customer issue drops.

You need to be specific. You need to understand what's going on with that specific customer. You need to ensure that that information is easily transferable to the person who can actually help them.

And you also need, in order to fully close that loop, you need to follow up. So it's about ensuring not just that the hazard has been resolved to your satisfaction, to the satisfaction of the professional who's dealing with it, but also that it's been resolved to the customer's satisfaction. And at that point you have then closed that loop. That issue has been resolved.

You are also looking to resolve the outer loop. Now, this is where we are bringing together all this data. We're identifying patterns. We are at that higher level looking at how can we continuously improve our overall process so we're making customers' lives better as a whole.

For that you need a couple of things. The first and the trickiest, I think, is probably united data sets. Customers — we always say this at Wordnerds, and we mean it in the warmest possible sense — customers are weird. And they will tell you offhand about a hazard in a phone call that's about rent, or they'll tell you about it in a really positive survey response. "Oh, but there's this — the wall's gone a bit black, but everything else is really great." So bringing together those different data sets and having a way of analysing them all at once is your best bet to identify the patterns that we're talking about here.

You need to be able to analyse it at different levels. So some people will be focused in on a specific issue or a specific group of issues. Others will want the kind of helicopter view to be able to see everything at once. You need to be able to present that data to them so that the right person understands exactly what they need to work through.

And that also requires a sense of transparency, a sense of this is not something that you are holding onto. This is something that you have ways of reporting back to your senior team, to your customers, and to the regulator.

And Zoe's going to now show us how Guinness Partnership has taken on the challenge of closing both the inner loop and the outer loop so that they can make life better for their customers. Pass back to Zoe.

Zoe Wilson: Great. Thank you, Steve. Before we move on to look a bit more at some of the Power BI demo, what I wanted to talk about — again, with massive thanks to Guinness for being so generous in allowing us to share their ideas and their approaches — is something that they are doing to tackle that inner-loop challenge as well.

Like many housing providers — I'm sure like many people on this call — Guinness send transactional repair-satisfaction surveys across all of their different trade types after a repair is completed. And this obviously includes any repairs related to damp and mould.

Beyond just analysing that feedback at scale, what they are now doing is isolating those damp-and-mould repair cases and they have split those out. And from those particular cases they're sending a further follow-up survey which is less about satisfaction to do with the job and more whether the customer is confident that that problem with the damp and mould will not recur. So this is really aiming to target those individual cases and understand how the customer is perceiving that in terms of, is this issue actually solved at the root cause?

And then to maximise the effectiveness of this approach, what Guinness are doing is integrating this with their CRM. This means that if a tenant indicates that they are not confident in the resolution of that damp-and-mould case, they're actually going to trigger that — the case will automatically reopen and that will alert the assigned case owner. What this means is the case owner can then follow up directly, understand why they are not satisfied that that problem is fully resolved and put actions in place to tackle that even more. Really creating that more closed-loop system to ensure those persistent issues receive the immediate attention that they need.

Just to round this up as well, Guinness do a really great job of connecting this up by integrating their data across systems across the different case owners. Which just means that that all flows directly into their internal Microsoft Fabric system. And those responses are rapid in the way that they need to be for the particular issues.

So just one other example of something that Guinness have implemented to really keep on top of these issues. The next thing that we're going to move into now is looking more at that outer-loop approach in a bit more detail.

Hopefully my screen will load. There we go. Where I am now is just back in our demo BI dashboard. What we looked at before was really more around understanding the high-level numbers of each of the hazards. But the question then becomes, how can you actually use that data in order to drive meaningful action and start closing the loop at that higher level?

So this is where it's really especially powerful to start bringing in the metadata that you have, so information about the customers who have given you the feedback that you are analysing. So again, everything you see here is based on data that has been loaded into Wordnerds, enriched and categorised into those hazard categories and then output into this Power BI dashboard.

What we've got on this view here is a particularly powerful view that we have called a cross-table. This is simply allowing us to compare two different aspects of our data and see proportionally where the issues lie.

So on this view at the moment I'm just set up, I've selected to look at all of my individual HHSRS hazards. So these are the themes which I've trained on the Wordnerds platform, and I've chosen to look at how these are being surfaced in each of the different wards on my data set. So we've got north, east, south and west. So I'm presenting the wards in my columns and the hazards in the rows, then looking at the crossover in the bottom table here. And again just to confirm, I'm looking at this at the moment as a percentage of my columns.

What I can straightaway see is which area each hazard is most prominent in. We've also integrated some heat mapping here, so at a glance I can see straight away this crossover between west and damp-and-mould growth is something that my attention is drawn to. So what this is indicating is that looking at it as a column percentage, 23% of all of my comments from the west region are related to damp and mould — which is good to know in itself. But the real power here is that I can understand that that is proportionally a lot more than how each of my other wards are talking about this particular hazard.

At each stage in our dashboard, this is really interactive. So now I can click through and zoom in on this particular crossover and see exactly how many comments this is relating to. And each time I can click and drill through to see the direct verbatim which has come from customers in that area about my particular hazard. Again, just really keeping that voice of customer front of mind whenever we're looking at this data from a more numerical point of view as well.

So this is where you would continue drilling through your data, aiming to understand more. Now I've used this tool, identify an issue. We don't have time to go into absolutely everything today, but you can see within this dashboard we've got lots of different pages set up and the aim is to give each different stakeholder the information that they need, in order to understand and hopefully put resolutions in place for different issues.

So what I want to show now is how it's really important to obviously identify those issues, but equally important to take action off the back of those and make changes, and then also track them to see if that intervention is actually making a difference that you'd like it to. So I'm just going to go into a final page that we'll look at in the Power BI Dashboard now, which we call Key Dates. The idea of this page is that it allows you to track how particular interventions that you've made might be having an impact in your feedback here.

What we're looking at is just carrying through that example. I've selected to show on this graph how the volume of comments about damp and mould has changed over my different time periods. And again I've just filtered down to look now at only data from those who are in the West. In my example here, because I noticed that issue that was occurring with my data in the West, I chose to make an intervention where in April I released a targeted campaign to all those people who had reported damp and mould, around how we were going to go about resolving those issues. And following that, what I can now use this to see is — has this had an actual impact in terms of how people are talking about damp and mould in that west region after the point that I made that intervention?

So in this example I can see that the volume of comments has decreased. But also importantly, I'm able to understand this as actually a percentage of all of the feedback from this region, to understand proportionally how that is occurring. Obviously this wouldn't be the final point. We need to make sure there's a real correlation between this. And again, I can continue diving through the feedback, seeing the verbatim. But this just allows you to not only identify the issues, but start making those actions and then also helpfully start tracking the impact and sharing this a bit more widely with the stakeholders who it's really important to.

So if anyone has any questions about the Power BI setup or whatever, please do let me know. I know that's been a bit of a quick whistle-stop tour today with some of those useful pages. But hopefully that's helpful to give an idea, and I'll just hand back to Sarah now.

Sarah Wilson: Fascinating. Thank you so much, Zoe. That was great. I love that Key Dates visualisation at the end. As always, get your questions in the chat — really anything about how that was implemented, how it's been set up, the views that she's shown, the categorisation models. Get your questions in and we'll answer those in a couple of minutes.

So you might be looking at this and thinking, you know, this is great. You've seen how Guinness have done it. You can see why it matters. But you're probably thinking, this looks great, but I really can't move forward with my data in this way.

When we speak to organisations at the start, there's generally three things that they mention. They are worried the data is too far back or too messy, that there's repairs in one system, surveys in another, and nobody owns that full picture; or they find it hard to get buy-in internally — so the culture is not there, perhaps in the more senior leaders. Changing how feedback is handled is really tough when your teams are already stretched to the limits. We know how tight budgets are and how small-resource the teams are. And also technical capabilities — so not everybody has the kind of AI infrastructure in house. It can take a long time to build this kind of solution, this kind of dashboards, and most teams don't have that resource. But honestly, you shouldn't need to have that resource.

So they're all genuine blockers, definitely. But I guess the one thing that I'd say is that you don't have to do all this at the start. So what we always recommend is that you start with one source. And that's what Guinness did at the start — they didn't do it all in a day. They didn't, you know, you can't build Rome in a day. They started with one data source, one team, and one question, and that was, what's hiding in the data right now? What can we be working on straight away?

If you are looking at this and thinking, oh, I'd love to give this a try on my data — we do offer a report that does exactly that. It helps you to see what your data looks like and how it's performing, sort of areas for improvement, deep dives. We take a slice of data at the start — so it's usually complaints or some surveys like the TSM, your transactional surveys — and we do the process always shown just there. So we take it, upload it, run it through the categorisation models, look at the prioritisation issues, and then hand you back a hazard identification report and that sort of prioritised risk list to make sure that you can see what you need to do first — what's going to have the biggest impact. And there's no real sort of massive procurement process or like a heavy integration there. It's a very straightforward way of getting started. It's sort of a clear way to get yourself in a current compliance position and show you what you need to do next. So you're feeling quite prepared for your next board meeting.

Now, before we open up the floor, we've got a couple of things to offer you today. So we've written a guide specifically for the Welsh housing associations, and it's all around HHSRS hazard response. What you need to do, what the timelines are, the specifics of it, and what systemic monitoring means in practice. So we'll share that with you after the call, and straight away after this call you'll get a full recording of today's webinar. So keep an eye out for that in your inbox.

If you're looking at this and thinking, I'd love to give this a try, I'd love to see what this looks like with my data — we are offering a launch offer for Welsh housing associations. So if you were interested in this kind of report, just get in touch with me and I can chat you through the whole process, what it looks like for you, the kind of timescales and that kind of thing. We do that for less than £5,000 because we tend to find that at that mark it tends to go down — it's quite easy to get through procurement, that kind of thing. There's not a great heavy lift to have it approved, so you can get it quite quickly and we can turn it around in quite a short time frame. So generally four to six weeks.

So with that all being said, I'm going to move very quickly onto the questions and give us plenty of time for discussion. So let me move to the questions. What have we got, Stella? Let me have a look. So we've got — can everyone see that there? I've managed to pop my chat box into a separate tab. I don't know how I did that, but hopefully you can see the question on the screen. So Alan has asked — "I'm interested in how Guinness ascertain vulnerabilities with each customer." And that's to Zoe.

Zoe Wilson: Yeah, I'll take this one. So, really interesting question, Alan, and I think there's a couple of different points to talk about here. So firstly, sometimes customers will actually — you'll know about the vulnerabilities that they might have from the metadata that you hold about those individual customers. So something Guinness have worked on recently is completing a customer census, which I know a lot of other housing associations that we work with have done as well. They have essentially asked questions around information related to that particular person living in that particular property. So for example, do you have children? Any physical health conditions, mental health conditions, etc., that they are comfortable to share?

I think it's important to make the point that none of this is mandatory. It's all just with the aim of helping to improve the life of that customer as well. Then using this actual metadata, they can connect that up with the records that they hold about that customer, the feedback that customer has given. And we can use that in relation to their qualitative feedback by segmenting that across larger customer groups as well.

So are there any challenges with which people with children for example face, which are more common to them than people without children? And we can do that using a similar approach that we looked at today, where we would use the kind of vulnerability categories as the segmentation, as opposed to the area, for example.

So that's one aspect. And I think the other aspect is that sometimes customers might actually tell you things in their feedback which you perhaps weren't even aware of about them. So someone might make a comment like "I've got two children and I've got no room for them", or "there's mould growing in their room, it doesn't feel safe", etc. This is where having a tool that actually analyses your feedback at scale like that is really important.

So what Guinness have done in particular is again used our categorisations on the platform, and trained models on each of those different vulnerabilities that they might want to be aware of. So children, and people talking about impact on physical or mental health, etc. That means that if people are telling you something in a survey response, you're always going to be aware of it and take that into account when you're understanding that feedback as well.

I hope that answers your question, Alan.

Sarah Wilson: That's a great question. Thanks for that, Zoe. I'll take this one. So, question from a fellow Sarah. How much effort is required from us for the hazard report and how long does it take? That's a brilliant question.

In terms of effort, we try and make it as light-touch as possible. So all you have to do is — we have a project agreement and then you send us your data. Generally one to two years of historical data. And it's just a CSV dump. So it's quite a straightforward way of getting it into us. And we do the rest. Well, when I say we, I mean Steve and his team, his brilliant team. I do absolutely nothing, which everyone's grateful for.

So what happens is we have a very short kickoff meeting where we discuss objectives. That's what we care about the most. What does success look like for you? Is there any areas in particular that you want to look at? Just really exploring it in a bit more detail. And then we also have a draft-report meeting where we'd run you through that data. And the rest of the time is — we hold a stakeholder workshop at the end. We take you through the journey that we've been through on the report. We allow people to ask and answer any questions ultimately. And Steve even throws in a quiz now in the most recent one, so that's been very popular. But it's just a chance to bring everybody together on the journey and show them why we did this, what the results are and what you can expect from this sort of piece of work. Which is a really nice way of ending the project.

In terms of timescale, so we generally turn them around pretty fast now. So within four to six weeks, we say — we obviously are at a busy time because it's the end of the TSM submissions. But yeah, we're standing at sort of report delivery between the start of April and the end of April now. Hope that answers your question.

Any more? Oh, good question from Michael. That popped up there. Sorry. Yeah, Steve, jump in.

Steve Erdal: Sure. Thank you so much, Michael. This is a great question that we get a lot of the time. We don't transcribe the data ourselves. We have kind of colleague organisations that are more focused on that. Or you may have, with your provider of your call data, opportunities to have transcription there.

We do analyse voice recordings of phone conversations. And they can be such a useful tool. Not just because obviously issues will be mentioned there, and because they're more wide-ranging, it allows customers to talk more about things that specifically matter to them. Or, as Zoe just mentioned, it allows them to — you know, if there are vulnerabilities or concerns that they have, they're more likely to talk about it in that kind of environment.

It also though allows us to look at how those conversations are resolved with the customer-service professional. Which types of issues are customers leaving feeling less resolved on, which ones are more likely to be long or short conversations, what interventions from that customer-service professional are likely to really satisfy and resolve that issue for the customer. So it's a really rich data source, and I think one of the great things about this sort of technology is it allows you to look at this sort of thing alongside survey, alongside email, alongside complaints.

So the short answer to that is yes. Do you see why I get the reputation for rambling on at things, being a talker.

Sarah Wilson: That was brilliant. Thank you, Steve. And just to add to that, Michael — something that we're hearing a lot in the last few months especially is that we can include things like door-to-door consultant — you know, if you're having door-to-door visits, consultations, customer-voice meetings, we can ingest any form of text but that can really be in lots of different formats. So not even just the surveys. You think of like the CRM entries, the chat boxes that you have — any form of text, we can have a different sort of source and pull that together. And I think that's where it becomes really powerful because you can see across the different sources how people are feeling, what kind of things, what kind of answers they're giving to these things. And that makes it a really powerful piece of work. But that's a great question.

I've got a couple more. So let's go to the next one. So — "is the perfect purpose of this software to give an overview of the data based on survey data, or can it be linked with incoming calls to the call centre, and can it distinguish which cases present significant risk to the residents?" I think that's similar to what we've just been answering. Steve, I see you're on mute now. Have you decided to leave that one to me?

Steve Erdal: I'm done for the day.

Sarah Wilson: Yeah, that's it. You're clocking out. No, that's a great question. So yes, it can be linked to any form of text. And that includes call-centre transcripts like we said. So it can work out which are hazards. And it can set up an alert system that can trigger — I think we say within less than five minutes you can get an alert of a particular hazard that's come through in the text. We can set it up like that. And we can't distinguish between the different sort of emergency levels, I suppose. But we can definitely tell when there is a damp-and-mould mention being made, for example, and then set up an alert system to the relevant person within the department.

Is that fair?

Steve Erdal: Yeah, it's a great question. I think the kind of process that we've described there will be slightly different for each housing association. Depends on the format that your data comes in. But I think what Sarah's described there is the best practice that we would look to work from.

Sarah Wilson: Great, thank you. The last question from Kevin. We should even be on time, so that'll — wonders never cease. "Is this AI system separate from what would be the reactive requirements for the WHQS element 1c? Only significant hazards fall into that category and have set response times. So I'm interested in how your system distinguishes those from general hazards." This is such a great question, Kevin. Sorry to jump in.

Steve Erdal: Yeah, I mean, I think this has been one of the real teething issues that we've seen in the English data set, which is — you're absolutely right that only significant hazards would fall into that category. And the issue that we've seen in England is, who decides what is a significant hazard. A lot of the time that is down to the policy of the specific housing association. So understanding what you would deem as a significant hazard, I think, is the first step there.

Once we have that information, though, I think one of the great things about what we're talking about here with the Wordnerds platform is you can train these small AI models to distinguish that. So you can say, I want one that's general — I want one that's just any kind of potential mention of something that might possibly be a damp-and-mould issue. And then one that is, I really want the specific — this is causing or exacerbating a health issue. This is an emergency situation. We talked about vulnerabilities — this is something that is affecting somebody with a vulnerability that pushes it to that kind of next level. And that prioritisation stage that we talked about, I think, is really kind of a key part of that.

So in terms of, does it distinguish — yes, it does, and you can have it at both levels. So you can get the broad and the narrow. But I think the key question here is, what do you decide significant hazards look like? And in a way that is then defensible to the regulator.

Sarah Wilson: Perfect. And look at that, we're bang on time, Steve. So thank you so much for a brilliant session, everybody. Really appreciate CHC setting us up, putting us in touch with you. We hope you've enjoyed it. I certainly have. I thought it was a great session. So thanks so much to the speakers, Steve and Zoe, thank you for your time. You've got my — you've got a little button at the bottom that says "register your interest", and that comes straight through to me. So I'm just sarah at wordnerds dot AI if you want to discuss any of this in more detail, absolutely no obligation. If you've got questions around the AI or really anything about the hazards, just let me know and I'll get back to you. But for now, we'll leave you. Thank you so much, everybody.