It looks, from people's messages, like we're getting sunshine from almost everywhere today, which I think is a personal best at a Wordnerds webinar. Normally someone's saying it's very cloudy where they are, so this is great news.

Before we start, we're going to put a poll on your screen. We just wanted to begin by seeing where you are in terms of Awaab's Law — how prepared you're feeling for it, going from "I'm feeling under-prepared" all the way through to "I'm feeling ready for all the phases", because there are three different phases coming in. We'll let that roll for a little while as other people come in.

We've got 172 of you in the audience at the moment, which is very exciting and slightly nerve-wracking, but it's wonderful to have you all here. We've got a lot of people getting ready for next month.

So, warmest of welcomes to what is one of the most-requested webinars we've ever done. We've got a mixture here of long-term customers right through to several of you who I'm sure have never heard of us before, so we thought we'd start with a really quick introduction for any newbies. My name's Steve. I'm one of the co-founders of Wordnerds. I am a recovering linguist, and my role is to work with our customers to ensure they're getting the most out of their customer-feedback data.

We are a customer-feedback platform. We work across a range of different sectors helping organisations automate their customer feedback. We were minding our own business in places like retail and travel and hospitality when housing associations like Town and Country, Guinness Partnership and Sovereign came to us and started saying, "we have this challenge with understanding our customers." Since then, housing has become an area that's very dear to our hearts.

Alongside me today, I'm delighted to welcome two of our finest Wilsons. We have Sarah Wilson, who is our in-house social-housing expert. She's been eating, sleeping and breathing the social sector for several years now, and she'll be taking you through a framework that will allow you to get ahead of Awaab's Law and track hazards individually and collectively. We're then going to hear from Zoe Wilson. Some of the customers on the call will recognise Zoe — she's one of our top customer success managers, and she's been working very closely with the Guinness Partnership on the practicalities of creating a data strategy and developing the data so that you can get the most out of it.

We're super grateful to the Guinness Partnership — I think there are a few of their representatives on the call today. Thank you so much, not just for the work you've done in innovating around Awaab's Law, but also for allowing us to talk about it to everyone here today.

One of the reasons Wordnerds got so involved in social housing is that, amongst all the areas we work in — retail, transport, travel, hospitality — this is an area where the work you do directly impacts people's lives. That's why we love it, and we know how seriously the vast majority of people on this call take that responsibility. In the face of this tragedy, everyone on this call is committed to ensuring that sort of tragedy never happens again. There's an enormous challenge in that, and a lot is being asked of you, but it's also a real opportunity to shape your organisation with the customer at its heart.

I want to stress at the start that we intend this as a dialogue. The truth is no one has all the answers when it comes to Awaab's Law — it's too new, and there are so many ways you can approach it. We're going to show you what the most data-savvy organisations we deal with are doing, but please do use the chat, ask questions, and share your own best practice. We have Vic in the background helping us pick out the most representative questions, and we're committing to answering every question — if we don't get to yours, we'll give an answer in writing.

So, in terms of the new law: your legal responsibility is changing, and it will now begin the moment you're told about a potential hazard. You'll be required to make an initial determination as to whether the hazard is potentially significant and whether it could be classed as an emergency. You've then got to assess that risk and take the tenant's personal situation into account — age, health conditions, disability — determine the severity, categorise and triage it, and then report on the solution you came up with.

One of the big shifts is that information must be captured from all communication channels, not just formal complaints. That could be a survey, live chat, a social-media post, a Google review. We always say customers are weird — they'll tell you things in places you wouldn't expect. In the same way you get positive comments in the middle of complaints, you'll also get potential mentions of hazards in places you wouldn't expect. All available information must be used.

So there's an immediate challenge around clear definitions — what does an emergency hazard mean, what does a significant hazard mean? — because those two things trigger different timelines. And you've got to show an effective system for how you're collecting, categorising and prioritising that information. If you're wondering what that rather sad-looking Frankenstein is doing on the screen: we've had this warning from the Ombudsman against landlords simply attaching these new legal duties to existing policies. It's not about having an existing policy and bolting the new HHSRS elements onto it, because that could create — in the words of the Ombudsman — a disjointed and ineffective "Frankenstein procedure". What they're looking for is a comprehensive, well-integrated system.

That challenge affects you regardless of your team. We picked out from the job titles you gave when you joined three main groups. If you're a data and analytics person, your job is to transform that scattered, siloed feedback into an early-warning system — spotting patterns in things like damp and mould, not just counting complaints but identifying which properties need urgent action today. If you're in a customer-experience team, your job is to take that intelligence and turn it into action, seeing where your interventions make the biggest difference to customers' lives — and that's especially true if your team also has some responsibility around contractors, because under the new legislation it doesn't matter whether it's your people or contractors you hire dealing with an issue. And if you're in a business-improvement, strategy or leadership role, your job is to prove that prevention beats cure with hard numbers, building a business case that goes beyond "it's mandatory" to a no-brainer, through the savings in emergency repairs and legal costs.

A couple of final slides on the dates. 27 October 2025 — we have a month and two days — and at that point the categorise-triage-report process is required for any report of damp and mould and all emergency hazards. Throughout 2026 this expands to more hazards: cold and heat, falls, fire and electrical, hygiene and food safety, water supply. And the following year all the phase-one and phase-two hazards, along with all other hazards — light, noise, space and crowding — so that by 2027 there'll be 29 different hazards that need to be triaged and categorised the same way you're working towards for damp and mould right now.

But it's worth saying there's a clear expectation from the Ombudsman that you don't wait — that this is not about doing what's required in 2026 and then expanding in 2027, but looking to extend into other hazards as soon as possible. That's a daunting thing, but again it's also an opportunity. The best housing associations are not using this as a standalone compliance exercise or something to bolt onto existing policies, but as the opportunity for a complete redesign with the customer at its heart. How does one go about doing that? I'm going to pass on to Sarah.

Sarah Wilson: Thanks so much, Steve, that was a really helpful background. What I'm planning to run through this morning is a framework we've developed alongside our housing-association customers, so you can be confident it's built and tested in the real world. There's nothing worse than some abstract framework nobody ever uses.

The core idea comes from Bain & Company — the same people who created the Net Promoter Score. They talk about two loops in the complaint-handling process. When I say "complaint" today, I'm using it in the Ombudsman's sense of an expression of dissatisfaction, however made — so this covers the whole spectrum, from a grumble through to an emergency hazard.

First you have the inner loop. This is what we all spend a lot of time on — the day-to-day resolving of an individual complaint for a specific customer. Someone raises an issue, you fix it, you close the ticket, job done. Then there's the outer loop, which is about stepping back and looking at all the complaints to find trends and patterns, so you can stop these problems happening in the first place. You might be thinking, "we don't really do that" — and if so, you're not alone. I was talking to a head of complaints this week at a 30,000-home association who said he never gets off the hamster wheel to do that strategic overview work. It's very difficult in the day-to-day firefighting.

A best-in-class process closes both loops, and getting to that two-loop approach is what's really going to help meet the Awaab's Law requirements. It's not just best practice anymore — it's essential. The inner loop ensures you meet those tight deadlines for individual cases. The outer loop is where the real transformation happens — it helps you find systemic issues in your properties before they become widespread. The providers really excelling at this don't just track complaints; they categorise all their tenant communications against the HHSRS hazard categories to spot trends early. Instead of just reacting to fires day to day, you can proactively prevent them. That's the real difference between just complying with the law and truly improving tenant safety, which is the heart of Awaab's Law — making sure every family has a safe and healthy place to call home.

Let me break that down. The inner loop is all about speed and accuracy — identifying and fixing individual tenant problems before they escalate. That starts with capturing feedback from absolutely everywhere. Under Awaab's Law, the compliance clock starts ticking the moment a hazard is reported, and it doesn't matter how — a phone call, chat message, email, social-media post, even a casual comment. So many providers have a blind spot here because they only monitor formal complaint channels, and that's a huge risk: by the time the issue is logged formally, you could already be days behind on your compliance time frame.

The second critical step, once you've captured everything, is to categorise it against the framework instantly. This isn't just about finding hazards, it's about confidently ruling out issues too — what's in is just as important as what's out. You simply can't have staff manually reading every message. Most people on this call are using generative AI tools like Copilot and ChatGPT, but that only gets you so far — top-level summarisation. This is where an AI-powered specialist tool designed for the housing sector becomes essential. A good system needs to understand context, not just keywords. A tenant probably won't say "I have damp and mould" — they're much more likely to say "the paint's bubbling on my wall" or "my bathroom is black". The system has to be smart enough to understand these phrases all mean the same thing and signal a potential hazard. Right now the focus is damp, mould and emergency repairs, but a truly effective system is one you can train to recognise all the hazards, so you're ready for the next phases without starting over.

The third part is routing and resolution. Once the system flags a hazard, it has to get to the right team immediately, automatically triggering an alert to your health-and-safety teams with priority flags based on severity. And that's the inner loop done — from identification to categorisation to resolution, ensuring you respond well within those mandated time frames.

We can't just stop there, though. The outer loop is where you take your system to the next level — moving from reactive to truly proactive, preventing problems on a wider scale for everybody. The first step is to identify patterns. Bringing all that data together, you can ask big strategic questions: are we seeing clusters of damp in specific blocks or estates? Do condensation reports always spike in November? Are our 1970s buildings reporting more structural problems than newer ones? There was a really nice example from Guinness about a year ago — damp-and-mould cases were increasing and they weren't sure why, so they did some correlation analysis in the platform and found that people who mentioned plants or ivy were reporting increased instances of damp and mould. They trained a context theme on any botanical mention, used that to put figures on their hunch, and ultimately cleared the ivy and reduced those cases. You'd never see that looking at one case at a time — it's very difficult as a human to pick these patterns out. The outer loop makes the invisible visible.

Once you've got the pattern, the next step is prevention planning — where insight turns into action. If damp is recurring in properties with a certain type of ventilation, that should trigger a property-wide inspection or upgrade programme. If one contractor is constantly being mentioned and causing follow-up repairs, that's a trigger for a quality-control review. The providers nailing this get everyone in a room to review the insights — maintenance, asset management, resident engagement and compliance, all working together to fix the root cause. Even cases where no hazards are reported become valuable data points: knowing your silence helps establish baselines and confirms prevention strategies are working. The final part of the cycle is continuous improvement — setting clear goals like reducing damp reports by 20%, measuring the impact of your fixes, and documenting everything, not just for your own learning but as evidence for the regulator. The most successful providers run quarterly outer-loop reviews where data insights drive strategic maintenance planning, rather than just responding to the loudest voices.

I want to finish by talking about something I don't think is getting enough attention yet: the tenant's right to interrogate your data, which is coming in 2027. That sounds years away, but it'll be here before we know it, and it's a massive shift in transparency. You've got to be ready when a tenant asks "how many damp-and-mould issues have been reported in my building in the last year?" or "which of your neighbourhoods have the highest rates of electrical-safety issues?" These aren't just information requests, they're legal rights now, and you've got to answer them systematically with data broken down by geography and even demographics. For many, this will be a complete rethink of how data is structured — information can't be stuck in departmental silos anymore.

The key to unlocking this is taking advantage of your metadata — and don't let that word scare you, think of it like a tag. You're not just logging the complaint, you're tagging it with property type, building age, location, demographic group and so on. When you consistently tag all this feedback with metadata, you've got a powerful foundation for analysis. This isn't just about being ready for tenants' questions in 2027 — it's about using that exact same data infrastructure right now to find and fix problems before they affect tenants. It turns a compliance tick-box exercise into a strategic advantage. By closing the outer loop, you're not just complying with the law, you're changing how you operate and providing safer homes for all your residents. Back to you, Steve.

Steve Erdal: Brilliant, thank you so much, Sarah. Great point about 2027 being here before we think. In my head, when I hear 2008 my brain still says "wow, five years ago" — so yeah, it will be here before we know it. Thank you for talking us through that framework. If you have any questions for Sarah, we'll do a Q&A at the end, so please do get them coming in on the chat.

So Sarah's talked us through how a framework like that would work. How does it work in practice — how do you turn that into something you can use day to day? Delighted that Zoe's now going to talk us through the work she's been doing with the Guinness Partnership to allow them to do that reporting and triaging across all the different hazards attached to the three phases of Awaab's Law. I'll pass over to Zoe.

Zoe Wilson: Great, thank you, Steve. Very shortly we're going to jump into a live demo using the platform to show how you can use Wordnerds to track these all-important issues from all your different customer-feedback points. To frame this, I'd love to share an example from our customer, the Guinness Partnership. Their insights team are doing really impressive and innovative work with their data and the strategies they implement across the business. As ever, we're really grateful to Chris Haynes, the head of customer insight at Guinness, and the whole team there for generously sharing their approaches with us and with everyone today.

A bit of context on Guinness's approach. While they're still finalising things to be fully compliant, there are a few things in motion. Firstly, like many providers, Guinness send a transactional repairs-satisfaction survey across all trade types, which includes damp and mould. The next step is splitting those damp-and-mould cases out of the whole repairs selection and sending a different survey — less about satisfaction and more about whether the job is complete and whether the customer is confident the problem won't recur. To maximise effectiveness, Guinness are planning to integrate those responses with their CRM, so when a tenant indicates they're not as confident in the resolution as they should be, their case reopens and alerts the assigned case owner. That's helping create a closed-loop system, ensuring persistent issues get immediate attention.

Some of the steps Guinness are focusing on are optimising their data workflow. Their current repair survey includes questions about unaddressed damp and mould; the next step is speeding up the action and response — hopefully using an API to get real-time results from their survey provider, ingesting that data into Microsoft Fabric, which lets them enrich it with property-specific details.

The next part to focus on is how Guinness have built a framework in Wordnerds to track mentions of those Housing Health and Safety Rating System hazards — the HHSRS hazards. They've trained their own classification models using Wordnerds to identify mentions of these specific hazards across their different feedback sources. By using this systematic approach, Guinness have enriched their monthly TSM reporting so they can show hazard-frequency patterns alongside relevant examples from tenant verbatim. So they really see both the quantitative and qualitative insights being used to drive preventative maintenance. The framework is constantly live, always on in Wordnerds, so they can apply it across all their feedback points — not just the TSM survey, but complaints, emails, anything where customers are interacting. The particularly impressive thing is that it's helped position them ahead of the Awaab's Law requirements and raised awareness of recurrent issues. It's a model we're seeing other forward-thinking providers beginning to adopt.

So let's look at that HHSRS hazards framework in more detail. It's not quite as simple as it sounds. The framework contains 29 distinct hazard categories, and the challenge is that these must be consistently identified regardless of how tenants describe them. This is why traditional keyword tagging, or generative-AI systems, isn't really going to work at this scale. Take excess cold — a tenant could describe it in multiple ways. They might say "it's freezing", "it's like a refrigerator in here", or even "I can see my breath indoors". None of these use the word "cold", which exemplifies why keyword tagging isn't enough. Multiply that across all 29 hazards and it becomes near-impossible to manage manually. This is where the AI-powered theme categorisation Sarah introduced becomes essential. If you can train themes to recognise hazards in natural language, you gain confidence you're not going to miss critical mentions, across emails, surveys, social media and call transcripts — creating that safety net that meets the "all available information" requirement.

What I'm going to do now to bring this to life is bring up the Wordnerds platform and show the example of how Guinness have built this out. Before I get started, I just want to flag that what I'm showing today is a demo project. It contains synthesised data from a fictional housing association — there's no real customer data being shown or shared here today.

What you can see now is our theme setup in the platform. A theme is simply a way to group your data in a way that matters to you, creating the distinct categorisations you need to track mentions within your feedback. We've pre-trained some themes on this synthetic dataset based around the HHSRS hazards — 29 distinct themes, one for each hazard. This is just one example of how the platform can track those hazards; it's completely customisable to what you want to listen for. It's also easy to get any text into the platform via CSV upload or an API connection, so you can bring all your different feedback points into one interface to group, interrogate and report on your issues.

Let me show how these themes work, using excess cold. Thinking of the inner loop, we want to create a way to track any mentions where people talk about excess cold — say in an email. This is where the Wordnerds context theme becomes especially powerful: it lets you group conversations where customers are talking about the same issue but using completely different language. The first thing we do is give the system an example of something to include — "my house is freezing" — thinking about how a customer might talk in natural language. The next thing is to establish a boundary between what we do and don't want, so we give it an example of something not relevant — "my freezer is broken" — a similar word used in a completely different context.

The next screen is our theme training screen, where we teach the model what to listen for. Straightaway you can see the results on the right-hand side — these examples are taken directly from my customer verbatim. I'm already getting examples like "draught coming through making it freezing", "cold air". Some might not be quite relevant, but it's doing a pretty good job already. I then go to the training list on the left and simply tag whether each example is or isn't relevant, then click retrain. The more we teach the model, the more it learns what to classify. After training, you start to see all the different ways people reference excess cold — "freezing cold air", "cold spots on external walls", "she's sitting in the cold". Once you're happy, you can apply this across all your data sets — anything in your project, and anything you upload from this point forward, will automatically categorise any mentions, saving all that time and effort.

Once that's set up, the next step is the outer loop, understanding these issues at scale. What I'm showing now is a purpose-built housing BI dashboard — an effective way to visualise, interrogate and share your qualitative data so multiple stakeholders can understand how customers are talking about these issues. Everything here is data we've already classified using the Wordnerds themes. First, our hazards framework, showing each hazard and how many times it's been classified. Looking at this data filtered to the first quarter, perhaps unsurprisingly, damp and mould is my most-mentioned hazard with 933 pieces of feedback tagged to this theme. Scrolling down I can see each of the 29 hazards, some at much lower volume — but it's important to have that visibility into what is and isn't being mentioned. Following through with the excess-cold theme, I can see 66 mentions within my feedback.

This dashboard is completely interactive. I can click into the cold theme and the page refilters to show those relevant results, so the slightly crazy line graph with all 29 themes becomes a lot clearer. I can see how mentions have changed over time to understand any peaks across the year. Underneath, you can also see crossover themes — where you use other ways of classifying your data to get into the detail, like a theme around doors or heating crossing over with damp and mould. I can click into a theme and drill through to the direct verbatim — because although we're looking at high-level numbers, the voice of the customer is still super important.

The final bit of the demo shows how to use your metadata to go a step further. The page we're in now is a cross-table, comparing two different elements of your data. I've selected a couple of my hazard themes — excess cold and damp and mould — and chosen to compare them against my different wards, which are just areas in this demo dataset. The cross-table underneath gives significant insight: of all my comments about damp and mould, which areas are they coming from? Using the heat-mapping, at a glance I can see the west area has a far more significant proportion of damp-and-mould comments. That gives me a really targeted starting point to dive in, understand why, and put preventative actions in place. This is just one example — you could present property type, age group, different demographics in these columns, and even cross damp-and-mould-in-the-west with particular property types. So there's so much you can do to get to that outer-loop level and understand your feedback at scale. Hopefully that's been a good overview of what AI-based classification can enable. I'll hand back to Steve to take us through to some questions.

Steve Erdal: Thank you so much, Zoe, that was fascinating. As Zoe said, I want to give as much time as possible for questions, so please keep them coming. It's worth quickly saying that we've covered a lot of ground today and the scale of the change can feel overwhelming. I want to reiterate: you don't have to solve everything at once. Only 2% from that poll at the start said they were fully prepared for all the phases — everyone else is on that journey somewhere. Guinness have taken a lot of steps to get where they are, with Zoe and the Wordnerds team alongside them every step of the way. And the truth is that Awaab's Law does require automation to get to the point you've seen.

So, briefly, what does it look like if you want to get there? The first step is to contact Sarah, our resident housing expert, to talk through what you're currently doing and get a sense of what other associations are doing. You've all met her now — completely no obligation, but her time is finite, so if you'd like to book some time, her calendar is on screen. After that, to get a sense of how your data would look in the framework, we generally suggest a proof of concept: you give us a data dump of your complaint or survey data, we put it through a framework similar to what Zoe just showed, and within about four weeks you receive a workshop, a Power BI file that's useful to you, and a report you can show your team — these are the key issues we're facing, this is what our historic data is telling us. That gives you a benchmark for the triage you're going forward to do. To get the always-on model you saw with Guinness, that starts at around £24,000 a year, ensuring all your data is automatically tagged and passed back to you so you get that triage and categorisation done really quickly, at source, in real time.

Next week you'll receive a wee care package with these slides, the recording, and a guide we put together on Awaab's Law. Our next webinar I'm really excited about, because we're unleashing our data team on you — if you're sick and tired of people talking about AI and want more of what's actually going on under the hood, that'll be a really interesting session with two of our top data scientists. We're also going to be at the Housemark Data Analytics Summit, so please come and say hello — and there's quite an exciting announcement to make there, some more work with Housemark. So watch this space.

Thank you so much for the questions coming in. We've got about 10 to 12 minutes to go through as many as we can. I think it's worth starting with the first one that came in, and Sarah, this is probably one for you — the contractor aspect we talked about, around the fact you're responsible for your contractors' work as well. What can be done to bring third parties on this journey with housing associations?

Sarah Wilson: It's a great question, and one we get quite a lot — especially, as you mentioned, it's going to become even more pertinent. There are a few things we recommend. First, sharing the insights from your analysis: when contractors and third parties see the actual tenant voice and understand the patterns we're identifying, it makes it much more of a partnership relationship. We've got a nice example from Town and Country, which we'll share after the call, where they used the data to improve satisfaction around window cleaning — they noticed people complaining about dirty water being used, fitted some new taps around the blocks, and it was quite a low-cost way of increasing satisfaction.

Using the early-warning system helps too — if you see multiple instances of a ventilation problem, for example, sharing that proactively before it escalates is a good way to bring contractors along. And having shared accountability, with reports and dashboards you can both access, is a good way of handling it. At the end of the day, it's their risk mitigation too — they face reputational damage if they're associated with non-compliance. So making sure nothing falls through the cracks is the best way to handle that.

Steve Erdal: Thank you so much, really interesting. That shared purpose is something we often see — contractors don't necessarily agree with situations, so having that data there is so important.

The next question, and again if we don't get to all of them we'll send answers in writing — probably another one for Sarah. Should info be gathered every time a worker goes into a property — a gas service, an electrical inspection — and should this be treated the same way even if a customer hasn't actually complained about an issue like damp and mould?

Sarah Wilson: Yeah, absolutely. It's not just about responding to complaints specifically. People doing routine visits will often spot some of the early-warning signs we've discussed — damp, mould, condensation problems — before people complain. It's about having that predictive eyes-on-the-data approach. It helps you capture the spirit of the law, not just the box-ticking part of it. So definitely, yes.

Steve Erdal: Absolutely. The first time you're made aware of something can come from any situation, including these initial visits, so having a way of analysing that alongside the surveys and complaints is really important. From the Ombudsman's perspective, there's no difference in terms of how you learned about it — once you've learned about it, that sets the timer off.

I think we've got one here which is probably for Zoe — someone's asked, is the Wordnerds platform suitable for teams who are less technical?

Zoe Wilson: Great question. Short answer is yes. We've created the platform so anyone can jump in, get to know it and use it. We had a couple of our wonderful customers in the chat explaining they can use it in great ways — not that they're not technical, I wouldn't agree with that. And you get really great levels of support from a dedicated CSM in terms of actually being able to use the platform. So it's really easy to pick up and use to analyse your customer feedback. Honestly, if we can use it, then really anyone can.

Steve Erdal: Thank you as well to a couple of our customers who jumped in — Howard and Rebecca, thank you so much. You do yourselves a great disservice, because you do great things with the technology.

I think we have another question here around audio data — can we scan audio, and can that be incorporated from an incoming-calls platform? Zoe, I don't know if you've got anything to say to this — I know you've done some stuff in the past with transcripts of calls, which isn't exactly the same thing.

Zoe Wilson: Yeah, essentially we can work with call transcripts, but the technology doesn't do that transcribing process for you. We do some great things with those transcripts in terms of understanding root cause of issues, and being able to pinpoint things like length of call and how complex the hazards people mention are. So we can totally get that data in, as long as it exists in a text form first.

Steve Erdal: Brilliant, thank you. And with regard to the incoming-calls platform, Philip — it would depend on what that platform was and how we could get the data out of it, whether it transcribes automatically as some do. We'd love to pick that up and talk to you about it further, so do feel free to book some time with Sarah if you'd like to go into that in more detail.

Another really interesting question — about improvement notices. Do you have any examples of improvement notices and what the expected outcomes are, say on excess cold? Sarah, is this one you'd like to take? I could give a few thoughts if not.

Sarah Wilson: I could comment, but yeah, go for it, Steve.

Steve Erdal: In terms of what you're hoping to achieve, we don't have any examples or templates, again because this is so new. As and when we do see what those values will be, we'll definitely be utilising them in the analysis of the data — being able to break the data down by what's expected of improvement notices, in terms of those outcomes. The only thing I'd add is that it's not just about the hazard, it's about the person. As Zoe showed us, you can find areas where there's a particular issue with damp and mould — you can do the same thing with vulnerability. So not just hazard, but vulnerability, and uniting those two things would be a key thing for improvement notices and for the categorisation and triaging. But in terms of improvement notices specifically, please do watch this space — we'll be gathering this information and working to make that better. At the moment there are lots of different approaches and no one knows which one's best yet. Is there anything you'd add, Sarah?

Sarah Wilson: No, I don't think so. Just the capturing of something like cold, which is very hard to capture using keywords. People tend to say "freezing", or "I feel Baltic", or whatever — so just capturing any mention of that is important.

Steve Erdal: We've got an interesting question next, just about the AI aspect. With it being AI, would our data be used to train large language models? Zoe, would you mind speaking to this one?

Zoe Wilson: Yeah, sure. Again, short answer is no. The beauty of Wordnerds is that when you're training your themes, you're training them against your own data. All your data lives in its own project, in its own specific area. So it's not used to train LLMs. And you can get really specific in how you're training your themes based on how your customers are actually talking, specifically because of how that AI is set up. I don't know if there's anything you'd add, Steve.

Steve Erdal: You summed that up beautifully, Zoe — in the interest of time, I'll keep going. Your data does not interact with anyone else's data. It's not used to train any model except the small model that's used to go through your data and find the useful stuff for you.

Next, a request from a resident wanting to know where the most electrical hazards are. That seems like an unlikely request, but in terms of what could be seen as a reasonable request, given there's an unknown volume of them — Sarah, do you have any sense of what might be seen as reasonable here in terms of volume?

Sarah Wilson: It's a really interesting point about volume. For many of these hazards you're only going to see a couple a year. Ensuring you know what's reasonable to expect, in terms of being able to say to residents who want to know where the most likely electrical hazards are — given you might not know the volume — is the challenge. Our response would be: what's the closest you can give them to that information? In terms of unknown volume, it's really hard to get an exact number — an exact number of electrical hazards, an exact number of people suffering from damp and mould — that's a really difficult thing.

The best you can do is have some way of understanding the relative size of it, and the customer-feedback data will tell you that. You might not be able to pinpoint every single electrical hazard through feedback data, but it'll give you a sense of broadly where it's happening and how it compares to other hazards. Relatively speaking, is excess cold a bigger or smaller issue for your association than electrical hazards? The key thing is that it's not reasonable to expect you to know about every single thing. The legislation is clear that it's when you're told about it.

Steve Erdal: When you're told about it, it's expected that you can say "this is the number we've been told about". Is that fair, Sarah?

Sarah Wilson: Yeah, that's fair. And that touches on another question — very aware we're at time now — there's a question around, with the analysis of verbatim, is there a view on whether housing associations will be on notice to undertake an investigation within the prescribed timescales? The answer is yes — as soon as you're made aware, there's a ticking clock. Totally understand that's challenging, but this is why the automatic categorisation comes in. Having that in real time is absolutely crucial, or you're just going to be wading through an impossible volume of requests.

Steve Erdal: Brilliant, thank you so much, Sarah. As you say, we're just beyond time now, so we'll leave it there. We've got so many really interesting questions — we will get in touch with an answer, and we might do an FAQ sheet so we can give everyone the answers we give to those individuals.

Thank you so much — we've had loads of really great questions today. Please keep us informed on what you're doing in this area; we're really keen to learn from you. As with Guinness, we're magpie-ing good ideas here, so please let us know what you're up to. This feels like a stressful and overwhelming time, but for data-savvy organisations this is also an opportunity to do something really powerful for your organisation and your customers. So best of luck, enjoy the sunshine, and we'll see you soon. Thanks so much.