How can you use AI to analyse online reviews?

Ross is one of the principal product managers at Trustpilot. His specific area of expertise currently is their API, and in that capacity he is super well placed to talk about how Trustpilot works with other organisations and how it fits into the ecosystem of customer experience. We've been dealing with Ross for a few months now, and we're really excited to welcome him today. We've got an inkling of what he's going to talk about, and it's some next-level stuff. So thank you, Ross — appreciate your time.

We also have on the cast Stella. Stella is part of the graduate program at Wordnerds; she's one of our analysts. She's loaded with cold and feels utterly miserable, so she's putting on a brave face — if she looks a little less bouncy than usual, that's what it is. Stella comes here to do live data science demos, because — yes, I know we're spoiling you. Nothing says Thursday lunchtime like a live data science demo. More on that in a moment.

If you don't know Wordnerds, we're a customer feedback analytics platform. We have the pleasure and privilege to work with some of the UK and world's most data-focused and customer-obsessed brands. We work in four main areas. In social housing, we work with a bunch of organisations, largely because of the regulation forcing them to do more with resident feedback. We work with people like Lloyds Bank and True Potential in financial services. In retail, we've got clients like the wonderful Sainsbury's, Marks & Spencer and B&Q. And in travel and hospitality, people like the Dorchester Hotel and Hotelplan. So welcome to those of you on the cast today from some of those brands — and if you're new to Wordnerds, thank you for joining us as well.

Why are we here? Online reviews are an amazing resource — a figurative gold mine of data you can use to your advantage in a bunch of different ways. We at Wordnerds deal in the customer experience space, so we look at it through a customer experience lens, but as Ross is about to elucidate, there are other things this information is useful for as well. Most brands we talk to fail to utilise reviews to their max, and the point of today's webinar is to explore how you can do more with online reviews: what you can do, who's currently doing it, and what your first steps are.

We're going to do a bit of a now, next and future of online reviews. Ross is going to take us through a little about Trustpilot — the story so far, the importance of content integrity — but the really exciting stuff I'm here for is how this AI revolution is changing e-commerce as an industry, and the implications of that for reviews. After that, we'll talk about how brands are using online reviews right now. You may have noticed that in the ads for this we were meant to be joined by Ruth, who was going to do this piece. Unfortunately Ruth can't join us. I am not Ruth; I am a poor facsimile of Ruth, and I'll endeavour to deliver her slides anywhere near the quality she would. I beg forgiveness, but bear with me.

I'm going to talk you through benchmarking, using reviews for positioning, and how to work out what actions you can take from reviews to improve your customer experience — with a very short three-step process — so we can get to the main bit we're here for: Stella getting her hands on some actual data. We are Wordnerds; our idea of a good night in is a big stack of data and a bottle of wine.

We've chosen, for the demonstration, a real-life retail example. Everybody shops at supermarkets; it's a point of reference we all have. So we've picked three supermarkets from different ends of the spectrum: Waitrose, Morrisons and Aldi. We've downloaded some review data from Trustpilot — thank you for that, Ross — and Stella's going to look at what that data tells us about how their customers experience those brands differently. If you were working for those organisations, what could you learn about your service and product delivery, and what might you do next?

So that's the plan. There will, after that, be a very quick plug for Wordnerds — no such thing as a free lunch, and we haven't even provided lunch, isn't that rude? It'll be very quick and painless, and then we'll go straight into Q&A. Please drop your questions in the chat at any stage. Behind the scenes we have Alex and Vic making sure all the levers are being pulled and the whole Wizard of Oz thing is going on. Starting with a little audience interaction: there should be a poll popping up on the right-hand side now — we're really interested in how you're currently analysing your online review data, so have a look and click one of those answers.

It looks like we've got a good mix between people who are actively managing review data and people who want to dedicate more time to it, which is a good start. So you're all aware of it and using it to different degrees — it feels like the right audience for us. Thank you for that. Ross, without further ado, over to you, mate.

Ross Hancock (Trustpilot): the story so far, and how AI is reshaping e-commerce

Ross Hancock: Awesome. Thank you so much, Pete — really appreciate it. Trustpilot is, as I'm sure most of you know given the poll, the largest independent platform for customer feedback in the world. We have over 300 million reviews, and that is growing at an incredible rate — at least 60 million reviews per year, and increasing. We have reviews from all industries across all geographies, from Uzbekistan to the USA.

But what's interesting is where Trustpilot started. It actually started with a washing machine, which might seem a bit weird. Back in 2007, our founder Peter went back to his family home. He was sat on the sofa and his mum comes up to him and says, “The washing machine's just broken. I found this one online — should I buy it?” He has a look and goes, “I don't recognise that business at all. I'm really not sure if you could trust buying it or not.” It struck a chord with him, and he realised he could probably solve this problem: if I can allow real users to provide feedback on businesses, then I can have confidence in buying online, and I can help people like my mum.

One byproduct of that, which is hugely beneficial, is that it allows businesses who are trustworthy and authentic to find the consumers for them. But underpinning all of that, for it to work, the data has to be trustworthy. It has to come from real people, and it has to reflect real experiences. At Trustpilot we invest heavily in ensuring that's the case — through a large volume of humans that moderate content, but also through a vast number of sophisticated technologies to detect and ensure the content is trustworthy. We look at a variety of signals: consumer verification based on IDs to make sure a Trustpilot account is operated uniquely by a human, through to signalling from location and patterns in time-stamping, some of which is proprietary. That ensures the content on the site is as trustworthy as possible.

All of this is really important because we're seeing a major paradigm shift now in e-commerce with the introduction of AI. If we think about the history: when e-commerce came along, you could start buying from businesses on the other side of the world rather than the local shop. When mobile came along, you could buy anywhere — on the go. The big change we believe is coming now, and seeing in real time, is with AI. At the very least, we're seeing it in two ways. The first is the introduction of interactive advertising.

If we think about the classic e-commerce journey, you'd see a static advert on a social site, click through, find the product in the shop, select the size, add it to your cart, enter your address and payment details, and finally make the purchase. What we're seeing now — and I'm sure will be released within months — is the ability to interact with adverts directly, so you don't have to leave the platform you're viewing the advert on. For example, you may see an advert on Instagram selling t-shirts. You could interact with it directly, through text or voice, saying “Is that t-shirt available in red? Can it be shipped to my location? Let's go ahead and buy it.” And that purchase takes place on that social online store.

The second thing we're seeing is the introduction of agents. An agent can automate parts of your workflow, and in the purchasing journey it can find the products you want without you having to search yourself. One thing we firmly believe, however, is that this isn't like Terminator — we're not just going to give control over to automated agents. We still believe the human stays firmly in control.

So as you imagine that customer journey, where you explore a product and evaluate whether it's appropriate, we believe the purchasing decision will still be made by a human — it's the point at which a human approves a purchase for an agent to make. That means it's still utterly critical, maybe even more critical, to get the signals that tell you what the human's experience is, because the human's experience determines whether they make the purchase. That's why it's really valuable to have data that reflects the true human experience, so you can understand it, improve it, and therefore improve the likelihood the consumer purchases. Does that give you a good overview of where we're seeing the direction?

Pete Daykin: It really does, and I love it. I love the idea that all of a sudden you can buy directly from ads without having to go into people's websites. We're seeing with Gen AI — not necessarily the death of search engine optimisation, but the fact that you put a search in and the results come straight to the top without you having to go to people's websites. It's changing the way we consume information. Is something similar going to happen with e-commerce, do you think? As the people who provide the social proof, are you going to have to completely change the way you deliver the results of reviews, because people are seeing these things out of context of your site or the e-commerce site?

Ross Hancock: Yeah, absolutely, spot on. To give a concrete example, Perplexity have released a shopping agent — I think it's only available in the States for the minute. Perplexity, for folks who don't know, is a chatbot interface similar to ChatGPT or Gemini. In their shopping agent application, you can say, “It's my daughter's birthday coming up, I need to find a present, she's four years old.” If you've interacted with the agent previously, it already knows that information, so it can hone down and personalise what you see.

It'll return different products it thinks are appropriate based on consumer reviews, but also that personalisation aspect — based on the interactions it's had with other people, what's appropriate for a four-year-old and what's popular. And it's not only automating that exploration and evaluation stage, but also including the purchasing within the interface. You can simply click “buy with Perplexity” and select the attributes of the product. So we're seeing that brought into different parts of the customer journey, as opposed to that traditional search-engine-to-online-store-to-purchase. What's important is how we bring that social proofing into those applications to make sure consumers are buying from trustworthy businesses — and we do that in a variety of ways, one of which is providing Trustpilot data through an API so those platforms can consume that content.

Pete Daykin: Interesting. And I guess particularly if you're seeing it out of context — there are probably hundreds of decisions we make about how trustworthy a business is when we go on their website. If you're divorced from that process, that adds more weight to the review data you're seeing out of context. So what's your advice for brands about managing their review data and putting their best foot forward, so that when they're seen out of context they can encourage a positive response?

Ross Hancock: I think it's going to become as critical as ever to manage your online reputation. As these bots take control of that exploration and evaluation, they're going to rely on review platforms and other signals — there's no option that those bots won't do that. So managing your online reputation is just so critical. And obviously the way to improve that reputation is to improve your business, and to improve your business it's all about improving the customer experience. The only way you can do that is by truly understanding what the current state of that experience is.

Pete Daykin: That's just fascinating. I love this idea that there's going to be a little bot following me around that knows more about my purchasing journey than I do, and will recommend me all these dull middle-aged things I buy these days. It'll just put those things in my lap. I hate shopping, so that feels great.

Ross Hancock: What I find fascinating is this isn't far out into the future — this is happening now. It's being piloted in different geographies and there's great consumer adoption. As I mentioned, Perplexity, but other big vendors are doing this now. It's being put into the hands of real consumers.

Pete Daykin: It's amazing how rapidly this AI revolution is changing every facet of life. We did a hackathon at Wordnerds a couple of weeks ago to automate processes that were annoying us, and the progress we made in one day — from illustrations to dull infosec processes — was unbelievable. To see it affecting e-commerce this way is incredible. Ross, thank you — we'll have lots of opportunity for people to ask you questions at the end.

Pete (Wordnerds): how brands use online reviews now

Pete Daykin: It's a beautiful segue, too, talking about the importance of really understanding your customer — I completely agree. A lot of the brands we talk about proactively manage their reputation; they respond to reviews. But remarkably few of them analyse reviews and understand the content. That's what we want to talk about today, picking up from Ross. Review data is, as I've said, a figurative gold mine of useful information. But there's a challenge: a lot of the information can be old, there's far too much, and all the usual problems with text analytics exist in spade-loads with review data. The challenge for most people is: where's the gold? Where's the stuff that's actually actionable?

What is actionable depends on the question you bring. Generally: what are our customers telling us that we didn't know — the unknown unknowns? Or what do they think about something very specific — if you're a holiday company with a particular location, or a supermarket with some new vegan sausages, are they being well received? Sometimes you bring hypotheses. From a positioning perspective, we get a lot of people asking whether they own a particular part of the customer mindset. We did some work for a high-street retailer a few weeks ago: they'd always been the brand that owned value for money, but after a couple of years of double-digit inflation they'd had to put prices up, and they were nervous it might affect their reputation as the value brand. By looking at review data and comparing them to their competitors' customers over the past 24 months, we plotted exactly how that had shifted.

One thing we find is that your survey data and complaints data — the information you have directly from your customers — is the highest-quality data available to you, and it will definitely tell you when you have a problem. What it won't tell you is the context of that problem. We do a lot of work for train operating companies. Every train operating company in the country has people complaining about late trains. The question isn't “are people complaining about late trains?” — it's “are you outperforming or underperforming the market?” Just because lots of people are talking about it doesn't mean it's going in the wrong direction; it might have improved massively. One of the joys of review data is it lets you understand the context: is it just an issue for you, or a wider industry issue you're actually overperforming on?

One of the big things insight and data teams are spending more time on, as technology gets better, is action. A couple of years ago the thing we heard most was “we're data rich and insight poor.” Now everybody has insights up the yin-yang — we've got more insights than we know what to do with, and which of those to act on becomes the more pertinent question. And then people ask: there's a new brand on the scene seen as really good at this thing — how do we understand what they're doing and apply it to our customers? Depending on the question and the data you've got, you'll get a different answer. But regardless, there's a very simple three-step process that lets you do more with review data, and it involves tools, process and methodology.

When it comes to text analytics, there's a definite maturity model we see. Everybody started this journey doing manual coding. A few years ago, before this AI revolution, text analysis wasn't very good — sentiment analysis was limited, and most people were reading the verbatim, five-bar-gating it, doing it by hand. You'll be shocked how many people in 2025 are still manually coding most of their customer verbatim. If that's you, don't worry — everybody's on a journey; more people we meet are still doing this than aren't.

Once people have been through the manual-coding phase, the next thing they try is getting one piece of software to do it all — often a Qualtrics or a Medallia, or a stack like the Microsoft one. It's a perfectly rational response and will solve some of your problems. Qualtrics and Medallia are brilliant tools for generating customer feedback and bringing lots of data into one place, but the text analytics they provide isn't quite what they were designed for, and it's quite limited. We encourage everybody to have tools like that in their stack, but not necessarily for review data and wider customer data sources — call-centre data, things your staff write about your customers. Wherever there are words from or about your customers, you need to be able to process that.

The next thing everybody tries is Gen AI, often Copilot, because it's bundled with the Microsoft stack — it can be ChatGPT too. That stuff is getting better all the time. We did a completely separate webinar a couple of weeks ago on “can I use Copilot, can I use Gen AI to analyse my customer feedback?” — we'll send a link. Stella got her hands dirty with the data and compared Copilot to other methodologies. We talked about why that technology is set up to do some things really well, but analysing your customer data numerically and factually is not one of them. That might change, I'm sure it will. But Gen AI is not yet the answer. And — shock horror — we are a specialist text analytics tool, so our conclusion is you should use a specialist text analytics tool.

When we say that, the things we think are important about a specialist tool are: firstly, a federated system of software providers that each do very specific things very well. Whatever tool you use needs to be integrated both in and out. It has to sit under all your frontline tools — different parts of your organisation already have software they're happy with for complaints, social media and call centres — so it has to pull all that information in, and push it out the other end too. Nobody wants another platform just to look at reviews on. Increasingly we're reporting this stuff in Power BI and other BI tools, because that's where people already are with their quant data. Quant data is great at telling you when there's a problem, but to find out why and how to fix it you need the qual data — the customer verbatim — and reuniting all of that in your Power BI dashboards is a really important step. And Power BI doesn't play nicely with text data, so that's a challenge.

It needs to be configurable — you need to be able to listen for anything at short notice. Gone are the days when software would just give you 10, 20, 30 things and say “this is a delivery issue, this is a quality issue.” You need to get right into the minutiae and find very specific problems. And it needs to be accurate and transparent. One issue with Gen AI is that it'll tell you something is a particular problem, but not how it arrived at that decision. We get into very nuanced conversations about what makes something fit a category. We did some work at B&Q where they wanted to find out whether people could find the toilets in store — and as a company that sells toilets, the difference between “I couldn't find the toilet” and “I couldn't find the toilets” is subtle but really important; they both have completely different meanings. No technology is 100% accurate where words are concerned, so what you really need is technology that tells you how inaccurate it is, so you can decide whether to trust it when making decisions.

The second part is a three-step process for analysing reviews. Training AI to listen for stuff is great, but it's not really listening — it misses valuable parts. The first thing it misses is the stuff you didn't know to listen for. We call it zero-shot topic analysis — telling you just what's in the data set. We use a lot of corpus linguistics for this: what are the surprises, the unknown unknowns, what's new and growing this month that's going to be a problem in six months that you've never come up against? Then there's the classification piece — setting up a framework to put things into buckets so you can see the size and sentiment of each. And then, increasingly, the game is about how you prioritise that data: if you're a massive supermarket, the number of things customers can tell you — from problems in the car park to issues with specific products — is myriad, so how do you work out what to act on first?

The answer to the prioritisation question is around methodology. You need to apply a bunch of different methodologies to this data, and bringing in tools like Power BI after classification opens up so many possibilities. We were talking this morning about a project with a housing association, looking at customer feedback early in a tenancy to build a predictive model of whether it's going to be a success — whether those people are going to stay. That's completely different to “can I work out what's driving my NPS score down, and will changing five things this year make it go up?” All of these are possible — you just decide which methodology you care about depending on the question, and set up the process to deal with it.

The methodology Stella's going to show you today is one we've been working on with Professor Alamanos from Newcastle University. He's a professor of marketing who's done a lot of work on omnichannel customer journeys in e-commerce — particularly complex journeys with both an in-store and an online presence. There are two areas he's interested in. Firstly, the experience values — the steps people need to go through to be happy before they buy (you can see some on the left-hand side), each breaking down into smaller things we can listen for and measure in real time on review data, for your customers and your competitors' customers. Secondly, the customer journey, which is increasingly nonlinear — and the agentic AI Ross spoke about makes it even more nonlinear. It starts with planning and being told about stuff, goes through the actual purchase journey, and then there's a big post-purchase part: when people are using things, are they reviewing the product, experiencing issues, talking to the brand? We've taken that methodology and applied it to the data.

So that's the theory — I promise you Ruth does it so much more succinctly than I do, so apologies for my delivery. It's all very up in the clouds, so to give a good example of what it looks like in real terms, I'm going to bring Stella in to take our Waitrose, Aldi and Morrisons example and show you what she's found in their data, which I'm excited about.

Stella Dooris (Wordnerds): live Trustpilot benchmark — Waitrose, Morrisons & Aldi

Stella Dooris: Yes, hello. My plan is to take you through a real-life example from the Trustpilot reviews of the three supermarkets. To begin, this is the overview page on Power BI, which we've taken the reviews onto. Here you can see the difference in the number of reviews and the sentiment of reviews — a very broad view of all the supermarkets. You can see at the bottom that sentiment has been increasing over the past year, which we found interesting, and we thought: why is this?

So we looked at the individual star ratings of each company, and you can see that Morrisons over the past year is a big reason for that increase in star rating and sentiment. Aldi's has stayed relatively similar, same with Waitrose — if anything that's gone down. There can be many reasons Morrisons' star rating has gone up. We trained themes to see if any stuck out, but one reason we found was the difference in engagement on the Trustpilot pages.

Morrisons reply to 96% of their negative reviews, typically within 24 hours. Waitrose reply to 62% of their negative reviews, typically within one week. And Aldi don't reply to their reviews at all. We found that interesting — when engagement goes up, star rating may go up, in that customers feel listened to. It could be a reason for this change.

To look deeper, we can look at what matters to people. On the right-hand side are themes that we as Wordnerds have trained, in that customer journey Pete mentioned, to see what people care about. This is ordered by volume, and you can see emotional effort — things like anxiety or stress towards their shopping or journey — is the top mentioned thing, followed by praise for the in-store experience, monetary themes, personal interaction, and negative staff interaction. These are clearly things customers care about and talk about most across the data set. We can also look at what's driving negativity across the three supermarkets — things like contact-centre experience, negative staff interaction, delivery and collection, and communication.

But you want to know how you can benchmark against competitors — what is one supermarket doing that others aren't? That's where we have this graphic of the customer journey. With the categories of themes we've trained, we can see the volume and sentiment at each stage. The size of the circle is the percentage volume of comments in each brand, so the bigger the circle, the more people are talking about it — so we can compare more accurately across different volumes. You can see Morrisons leads on most categories in this data set.

If we start with the starting stage, store vibes — and of course vibes are very important, everyone loves vibes — are high volume and high sentiment in the data. Compare that to Waitrose, who have lower volume and lower sentiment, and you start to see why Morrisons may be higher sentiment in this category. Next, the shopping stage — Morrisons again. This is probably a more important category because it's where the main things happen: positive staff interaction, the in-store environment, value for money are all drivers of positivity within Morrisons. If we compare it to Aldi, their negative staff interaction outweighs their positive staff interaction by quite a lot, and interestingly their value for money is lower than Morrisons, which I wasn't expecting because they're a cheaper supermarket. For Waitrose, value for money is even lower, negative staff interaction is higher than positive, and a lot fewer people are talking about the in-store environment.

Finally, the following-up stage — the after stage. A lot of people talk about being a loyal customer at Morrisons. This is people self-identifying — saying “I've shopped here for years” or “I'm a loyal customer.” It's high and positive in Morrisons. Compare that with Waitrose, and their loyal-customer base is actually quite a lot lower, which is interesting because you think of Waitrose as something people are loyal to. So it's not as big a mention as you'd think, which might be something to work on.

We can dive deeper into this loyal-customer theme. On the right-hand side are crossover themes — what people talk about when they self-identify as a loyal customer. This shows what really drives loyalty: positive staff interaction, in-store environment, store vibes, promotions and offers, navigating the store, value for money. These relate back to those top-volume themes at the start — so if people care about those themes and Morrisons do them well, that's likely why their sentiment is higher.

Diving deeper again, we can look at cross-tables, which is one of my favourite graphics. Here I've put the competitors as the rows and the themes as the columns. When people self-identify as loyal customers, 15% at Morrisons talk about being a loyal customer, compared to seven or eight percent at Aldi and Waitrose. If you look at churn risk — people saying “I'm never shopping there again” — it's kind of an inverse relationship: Aldi and Waitrose are much higher than Morrisons. Waitrose has the highest proportion of people talking about churn risk.

So I did another cross-table on people at Waitrose talking about churning, against brand perception — people talking about the brand as a whole, its standards. Interestingly, 46% of people who talk about leaving Waitrose talk about brand perception: the brand not being what it used to be. And 13% of people talking about leaving mention their expectations not being met. So that's something for Waitrose you can identify in the data — a reason why the negativity around Waitrose is such a thing.

Finally, this is where we get more actionable insights: what's driving negativity in each data set. For Morrisons, it's things like contact-centre experience, delivery, collection, communication, negative staff interaction — so those are what we'd focus on if we were giving recommendations. Interestingly, compared with Waitrose, Morrisons' negativity is a lot more external — contact centre, delivery, communication — not necessarily in the store. For Waitrose, things like payment experience, product information and returns are a lot more internal to the actual store. Those are things we'd start to look deeper into and turn into actionable insights. So you're able to benchmark, and see what your competitors are doing well or less well. That's me done — my conclusion is Morrisons seems to be the best and I'm going to get all my shopping from there in future.

Q&A

Pete Daykin: Thank you, Stella — well done, you absolutely smashed that, and you'd never even notice that cold. Never put a millennial in charge of naming themes, otherwise you get themes like “store vibes” — that's my learning from this whole experience. Like I said, what now follows is a party political broadcast on behalf of the Wordnerds Party, and I'll keep it very brief. If you like this stuff, we can do it for you. We can run a competitor benchmarking report, fully managed by the nerds — your brand versus competitors — and we can pull in your own data as well.

Where Stella was talking about, say, if you're Morrisons and you change your contact-centre stuff and that has the biggest influence on satisfaction, there's a step of “right, what do you do to solve that issue, what's the next level down?” For that, your own data is really helpful — bringing it all into one place and getting even deeper and more actionable is great. What you get is an interactive Power BI dashboard like Stella just showed, a key-takeaway deck where we pull out the interesting things, show the evidence and give recommendations, and then we tend to run an online stakeholder workshop where you and other interested people come together to discuss the implications. It's a great way of socialising this stuff and bringing everybody along. If that's interesting, we can do it for about six thousand pounds — there's a thing on the side of the screen saying “book a chat with Pete.” I love talking, so feel free to book a chat.

Right — there are loads of questions coming up in the chat, thank you so much. The first one, I think, is for Ross: have you got UK stats for the number of Trustpilot reviews?

Ross Hancock: Yeah, we have around 120 million reviews in the UK, and that's growing at around 20 to 25% per year now. We expect that to grow exponentially.

Pete Daykin: Fantastic. And when we spoke a couple of months ago, one of the things you were interested in is how you use this stuff across brands — to spot patterns within verticals and how consumer behaviour is changing. Is it worth mentioning a bit about that?

Ross Hancock: Yeah, 100%. There are a couple of applications. Really understanding how consumer behaviour is changing, and how that differs across geographies. And — we're going to see this more and more — customer feedback is going to be a key input into investment decision-making as well. The investors for your companies will be using review data and customer feedback data to manage their capital allocation across their portfolio. So it's become increasingly important, not just to deliver a better customer experience and grow the business, but for investment purposes as well.

Pete Daykin: Thank you, that's really helpful. Next question — thank you, Kerry: what kinds of businesses are adopting this AI? Is it just retail or broader than that? Could it plan me a holiday?

Ross Hancock: Yeah, you're spot on — retail is just one example. Holidays, the whole travel industry, is likely to be one of the first implementations. We expect OpenAI to release their operator product in the UK in the next couple of months, and one of the core use cases will be travel. You'll be able to ask it, “I want to go to Italy, book me the best hotel as rated on TripAdvisor,” and it'll go ahead and do that. The way it does it is by effectively taking over your screen — it'll screenshot the screen and do some image detection to navigate the website. It'll look through the TripAdvisor website, find the highest-rated hotel, but then also consume review data to personalise that — maybe you value a great breakfast, it'll look at the feedback and take that into account. So that's absolutely happening, including the purchasing experience, all within the interface.

Pete Daykin: Fantastic. It really underlines what you're saying about customer experience and how you need to understand it and be proactive about finding out what your customers are learning about you at the point they're consuming this stuff. There's loads more and we're running out of time, so I'm going to push on. Paul, thank you for your question: can you link the findings to a financial measure, i.e. improve revenue or value at risk? I can answer this.

Absolutely. We spend a lot of our time building predictive models for people, looking at their feedback: can we identify people who are going to churn, what's the chance they'll leave your brand? Because this is in Power BI, we can take any quant data as well and look at the drivers compared to those metrics — what makes spend go up, what makes it go down. Anywhere you've got numeric data, we can use it to segment this feedback. You can do that as a lagging indicator — look historically at what's driven that number — and then, the holy grail, build predictive models. We use all kinds of techniques; random forest analysis is one of our favourites at the minute. It's not perfect, but most of the time we're building models that start at 80 to 85% accuracy out of the box and improve from there. The beauty is, the more data you get, the better it gets. So yes, using this to predict revenue is a key thing lots of people are interested in.

Pete Daykin: Kerry, in your Power BI dashboard, Stella — I'm coming to you for this — can you see the verbatim of what the reviewers are saying?

Stella Dooris: So yes, there are ways to see verbatim within Power BI, but at Wordnerds we usually go into the platform and see the verbatim there. So short answer — yeah, you can do both.

Pete Daykin: And why do you tend to go on the platform?

Stella Dooris: Because on the platform you can go deeper — you can cross it against other themes, so you can get really into what a customer's saying and also see what else is present in that comment: what other themes are present, what else they're talking about.

Pete Daykin: Interesting, thank you. Just for my own satisfaction — we've got time for one more before we go. Ross, this feels like it's for you: what's the difference between Trustpilot and Feefo? In the past I felt Trustpilot was more an open platform rather than needing to be a verified customer invited by the company. Is this changing? What makes Trustpilot better, or different, as a platform?

Ross Hancock: Yeah, sure. Jimmy, you're absolutely spot on — our core differentiator is that we are open. Anyone can leave a review. That said, if we know more information about the consumer, that holds more weight — not in terms of the trust score that's displayed, but in terms of ensuring it's more valuable to you when you do your custom analytics. If you do some customer analytics on feedback from a private tool — say something like Medallia, where you're directly asking the customer for feedback — that won't necessarily be representative of that customer's experience, because there's inevitably some selection bias, and the information the customer provides also tends to be biased. So we've found that being open is absolutely core and critical, and the key differentiator for us. It provides far more honest feedback in order to improve. Pete, you might have some examples of where you've seen differentiation between private feedback and public feedback.

Yeah, so the core purpose is for us to give a true and accurate representation of how the customer is experiencing a certain product or service. And as I said, being open is the best way to achieve that.

Pete Daykin: We find all the time that brands get really frustrated because the way a customer feels doesn't always represent reality. They spend millions of pounds changing some core element of their service, but because they haven't communicated it to the customer well enough, the customer complains about something and thinks nothing's being done — and almost always it's a comms issue, not an actual service-delivery issue. So what they're saying doesn't necessarily reflect the actual situation, which I think is a version of what you're saying.

I love it. Thank you. We are at time — I could talk about this all day. I'm off to set up agentic AIs to go and research golf clubs for me, because that's how interesting I am these days. Ross, thank you so much for your time, it's been fascinating as ever — we really appreciate you spending your morning with us. Stella, you smashed that, well done. And thank you everybody for joining us. We'll be sending some resources out afterwards, including the link to the other webinar on Gen AI — do check that out. We'll also be asking you for your feedback. It's 12 o'clock — go and have some lunch. Thank you again for your time.

Ross Hancock: Thank you.

Stella Dooris: Bye everyone.