Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Episode 4 - Bertie Vidgen - Rewire - The future of business with AI
109 Plays2 years ago
In our fourth episode of Survey Booker Sessions we speak with Bertie Vidgen from Rewire, building socially responsible AI for online safety.
Artificial Intelligence (AI) is a hot topic and being used in a growing number of ways. We discuss:
🦺 How Rewire are using AI to tackle online safety
🛎️ The move towards moderating and deriving information from customer feedback
🤖 Will AI replace or support people and customer service teams
🛣️ Understanding your value proposition and when to pivot
🤔 Explaining new products without getting stuck with jargon
🛺 Understanding when is right to use AI or automation
👥 How to attract talent with a high competition for candidates
Recommended
Transcript
Introduction to AI and Human Synergy
00:00:00
Speaker
I'd never want to see a system given the performance of AI now where you don't have a human coming in and checking and saying actually is this sensible like we're making the right choices so you know and I think AI when it's used it is best and free up your human expertise to do the thing which humans can do which AI can't do you know you can't build a relationship with the customer you can't understand their problem in depth with AI like that is just not what it's there for
00:00:25
Speaker
That's where you want humans and people to do it. I think those jobs I don't see going and hopefully with more AI we can actually free people up to really do that more.
Launching Survey Booker Sessions Podcast
00:00:34
Speaker
Welcome to Survey Booker Sessions. Tune in to hear from people working in a range of industries and roles to provide you ideas that you can take away and use in your own business. I'm your host Matt Nally, the founder and director of Survey Booker, which is the leading CRM and survey management system for surveyors.
AI in Content Moderation with Bertie Vigen
00:00:48
Speaker
So on today's episode, we have Bertie Vigen, who is the co-founder and CEO at Rewire. So thanks for coming on today, Bertie. Yeah, thank you very much for having me. Really glad to be here. So what's your role in Rewire? What does Rewire do?
00:01:00
Speaker
Right, so I'm CEO and co-founder. It's a, Rewire is a company that I founded with the CTO, Paul Rotka. We met whilst we were at Oxford and the Alan Turing Institute doing some research into the use of AI to stop hate speech and to stop different forms of content that is illegal, harmful, or which just violates a platform's terms and conditions. So that's what Rewire does. We build AI systems that help to make the process of content moderation more efficient, more effective,
00:01:30
Speaker
both so that we can save money, because at the moment a lot of platforms are using humans to do this work, which is not a great way of solving the problem. You know, humans are very slow, they're very expensive, they don't scale, but also they suffer really serious harm. And so we don't just see this as being something which is driven by the money, it is also driven by the social impact and the challenges that people face. We also believe truly that AI can open new possibilities for how people are kept safe online. There are things you can do with AI
00:01:58
Speaker
that you cannot do if you're just sort of manually looking for content that might be violating your terms and conditions. So it's quite an exciting space to work in and of course we see a lot of this in the news at the moment with everything that's happened to Twitter and Facebook and it's a yeah it's a very hot area to be in.
00:02:15
Speaker
Yes, it's a massive area and I think some other bits will come on to you later in terms of the areas you're looking at with AI around customer service. But just take a step back a bit. What is AI or artificial intelligence and how is it different from, let's say, machine learning or just general computer software?
Understanding AI, Machine Learning, and Software
00:02:34
Speaker
This is one of those really never ending debates. So my background was as a researcher. And so people endlessly debate whether it's AI or it's machine learning or is it just statistics, but it looks a little bit fancy. And, you know, I think broadly we try to think of it as just technology that's automating processes now.
00:02:53
Speaker
The extent to which that's truly mimicking or reflecting or replacing human intelligence, I think those are very open-ended questions. Obviously, we've all seen recently OpenAI's chat GPT, which has just been absolutely amazing, and it's really started to, you know, it seems to be truly meeting human levels of writing, which we've never had before. We've never had generative AI that can do that. So it's been a huge breakthrough.
00:03:17
Speaker
But, you know, does that mean that AI is actually thinking for itself? No, you know, it's still not thinking for itself. It's still a deterministic computer program. So we're using AI in that sense of anything which can automate the process of content moderation. Yeah, that's what we've been looking at is AI, you know, we're at the stage of doom and gloom where AI will take over jobs or actually is it still there just to support and it's just a computer sort of program, I suppose, you know, doing things for you.
00:03:46
Speaker
I know you deal with AI mainly in communities but and I suppose more specifically within the confines of social media but there are other types of communities where you know AI fits in and works for you as a solution you can help with. I mean yeah I mean the way that we see the market really is that anywhere that you have
00:04:08
Speaker
people sharing content, interacting with each other, engaging, then you always have a risk of harm being created. And sometimes that risk is incredibly low. Like there are platforms that we've seen which are sort of internal message boards. And yes, people are sending messages to each other or like a Slack workspace. The risk of serious harm is low. It's never zero, but it really does start to become quite small. But there are other spaces where it's incredibly high. And so you have a chat
00:04:33
Speaker
chat service which is aimed at children or you have a live audio feed when people are gaming and so a lot of that is children but not exclusively and that really can create quite a serious risk of harm and in all those spaces I think AI can come in and help because really it's just a question of scale you know if you want to make this really simple it's
00:04:52
Speaker
You have so much content being created in real time. There is no feasible way to go through and check that. And then, you know, all the alternatives that people have proposed just become absolutely infeasible. Like you want to have a moderator check every video on YouTube before it gets posted.
00:05:08
Speaker
Yeah, that's not gonna work. People are gonna stop using YouTube straight away. And so something that we think a lot about is proportionality. And how do you make sure that whilst you're trying to protect from harm and build AI that can improve that process, you don't go too far the other way and you don't start ruining user experiences, shutting down free speech, kind of adding friction where you just don't need to.
00:05:29
Speaker
So one of the things you mentioned there was AI moderating chats. So is there something coming where AI will help with moderating customer service?
AI's Role in Customer Feedback and Moderation
00:05:39
Speaker
So for example, before you open an email, you'll have an idea of the sentiment of what's in that email and what type of content it's got. So is it a positive email that you don't need to prepare as much for? Is it a highly negative email that you might want to mentally prepare for rather than sort of opening it and just being a bit taken aback and shocked?
00:05:58
Speaker
Is that where we're going in terms of moderation there?
00:06:01
Speaker
Yeah, I mean, it's sort of fascinating because we did some work recently with Deutsche Bahn, which is the national railway operator in Germany. And they have this problem that they get a lot of real time customer feedback. And that's really helpful. So that's customers either they have a sort of like an icon, which you can scan on your phone when you're on the train and just straight away give them feedback. They have forms, they have Twitter and other social media profiles. And so it's really easy to get in touch with them and say, look, this is how I think your service is.
00:06:29
Speaker
The reality is that most people spontaneously give customer feedback when they're either incredibly happy and something has gone absolutely amazingly well, which is quite unlikely with trains because a train doing an amazing job basically means that you don't really think about it and so it's very rare.
00:06:45
Speaker
Wow, you were on time. That was fantastic. I mean, that doesn't happen. Oh, and they're really annoyed. And when they're like, you know, guess what? Your chain was late and I missed my next connection or I missed something really important or just, you know, the service was horrible. So they get a lot of I mean, we can't put a number on it, but they get quite a lot of toxic, angry people.
00:07:05
Speaker
who are writing into them. And the challenge is how do you separate that out so that A, your staff don't have to look at this stuff when they're not really ready for it. You know, they're just quickly looking at the feedback and you see all this horrible, and sometimes it's incredibly explicit, sexually explicit, sorry, almost violent content. You know, it's not really very nice to look at.
00:07:22
Speaker
But B, how do you also make sure that you don't just throw all that feedback away because it's so horrible? You know, there's lots of useful information in there. There's lots of helpful signals about, you know, more substantive criticism and useful feedback on
00:07:36
Speaker
where you could actually improve the service in a meaningful way. And so we wanna help them be able to capture that. So we've been doing some work with them to help filter out that sort of harmful content, the toxic feedback and help them to make sense of it so that they can actually derive some value from the more useful critical feedback. I suppose longer term then, is it something you see being used by SMEs or will it be, I suppose, more exclusively used by the larger firms who just have massive volumes of inquiries?
00:08:04
Speaker
I mean, this is a big question that we ask ourselves all the time is where is the market for this? Is the market for this really with the bigger firms? Because until you get to a sort of scale that's more than a hard for one person to go through and check, then however unenjoyable that work is, most firms just won't put the money into AI. They'll just put the money into hiring a member of staff and they can sort of handle it. At some point it becomes too much. It just becomes way too hard. So the big social media platforms
00:08:32
Speaker
They're always going to have to use some AI and even if you're like a moderate size company and you have a lot of customer feedback, there's a lot of inbound messages on your socials, you're probably going to need to use some AI to process it. I'm not sure where the bottom of that market sits and whether
00:08:48
Speaker
Smaller organizations will really want to use it because there's also a reality that AI is relatively expensive. To build bad AI and to get something off the shelf that's like complete rubbish, cheap, cheap and easy. And there's a thousand providers out there building pretty shoddy AI products. If you want something that's good, that works as you want it to, that can be live updated, that understands nuance and context and actually starts to match that human level reasoning,
00:09:16
Speaker
you need to put some more serious money into it. And even though OpenAI's chat GPT is amazing, by itself, you can't just point it at your customer feedback and say, right, tell me about this customer feedback and explain it all to me. It's still not quite there. You still need someone to come in who actually knows about the setting, knows about the problem, and get them to help.
00:09:36
Speaker
I'd be interested to see how that market does evolve, and as you say, where it eventually bottoms out. One of the things and the questions I have that I think it relates to surveying is, surveyors have to explain different services to different people at different times, and people have got different motives. For example, you might have a first-time mover, an up-sizer, or a down-sizer.
00:09:59
Speaker
But with everyone, there's sort of, you know, there is jargon within the process and there's the challenge to, I suppose, keep things simple and straightforward. When you're trying to market something like rewire an AI product, it's very new, there's lots of new industry terms and all that kind of stuff. Is there a challenge in getting your message across to potential customers and cutting through that sort of new language?
00:10:26
Speaker
Yeah, it's it's really challenging. I have to say, you know, we always have that problem of how do you simplify things which for us are kind of bread and butter, you know, so we'll throw around terms like, oh, you know, the precision, the recall of the model. We'll talk about gold standard labeling. We'll talk about the annotation process. We'll talk about inter annotated agreement.
Simplifying AI for Non-Technical Audiences
00:10:44
Speaker
And our team don't really think about it. It's second nature to us because we've worked on it for a few years and we do it every day. But for a lot of the customers that we deal with who tend to be non-technical, I will say that as well. That's kind of been a learning point for us that a lot of our customers are not deep technical experts. They have it best to kind of work in familiarity with some of these things.
00:11:02
Speaker
You have to really take them through what it means. And I think it's always the so what. We're talking about, say, precision, which is a metric to assess an AI model. So basically it tells you of all the things which your AI flagged as being hateful, as being toxic,
00:11:20
Speaker
how many of them actually were, you know, how many actually were, and it's, you know, it's a subtly different metric to accuracy or to recall. So, you know, you always have to be a little bit careful with how people interpret it, but we want to give the say what, we want to say, you know, this matters because if our precision is low, that means that lots of the things we're telling you are hateful or toxic, they're not, you know, you can't really trust the model, you can't trust the model results. And I think that's always been really helpful to us. And actually, it often helps us to not explain things that we don't need to.
00:11:49
Speaker
because we start going, right, well, we could go into this of, oh, like your F1 score is this, but it's like, but what, you know, the say what isn't there, the customer's not going to care. Like we care about it because it masses to us for sort of deep technical reasons. For them, they don't need to know this stuff. They need to know the sort of things they need to, which is obviously untrism, but it's something we do try to think about quite a lot with the more technical areas.
00:12:13
Speaker
So do you try and avoid jargon words entirely? I mean, is it best just to use plain English in all aspects, or is there a benefit to using technical jargon sometimes? One, to show knowledge, but two, because then you can go in and explain that in simpler terms. So when you're focusing on things in plain English, are you looking at the what and the why we were talking about earlier?
00:12:37
Speaker
Yeah, I think it's trying to find a way as well of communicating it without talking down to someone. So being like, you know, maybe you guys already know this, maybe I'm telling you things you know, but still, I just want to be really clear when I say this, this is the meaning of it. And I think that's been quite helpful. But obviously, if you do that too much, it starts to feel like a lecture or like a seminar or something. And that's not very good for a client call. But sometimes we do that just to make sure everyone's on the same page. I think
00:13:05
Speaker
We try to sort of say, look, here's the result. Here's the, so what? Here's the thing you need to think about. Now, if you want to go five levels deeper and really push us on every single assumption and all the detail, we have that. Like we have actually done that thinking. We have it say in the appendix to the deck. We have it in this very, because we actually write a lot of open source, open access research papers, which we publish. You know, it's all in the paper. Like it's all there. It will even try and share some of the data. So if you really wanted to recreate this, they'll obviously
00:13:36
Speaker
you know, if you want to recreate our work, then kind of what's the point of getting us to do it in the first place, maybe, but sometimes people do, they just want to know that they can recreate it. We say, yeah, look, absolutely, it's all there. And I think just saying that you have that and being willing to be transparent and open, that instills so much confidence that it is very rare that people really want to push us all the way down to the bottom. Though I have to say that we actually like it when they do, because we're like a bit nerdy, and this is kind of why we got into the games, because, you know, we enjoy those difficult, challenging problems.
00:14:06
Speaker
Okay, so that turns into another question I had. So, you know, every business has got a core service and core strengths and core offerings. And, you know, there are times where you might try
00:14:17
Speaker
a new service or a different angle on a particular service. You made a very clever move, I thought, moving from primarily being online social harm-based in terms of the AI products to, I suppose, repurposing in some respects and focusing on how you can help with the customer service angle.
Rewire's Strategic Pivot to Customer Service
00:14:37
Speaker
How do you go about stopping and thinking when is right to try and pivot and move on and try something else and what I suppose commercially makes more sense? What's your process there?
00:14:49
Speaker
Yeah, I mean, this is something we think about a lot, I've got to say, and I definitely do not have the answer to this, but it's a very live conversation internally, is basically corporate strategy. And that feeds very closely in, because we're such a small organization, into product strategy as well. But thinking about what is the real pain point that we can solve? And I think initially we thought, right, we'll just build AI. People are doing this with humans.
00:15:15
Speaker
Perfect I just built AI to replace the humans and that's just like a classic kind of business digital disruption. You have this slow boring manual process that's thrown some technology and see if we can fix it. We then realize that actually the way the market works and the way that people's workflows are organized meant that fitting in that AI wasn't actually possible for most companies out there most.
00:15:36
Speaker
Most organizations don't have the scale, don't have the workflow, don't have a budget really, don't have the sort of maturity of thinking about their trust and safety to start just chucking in AI. What they want really in many cases is an orchestration tool, a tool that will help them to just manage that workflow which we weren't building.
00:15:54
Speaker
So, then we kind of thought, okay, so who does need AI? Well, the big platforms do, but actually they build a lot of this themselves internally, and they have fantastic researchers, and we know a lot of those guys, and we have done work with Facebook and Google, and we have kind of done some projects with them, but we know that's not a huge long-term opportunity because, you know, they've got amazing people. They don't need to come to us to do that for them.
00:16:14
Speaker
We started to just think much more carefully about who is actually suffering from this, for who is it a real pain point, and who has money to spend, and where can we build something that they can't easily replicate, it would cost them two inches to build themselves. That did take us down the customer service route. It took us that space of every business wants to understand the customers better, every business has this problem of toxic, undesirable feedback coming in. If we can help them to filter that and make sense of it,
00:16:42
Speaker
you know, we have a big marketplace and that we can chase after and we can go for. So, you know, I think we still see ourselves as an online trust and safety company, but I think we're just being a bit more flexible because until we have that product market fit where we can go right, you know, we've got like a lot of companies who really want to buy this product. And if they didn't have our products, they absolutely would be super upset.
00:17:03
Speaker
you know, we just need to keep pivoting, keep exploring things and just being very honest with ourselves and very critical about where we actually add value. And maybe I'd say as a reflection a year ago, we didn't do that enough and now we really are.
00:17:15
Speaker
That's a good point. What's enough and what's too much? Do you find there's a sensitive balance in terms of how often you review things? Because you don't want to overreact and stop doing something before it's had a chance to bed in. Yeah, I mean, this is such a tricky thing, isn't it? If you just pivot all the time, you've never really explored an option. You've got a couple of data points of just the first few people coming back and saying, nah, not interested. And you've given up. And it's like, well, part of being a successful startup has been quite tenacious and saying,
00:17:45
Speaker
OK, but why? And maybe it is there's a reason you can address or maybe it's you still haven't quite found your right niche or segment. We didn't speak to the right person or there's so many factors which go into that. I always remember the Y Combinator story of Stripe where Stripe did this payments process very early and they were very effective at getting other Y Combinator startups to sign up because they didn't just say, look, here's a link.
00:18:08
Speaker
go sign up, go give it a go. They actually went over to people in the offices and said, right, you want to use it. I'm going to go on to a laptop for you and I will set you up and I will get you using this. And that slightly aggressive strategy has to be used as well, I think, for any startup. And if you pivot too often, you just can't pursue that. You know, you're moving around too much. So.
00:18:27
Speaker
It's hard. Yeah, it's very hard. And everything you do is kind of opportunity cost, right? Like by spending a couple of days exploring one option, that is a couple of days you've lost. And you've got to really think about that carefully. Yeah, I think the flip side is that if you if you don't take that time out either, then you've got the opportunity cost of missing out on something and potentially not realizing there are, you know, avenues you could go make more money from.
00:18:50
Speaker
I suppose from a customer service perspective, and we've been touching on that a little bit through this in terms of AI, do you think we're going to get to a point soon where AI will replace the customer service function because it's good enough to analyze the sentiment of what's coming in and some of these things that can write responses to those problems?
00:19:11
Speaker
Or is it something that will really be there to help analyze and look at sentiment and provide overall analysis, but really the support will still come and the value will still come from people within the process?
AI Integration for SMEs
00:19:23
Speaker
You know, who knows where things will go because there are some amazingly powerful AI tooling coming out. And I think some of the stuff that we built is capable of making human level decisions in some contexts already. And, you know, with more money, more funding, we can really start to push into that a lot more. I think the reality is that most AI and I think most responsible AI will be about augmenting and supporting human activities, not replacing them.
00:19:48
Speaker
because it's just too much to expect the AI to do it. And you should always have governance. I mean, if you're automating every part of your process, where's your human touchpoint? Where's your human just checking that this makes sense, that you're not making some catastrophic error? I mean, we deal with a fairly, I'd say high profile contentious issue, online safety, people expressing
00:20:06
Speaker
hateful language or abuse or terrorism in some extreme cases, but also making sure that we protect free speech and that we don't invade people's privacy through the AI that we're training. So it's very important that we kind of make that judgment call correctly. I'd never want to see a system given the performance of AI now, where you don't have a human coming in and checking and saying, actually, is this sensible? Like, are we making the right choices?
00:20:31
Speaker
And I think AI, when it's used at its best, can free up your human expertise to do the thing which humans can do, which AI can't do. You can't build a relationship with a customer. You can't understand their problem in depth with AI. That is just not what it's there for. That's where you want humans and people to do it. So I think those jobs, I don't see going. And hopefully with more AI, we can actually free people up to really do that more.
00:20:55
Speaker
Regarding support, I suppose in terms of supporting people, at the moment, I suppose the use of it is reactive. It's reading what's coming in from customers and looking at sentiment and so on. But will it ever move to being a proactive system where it's analyzing what you're about to send out and suggesting where something might get misconstrued, for example, and therefore get an undesired response from the customer because they've taken something in a way that wasn't meant?
00:21:24
Speaker
People, yeah, people who tried to make these tools that she has become a sort of popular area and like almost giving nudges. Like someone's just about to hit send on a message or a tweet kind of going, oh, are you sure you want to say that? Like, is that a nice thing to do? There was an app from the BBC, which did this. It was called the Own It app. And it was, you could download it on your phone. It was aimed at children. I will say that. And it would then just give them, it was like sort of a keyboard overlay. So it would give them a little nudges and saying, well, do you really want to say that? Is that a nice thing to say?
00:21:53
Speaker
The problem with it is that A, it makes a lot of mistakes. So, you know, you can't, it's not always going to get it right and that can annoy people. And obviously B, do we want that? Do we want nudges? A nudge is a good way of solving this problem. Are there an appropriate way of solving this problem? Is it not a bit nanny states in certain cases, especially for adults?
00:22:13
Speaker
But I think Grammarly-style tools for hate speech will be coming in. Of course, the one big challenge is that the people who tend to be most concerned about hate speech and who really want to solve that problem and who would, you know, think a Grammarly-style tool for hate speech is a good idea are often not the people who would be spreading hate speech in the first place and the people who actually are going to be spreading hate speech really would never, ever want to use that tool. So getting people to use this tooling can be a bit of a challenge.
00:22:41
Speaker
So what in terms of AI is available to SMEs right now? I mean, I know there's things like chat GPT that can help automate blog post content creation and stuff like that. And I've mentioned that previously on other podcasts. But what tools are there at the moment that companies can use to, I suppose, automate processes and speed things up, but keep that personal touch?
00:23:08
Speaker
Yeah, I mean, I think it's the best thing to do is to be problem driven is to say, what's the problem I'm trying to solve? What's the constraint I face or the challenge or where I am being overrun with too much say inbound content or something like that. And being really clear about what you're trying to do. Because I certainly have seen a lot of organizations just say, how can we embed AI in our process? And that's not a great way of thinking about it. That's been very solution driven, just kind of, you know, looking around for a nail, which you can hammer into. And I think it's,
00:23:34
Speaker
Yeah, it's not a great way of doing it. It's much better to start from the problem, be very clear about that, and then see if AI is the right answer. And in many cases, it's just not, unless you've reached a certain scale or size. I think once you do reach that scale and size, it becomes really important, and AI is like an amazing cost-saving tool.
00:23:54
Speaker
I think the next thing to think about is what skills do you have? It's good to be very open with yourself. If you're not an AI expert, if you're not that interested in it, can you outsource it? Can you bring someone in who could do that for you? There's plenty of companies who do outsourcing AI dev.
00:24:09
Speaker
or can you use a totally no-code solutions? This is quite an exciting area actually within computer science for the last couple of years, is recognizing that just writing code is a huge barrier to most people using any technical system, whether that's advanced AI, which is generating human-like text. If it's just a really simple predictive model that tells you, which customers should you prioritize this month? Well, you could use a very simple AI model to help tell you who you should be reaching out to.
00:24:40
Speaker
Any of those implementations you can now do without any coding experience. There are lots of cool tools out there and they just completely remove that barrier. So I think I would definitely be looking at that because the true high end of AI, like the amazing, fancy, latest, state of the art stuff.
00:24:57
Speaker
You know, you can get very good stuff, which is very close to that, but has none of the kind of technical barriers. You know, you won't have the absolute best AI in the world, but you might be 95% of the way there and you only had to spend a little bit of money and didn't have to code anything, which I'd say is like a pretty good kind of trade-off. Perfect. So I suppose moving on from the AI perspective, as a more generally as a CEO and co-founder, let's say you found your market fit
00:25:24
Speaker
It's starting to grow, but how do you really start to scale whilst maintaining quality of product and customer service and so on?
Scaling and Talent Acquisition in AI Startups
00:25:33
Speaker
How do you find the right people to bring them in and what are the keys to success for that side of things?
00:25:37
Speaker
Yeah, I mean, I think the first one actually kind of goes back to what we talked about earlier is around product and corporate strategy and just being really honest with like, what is your strategy? What is your hypothesis on the market? And I think that term hypothesis can be overused a little bit, but it is really helpful because then you start to think about, okay, well, what evidence do I need to support this hypothesis? Our hypothesis was that
00:26:00
Speaker
people want to automate their process of content moderation by using AI to replace humans. We actually kind of discovered that wasn't quite right. And as we dug into it, you know, it was a little bit more complicated. So, and again, if we think about, you know, when should you pivot? When should you not pivot? If you have a very clear hypothesis, like I think this segment will really want my product, what evidence do you need and decide that in advance before you go start doing it. So maybe my goal is to speak to 500% of your customers. If none of them express any interest, I know I need to move on to the next thing because clearly there's no demand.
00:26:29
Speaker
So I think that's been a huge thing, understanding the value proposition and testing hypotheses in a very structured way. I think the second thing is team. You can't be a good team. It is just the most important thing in the world. Sometimes great people come in and you don't quite know what the role would be for them, but you know they're great. And so you can find a way of bringing them in. You have other people where, okay, you need someone really technical because it's a technical role and they do have to understand what they've actually got to do.
00:26:56
Speaker
But that doesn't mean they're a great long term team member if they don't totally get your ethos and what you're trying to do. So we do sometimes bring people in on shorter term contracts because they have the right technical skill set, but they're not necessarily totally on the team kind of vibe. And I think, you know, team fit and just being very honest about that is.
00:27:14
Speaker
is important and it's fine. It's something when we're saying to someone, look, you're great. We really like working with you. However, we don't see you as our number three employee because that's a super important position to us. And then finally, it's just flexibility. You've got to be flexible. We definitely, I think we're a bit too fixed when we got started. We've been in this space for six, seven years working on online safety as researchers and as advisors.
00:27:39
Speaker
and maybe we weren't responding quickly enough to the market signals and the feedback and now we're starting to do that and actually it's been incredibly helpful. So yeah, that's been really key to us, but we're still trying to work it out, I think.
00:27:52
Speaker
It's never really, we've never totally nailed it. Yeah, that flexibility side is interesting because you have to be able to adapt to market conditions and also what feedback you're hearing in terms of, you know, as a common interest, what do people want from the solution? But equally, you have to have built enough of your core product and really solidified that before you can start.
00:28:12
Speaker
getting distracted with other things because otherwise the core offering just isn't there. I think another aspect I wanted to talk to you about was recruitment because in surveying, one of the difficulties has been there's a shortage of surveyors, there's lots of surveying firms that have been very busy and needing to attract surveyors into the business.
00:28:34
Speaker
equally, there just aren't that many spheres to recruit. So I imagine in AI, it's the same thing. There's a skill shortage where you've got people that could go into AI, they could go into web development, they can go into cybersecurity. What are the keys that you found for overcoming that shortage and attracting talent into the business?
00:28:54
Speaker
Yeah, huge problem, huge problem. You know, people who do in the UK nationally has a massive shortage in AI and data science skill sets. We know that the UK is not training enough people even to meet current demand, never mind future. And so that means that people can charge a real premium. They come out of university, even an undergrad, and they can charge really serious salaries, especially if they're London based.
00:29:15
Speaker
So it's hard. I think we've been really fortunate that we have a great network of PhDs and researchers and people from Oxford and Cambridge and Turing that we've worked with for a long time who are happy to come join us. Actually, Queen Mary has probably been the university that we've some reason had the most people come through from. But it is difficult. And I'd say that once you find someone who's really good, it's okay to be open with them, and especially as a startup for a small business and just say, look, we don't have a ton of money to pay you, but there will be other benefits that come through. Equity is obviously a great way to incentivise people
00:29:45
Speaker
And I think for us, we have a clear social mission. So that does appeal to people. They understand that we are trying to have a positive impact. I mean, we are a full profit, but still we are this very clear social mission. And the work is really exciting. We get to work on cutting edge problems with some great clients. And especially for junior people who are trying to develop skills, I think that makes a big difference. I think not everyone is just immediately trying to maximize money. They value these other aspects too.
00:30:13
Speaker
But it's hard. Yeah, I wish, you know, it also just takes up time, right? It's just a big time chain trying to find people. Yeah, it's funny you say that I was reading an article the other day, I think on LinkedIn that was basically saying employee engagement goes and longevity with the company goes beyond salary and the purple subscription. It's much more about the value is aligning and feeling rewarded by a team and training and
00:30:38
Speaker
and stuff like that. So I think there's probably an opportunity there, or there is an opportunity there, firms who they know their value proposition, you know, their, why they exist, what they're offering to customers and why. And I think if you know that, then you can probably better attract people and not have to worry about winning purely on a, you know, on a budget level.
00:31:00
Speaker
and engage people that way. Yeah, I mean, you know, I mean, money is important, right? You've got to be realistic. People do want to earn like a decent wage and they're never going to take a huge kind of cut to come to you. But yeah, I think we've been, I think we've been kind of fortunate. We definitely have some great people come into the company. I mean, it's, it's, yeah, it's really fantastic when you find people who just totally guess it, like they get the mission, they get our style of working.
00:31:23
Speaker
And it's literally just like having someone who can take work away from you and come up with things you would never have thought of. And that's where you're sort of like in a really sweet spot. But yeah, it's definitely a big risk concern. I'm not quite sure what the right term is, but it's definitely something we think about a lot.
Growth Strategies: Fundraising and Market Positioning
00:31:40
Speaker
Well, thanks for coming on today, Bertie. And I might just put you on the spot before we go and ask, what are your top three tips for success in growing? Oh, gosh, that's, yeah, that is putting me on the spot. I dunno, no one is like, think about money early on. Think about like, are you raising money? Are you taking out debt? Are you bootstrapping? And just really get into that straight away and actually raise early, maybe. Like the longer you wait to raise, I think the harder it gets.
00:32:08
Speaker
You want to have that network rolling. You want to be having those chats early on. Investors give amazing advice. Some of the best advice we've had is from investors asking us tough questions and making us rethink what we did. So I think that's just a big thing. Number two, like just be relentlessly focused on your customer. Like, you know, it's that I think that's sort of saying about one, 10, 100, 1000, get your one customer, get your customer who if you shut down the next day would be really upset because they absolutely love what you're doing and you become essential to them.
00:32:38
Speaker
and then get your 10, and then get your 100. And those are roughly as hard as each other. So just getting that first customer, not easy. So I think that's the next thing. And then, yeah, I don't know. Corporate strategy, product strategy, kind of things I've been saying the whole way through. Just be very honest with yourself, and then be honest with your investors, and honest with your stakeholders, and develop a sort of nuanced, sophisticated position on the market, and why you're the right people, and why you're building.
00:33:07
Speaker
even if it's not the fanciest thing in the world, is the right thing to be building and is what people will actually pay for. Awesome. I think they're good points. I'd agree with those. Yeah, thanks again for coming on today. I look forward to catching up soon. Yeah, amazing. Thank you for having me. It's been a real pleasure.