"Can Me Mam Use This?"

Episode 14: Steam Engines, 100 Out of 100 & Can Me Mam Use This? 🍺

Week 14. Veterans now, apparently. Kieron is off to pootle on the Thames on the beige boat of joy after this. Neil is enjoying the last few days of summer in the Lake District before the cold winds of June blow through. The pantomime horse of a podcast rolls on.

The steam engine analogy — and why it changes everything 🏭 Kieron opens with a cracking insight from behavioural scientist Rory Sutherland of Ogilvy. When electricity replaced steam power, early adopters simply swapped their giant steam engine for a giant electric motor — and changed nothing else. The efficiency gains were marginal. The real prize only arrived when someone realised electric motors could be miniaturised and built into each individual machine, allowing the entire factory layout to be redesigned from scratch. The assembly line followed. The analogy for AI is exact: most organisations are still swapping the steam engine. They're bolting AI onto existing workflows and getting 10 minutes here and there. The transformational gains come when you rethink the factory entirely. Not many organisations are there yet — but the ones that get there first will be very hard to catch.

AI curious, AI opportunistic, AI first 🎯 Neil's three categories of organisation, drawn from Gartner. Most are curious — trying one thing, dipping a toe in, not yet convinced. Some are opportunistic — using AI where it obviously helps. Very few are AI first — genuinely rethinking how they operate with AI at the centre. The gap between AI curious and AI first is widening every week. And the AI deniers? Still out there. One of them is close to Neil. The doghouse awaits.

Moving to GPT 5.4 — the nuances nobody warns you about 🔧 Leading AI has been migrating customers to GPT 5.4 mini — cheaper on tokens, better at reasoning, faster. But the move from 4.1 hasn't been frictionless. The same prompts behave differently. Tiny changes in how a model interprets an instruction can cause it to pull slightly different data, produce slightly different structures, introduce subtle errors that are genuinely hard to spot. Kieron has spent most of the week in the weeds of it. The quality of the writing and reasoning is better — but getting there takes real expertise, meticulous testing, and a lot of prompt rewriting. This is exactly why "our IT team can build this" is a dangerous sentence.

Prompt engineering is so 2025 — Nate B. Jones 💬 Neil references Nate B. Jones' latest Substack: in 2025, you prompted AI like a smart intern — detailed, structured, prescriptive. In 2026, with the newer reasoning models, treat it like a strategic advisor. Give it your hypothesis, your context, your data — and ask it to challenge you, push back, and find things you haven't thought of. Kieron accidentally proved this the day before: his pricing strategy prompt included "I don't know what I don't know, come back with other ideas" — and the output was brilliant. Neil's verdict: you were already doing 2026 prompting without knowing it.

100 out of 100 🏆 Neil's BidWriter story of the week. A customer used KnowledgeFlow BidWriter to produce a first draft, then used the built-in evaluation scoring to identify weak sections. They refined the weak areas, ran it back through evaluation, consistently hit 5 out of 5 for every question — and then the actual bid evaluation team came back with 100 out of 100. A perfect score. On a competitive tender. Neil has never heard of it before. Neither have we.

Healthcare and the move from single tool to platform 🏥 The public sector customer conversations are shifting. It's no longer "can I have Policy Buddy?" It's "how do I give a whole cohort of people — across primary care, secondary care, physiotherapy, any healthcare provider — consistent knowledge and consistent messaging?" Linked to the work Leading AI did with North East Lincolnshire and the Care Plus Group on multilingual cancer information. The private sector version: how does an organisation become the trusted hub of knowledge for their customers — explainable, reliable, verifiable? Context layer. Data in order. Answers that can be traced back to source. That's the differentiator.

The future of education assessment 🎓 Kieron thinks big. AI detectors don't work. Handing in written work is already mistrusted. But nobody's talking about the real prize: AI can assess on everything, over a whole career — every essay, every contribution, every presentation, every rugby match. A data file on each individual that tells employers far more than a grade in an exam taken on one morning when they weren't feeling great. No more failing people who happen not to suit academic-style assessment. There are vested interests in the way. But it needn't be like this anymore.

"Can me mam use this?" — Chris Quickfall from Cognassist 🏆 Kieron gives a well-deserved shoutout to Chris, CEO of Cognassist, who was speaking at the FE Tech conference this week about the upcoming SEND white paper and the growing neurodiversity challenge in colleges. His product design test: his mum is a teacher, not great with technology. If she can use it, they ship it. Four words. Best product design philosophy of the year.

Kieron is pootling on the Thames. If you see the slow beige boat of joy, give him a wave.

Two mates. A bar. Thirty years of business between them. And all they want to talk about is AI.

Pull up a stool — we'll get the beers in. 🍺

Transcript

SPEAKER_01 0:00

Right, shall we, Kieran? Shall we get this Pantomime Horse for podcast underway for episode 14?

SPEAKER_00 0:07

Look at us. We are we're veterans at this now and probably largely run out of anything useful to say.

SPEAKER_01 0:14

I disagree entirely. I disagree entirely. I think our audience will be really interested in this week's topics if he's not asleep already. He's already nodded off because we've bored him already. He's like, oh yeah, I've listened to their dull sit tones. I'm really I'm into my REM section now.

SPEAKER_00 0:34

Well maybe we'll be infiltrating him like osmosis. Exactly. And then uh then he'll he'll wake up all more much more knowledgeable and feel more refreshed than ever.

SPEAKER_01 0:44

All right, what's on your list for this week, Philip?

SPEAKER_00 0:46

Well, I want to talk about a great analogy about um uh AI and how really getting AI first and rethinking workflow. I'm gonna come up with that in a moment, which I think is fantastic. I was at an ed tech conference, I've got some things to share from there about assessment and AI and assessment cheating, really, in exams, but also how can you benefit from AI and assessment? Um, and I want to talk a bit about uh the challenges we've had this week moving to 5.4 mini. So we've been moved our model, as you know, the large language model that drives knowledge flow on our journey towards a multi-model, multi-model switching, model switching. Um, but right now we can get 5.4, so we've been pushing everyone to that. And there's been some interesting um challenges. So I would like to reflect on some of those. That's me.

SPEAKER_01 1:34

What about you? Um, I've got uh three customer stories that I I wanted to talk about from conversations I've had this week. Um, I wanted to talk a little bit about um product differentiation, the kind of the whole copilot versus knowledge flow thing that we come back to on a regular basis, but also wider than that, um some some things again reflecting on the Gartner conference from last week um and and how do we how do we differentiate ourselves and then kind of link to that a couple of new features that we are implementing, which I think will make um a massive difference to our customers. So um, yeah, those those three things are on my mind, but why don't you start with your uh your exciting analogy then first? Let's let's hear that.

SPEAKER_00 2:21

I will. Well, this um just to give credit where credit's due is the great Rory Sutherland, uh behavioral scientist, um Ogil V chair, I think he's is his uh role still there, um, love his stuff. Um he was talking about AI and uh the industrial revolution, which is often used as uh uh as lots of people will know, as the kind of AI is the next industrial revolution. Um and he was talking about the advent of electricity and steam-powered world. And so what used to happen, and uh and I have researched this to check that it's true, um, that so a factory would be set up with a massive steam engine at a kind of you know one end, if you like, and a huge crankshaft coming out of that running all the way through the factory, and all the machines would attach to that crankshaft and do whatever they do with with that. That led to obviously huge amounts of accidents because ultimately you're always on machinery, which is sounds terrifying. Um, but also just a huge amount of inefficiency because the very layout of the factory was determined by that single crank shaft giving you power, not how you would optimize if you were you know doing anything else. Then electricity came in, and uh the the early adopters uh were jumping and were changing their giant steam engine for a giant electric motor. So that now driving your crank and nothing else changes, and pretty light optimism uh uh optimization um and not much efficiency savings. When people realize that electric motors could be made much smaller than a steam engine, apparently a small steam engine. I'm no expert, but is pretty useless uh in terms of power, um, uh, whereas electricity can do their thing. So once they realize they could shrink them, then they could build them into each of the machines in the factory. Now you can change your layout of your factory, you can turn them off where they're not being used. Um, and it's arguably, of course, the research part of the reason assembly lines, the production line happened because you could now do that and move it around as you wanted to in the factory. The analogy, of course, is we are at the moment very much in the space of replacing the big steam engine with AI, putting AI into your existing workflow. Um, and there are efficiencies, but they are not transformational efficiencies. They are 10 minutes here and there, a couple of hours a week, which is great for well-being, but not enough to really make a big difference. So I I love the analogy as a way of talking about how when you rethink your factory layout, that's where the prizes really are, and that's where probably the organizations that succeed in an AI first future are likely to be redesigning and just rethinking the way they do business completely. So that was the that was that.

SPEAKER_01 5:05

Interesting. I think uh I think a few things. One is the um uh I like the idea of uh uh replacing just that transformational idea of of transfer transferring from one big machine into lots of small machines. You know, we come across that all the time, and and yeah, you'll hear everyone will have heard me say, you know, oh I've got co-pilot. It's like yeah, copilot can't do that. Actually, what you need is an individual feature with an individual solution to a specific problem. So whether that's policy buddy or whether it's uh bid writer or or or whatever else, you know, you fix the problem. Um, uh and that's what makes that's what will really make a difference. So I think it's a really interesting analogy, uh, uh, and and a bit like you, I uh I'm a I'm a huge fan of Rory, and uh uh uh as you know, I uh I reference him quite a lot in in lots of my stuff because uh he's got some really interesting stuff around behavioral psychology. I didn't know. It was brilliant, it was really interesting because what he does is he comes at it from a behavioral psychology perspective, and he does that for marketing, and I think we've been trying, and we will continue to try to do that from an AI and a technology perspective, because as we have found so often, this isn't about um technology, it's about people, and it's actually how do you get the technology to help people do a better job? And I think that that quote I gave from Gartner last week where you know people aren't worried about AI replacing their job, they're worried about their boss thinking that AI could replace their job. And and that's that's true. And the other thing that I was just flashing through my mind was the whole three tiers of or three types of organisation. There's the well, actually four types. The first one's an AI denier, and we've talked about that previously, but three types of AI. I know, just keep the voice down. Three types of A organization, the AI curious, the uh AI opportunistic, and and the AI first. And and not many organizations are AI first. Not many organizations are thinking about how do I re- actually do I need a factory or how do I reconfigure my factory now that I've got these new tools. Most people, uh a lot of people I think are in the kind of opportunistic or curious phase. Um a few, it's not the majority. I think the majority are in the curious phase where I'll try one thing, I'll try policy body, or I'll try a little bit of this, or I'll try and doing a bit of workflow automation because they're not yet convinced, or it doesn't do the things that they want or expect. And um part of that comes from the models. You mentioned the the move to 5.4 and some of the challenges. So I I've not been involved in any of that stuff. So so what's what's been what's been occurring?

SPEAKER_00 7:52

Yeah, well, let's talk about that and then um I'll come back and want to talk about um assessment, uh education assessment with AI. But um yeah, so we've um moved to 5.4 across a whole bunch of our tools. So we put that through our testing, of course, run it on our sandboxes. Um, the way we kind of test in lots of ways is we'll we we have copies of people's various setups and tools that we can then test to make sure that the outputs are are matching what we expect and that everything's working. Um so then we pushed it out to a number of clients, sort of you know, we're on a sort of phased approach to it, so we're not quite being silly and just pushing everyone overnight, thank God. Because the nuances, so 5.4, we were running on 4.1, which has been great, really good for RAG. It's really effective, but people, it's almost a pressure of um fashion rather than a reality because people will see 4.1 and it feels like we're not keeping up. And and in reality, actually, the task they want doing is probably better done by 4.1 in some cases than 5.4. Anyway, we've moved to 5.4 because lots of reasons, bit cheaper on tokens, which is great, um, and uh capable reasoning, better at reasoning than four the four models, but there's little nuances which keep popping up, and I've spent a lot of time this week. It's been a frustrating week because you know how your diary goes, and you've got maybe a half hour here and an hour there, and the rest of it's all meetings, where all those little half hours and hours have been swallowed up immediately by things that we can't fix. So it's been very frustrating, and it's in the nuance, it's in the things that our testing currently wasn't easily picking up, and that will be where it's using client data with 5.4 prompts.

SPEAKER_01 9:38

I was gonna say, can you give without without naming a client? Can you give us an example of the kind of nuanced problem?

SPEAKER_00 9:44

Um gosh, so I can. Yeah, I mean, so our how do I we're just gonna be able to do that. Yeah, that's what I was thinking. Yeah, yeah, that's what I was thinking. So you get a different behaviour in um some of our quality tools. So we we um smart targets, which is my favorite tool, so it's good to talk about that. Um, I think writing smart targets and parent reports is a fantastic opportunity that AI brings to educators. Parent reports, I think, probably is the one that is ubiquitous, and we should maybe focus on that. What happens when you put 5.4 on it is the prompt is uh treated differently, it's probably the easiest way to say than the 4.1 prompt. Uh, so we knew that, and we've been changing all of the prompts in the background, but then these little subtleties change and come in where it's no longer looks at the data that it was looking at because of the way the prompts are written with some really tiny piece, and instead of pulling that bit of data which was doing really well before, it's now putting something else into there. It's like tiny little bits, really hard to spot, and it's as always the testing observability side of our world is really challenging because of that. But um, I mean it's so much better. The the quality of the writing is so much better, the quality of the reasoning is so much better. But it's working through the how we have to change the language of our prompts, and it's uh we've been doing a bunch of that using our tool AI tools to help us, but they're only as good as their knowledge, and so you know, as the AI is ultimately pattern spotting for you based on everything it's seen. Some of our world here is not stuff it's going to have seen before, and so trying to kind of preempt that anyway.

SPEAKER_01 11:24

I'm rabbiting around it, but that's yeah, no, I think it's it it's interesting because because I don't know if you've listened to uh Ned's N B. John's, another one of our heroes along with Rory. Uh his um uh Substack post last night talked about um prompt engineering is still 2025. Actually, it's have you listened to it yet?

SPEAKER_00 11:47

No, I haven't.

SPEAKER_01 11:48

I've uh I haven't got people saying that phrase 2025. And actually, what you've got to stop doing is uh thinking about um his description was people in 2025 were using uh prompt engineering like having a smart intern. You've got to check it, you've got to really uh ask uh very detailed, you've got to structure it in very uh succinct pieces in order to get the kind of responses that you want. In 2026, in the later models, and he's talking about actually, you know, not not some of the private models because you can't always access them, um, uh, and I'll talk about that in a second. But what he's saying is you now need to treat your AI assistant like a super smart consultant or an executive, because you need to challenge it and you need to say things like um I think my hypothesis is based on this data or based on this information, this is occurring. I want you to go and check this information, but I don't want you to limit yourself to that. I want you to come back with some challenges and some thoughts and and some some nuance to it. And he said, That's the best way to get um uh output from from your AI agent. I think that just that shift of thinking is really interesting.

SPEAKER_00 13:04

Yeah, no, it's good. I like that. And I love the phrase of it's so 2025, it just really makes me laugh. But it's really good. Using it as a challenge, as a thought partner, challenge what you've do, what you're doing already, it's tremendous in that space.

SPEAKER_01 13:17

Yeah, but you did that yesterday with uh some of our our stuff and and it made a massive difference.

SPEAKER_00 13:24

That's amazing, isn't it? I think so. We've been grappling for our audience. We um had a team session yesterday where we really most of it was on pricing. And it's our eternal challenge is how do we price this stuff? Because uh, you know, obviously we want to be really fair first, um, uh, but we also need to make sure that we're not losing money on customers. And so we had a two-hour team session uh with everyone pitching in their thoughts and ideas, and you had previously done a kind of survey, if you like, or at least got people to respond with thoughts and produce a slide deck. So I gave the transcript of our meeting and the slide deck to one of our tools and gave it a uh, I think a pretty healthy, useful prompt. Um, and it has taken that and just partly organized our thoughts, which is just amazing because that is really challenging, but also brought in some of its pre-trained knowledge about what you know what the world's up to, and given us, I mean, I'm just it it's brilliant because you can see all of our conversation in it. It's not like this is just you've just asked an AI tool to tell you how to price your thing. Yeah, it's it is worked with that and come out with some brilliant. Oh, I think it's actually tremendous.

SPEAKER_01 14:33

But two things I would say. One is the prompt you give it, even though you haven't listened to Knit uh yet uh you the prompt you gave it was was quite long, it was two or three sentences, but one of the things was I don't know what I don't know. So come back with some other ideas. So you're already you're already doing 2026 prompting kit. Oh well, good. I'm pleased to hear it. You didn't know. But the other thing that I thought's not a long prompt, by the way.

SPEAKER_00 15:01

Some of our long prompts are two and a half thousand words. I understand that.

SPEAKER_01 15:04

Yeah, but that's still 2025, right? Uh the other thing that I really liked about it was yeah, uh David's quote was this, and it was the best quote in the entire thing. And Donald was absolutely right about this, and Mark is absolutely right about these things. So it was pulling the context of the conversation out and inserting it into you know, here's here's three things that you need to do, and here's why. I thought I thought the analysis was just brilliant. Really interesting stuff.

SPEAKER_00 15:31

Although I had to on my second read, I had I then uh took out because every time whenever it named me in there, it would say and Kieran was quite right when he said, and um you it gives you a little glow when you're like and I had to and I had to read it again, taking out the c the glow to actually read the facts. So it's quite it's yeah, interesting, but I was conscious that those ideas that I said that it thought were great suddenly for me felt like the thing we should definitely do.

SPEAKER_01 15:57

But back back to Rory's behavioural science point, it's kind of you know, two things. One is it it was clearly manipulating you, and it I mean it clearly knew you were oh who's Ben? Who can I uh but back to the model point of actually different models are better for different things, and um you know 4.1 is great at some things, and indeed better than some of the five models for others, but people automatically assume that because it's the latest model, it's got to be better, and that just isn't true with AI, it's different tools for different things, and uh that's one of the reasons why I think the work that we've done on model switching is going to become really important. I think for our audience benefit, you know, you can't actually access some of those things in the private, secure way that we uh some of those you can't actually access some of those newer models in a private and secure way that is super important to us and to our customers. Yet those things always lag. They they always do the kind of um I don't know if they crack you cheeky and call it the cracker cane model first, don't they? They give all the sweeties and then they and then sprinkle all the all the fairy dust on top of the free models just to give you a to get you hooked. Um but the private enterprise versions at the back end, they don't always make those available immediately, and and actually it it takes some time for those to come through. Um which is really because people don't understand that. They just assume that if they can go into the free version and they can do stuff, you can do it in the private ones, but you you just can't because that's not the way that that's not the way that the um the the the frontier labs are working at the moment and uh we have to work around that. But yeah, interesting behaviour that people have just assume that the because it's got a higher number, it's a better model, and that's not entirely true.

SPEAKER_00 17:40

No, indeed, yeah. Finding the right ones are the right job, and that's hard again because then there's a whole new load of variables involved in a in a uh a retrieval pipeline or whatever. So you got you said you had three customer stories.

SPEAKER_01 17:52

I did, but we um I'll I'll I'll I'll do them. But I I did want to come back to the FE thing because you were at an FE uh tech conference this week, weren't you? Yes. And I was interested to hear about to hear about that. What the what connection?

SPEAKER_00 18:05

Yeah, and I should talk about assessment, which is really interesting to me. Um so assessment and exam cheating, and um yeah, the the real point is uh have you been cheating in exams again? Of course not. Um I don't think I have. Uh maybe one. There was there was one that springs to mind where I was maybe had another another browser open on a different laptop. I was looking at that because it said they were tracking everything across there. So I was like, well, I've got another laptop, you can't track that. It wasn't very important thing. But um, assessment, there's there's ultimately a world coming and probably already existing right now where uh a student writes an essay with AI and it's marked by AI, and therefore you kind of think, well, what the hell is education for? Um and back to that age old. I mean, that really is where you end up with any conversation about this stuff. It's like, what is education for then? Um uh but the assessment part, there was a conversation on stage about how do we how do we use AI in assessment? How do you try and get around some of the cheating? Because basically any handed-in work now is mistrusted, um, even with the unfunctioning uh AI detectors, none of them work effectively enough to be uh reliable, so you can't use those. All you can do is hope. Um, and indeed, the cleverer people are encouraging use of AI to be referenced and uh you know ultimately arguably what education should be for, that you finish up knowing how to use AI to answer a question. Uh and in the sort of like exam assessment and do they mark and can they use AI in it and all of that? There's lots of conversations about that. I think though, take the steam engine and the electric motor again. I think that all nobody is talking about the big prize. The big prize is that because AI can take on board vast amounts of data, you could assess on everything. So your entire education career, everything you handed in, every contribution you made in class, every presentation you gave, every running race, rugby club, etc. And you could create a data bank on individuals, and there's all kinds of kind of questions about data and who owns it and all of that. You could, though, right now create a data set for each individual that knows them so much better than any exam on May the 23rd, May the 22nd, when they're doing psychology at 9 a.m. And if you weren't feeling up to it, well, sorry for you, your two years of psychology learning that you've failed in the exam, that's how we assess it. So terrible when you think about it, and we needn't do it anymore. But no one's talking about that idea. And you could have this data file, you could say this student is an A-grade student. I think that would not make sense in that world. I think you might just say, here's a data file. Um, if you're an employer, ask for access to it and pick your people from that. You know, maybe, and yeah, maybe you want someone that's great at woodwork, maybe you want someone that's great at creativity. It might not be the English and math schools. I think it's the I think it will be a future at some point, but there are so many vested interests along the way that it's quite difficult to see how you get to it. Super interesting.

SPEAKER_01 21:25

A couple of thoughts spring immediately to mind. One is um when we're at the Gartner conference, it all the people uh wandering around were super helpful and friendly, you know, joking, chatting, lovely, and you know, uh having whatever your question was, nothing was too much of a problem. And you can't test an exam for that. So how how do you yeah, obviously you interview and things, but you know, you for a from an employer's perspective, having access to all of that kind of history would be would be fascinating. Nick, come back to Ned uh again, he's he's done this thing where people can load up examples of their work. To prove that they can do the AI. So it's like an online CV. So rather than you just going, yeah, I can definitely do the uh uh back end settings, you've got to you've got to put in a piece of work to do it. But that whole um from an employer's perspective, you know, we employ people, uh our other companies employ people. Yeah, the whole kind of interview process and going on people's exam results, etc., is is just such a flawed process. Yeah. And being able to naturally, I here's here's here's uh here's our company values, here's the attributes that I need in general, here's the specific skills I need for this job. They're a bit woolly and and they don't come out in in air levels or graduate um exam results. It's they're fascinating, but you're right, there's a lot of vested interest. And I don't think that's a real challenge with some of that stuff.

SPEAKER_00 22:51

Yeah, I don't think that's going to be an overnight fix. But exactly, and as you said, just that whole kind of work readiness, everybody knows that graduates are arriving into employers just totally miles away from uh sort of being work ready. Uh, and you think about the things you actually want storytelling, as you say, personable, but being personable, getting on in a team, enlivening the place, all of those things which just no grade will ever give you. And then never not to not to um forget that most a load of people are failed by education. Let's be really clear, it's not the other way around. It's like academic style education works for a whole cadre of people, it does not for a whole other set. And they've got tremendous amounts to contribute, and they're just failed from bloody 16 years old. It's terrible. Needn't be, need not be anymore.

SPEAKER_01 23:43

Crikey, okay. Evangelist Kieran for education reform. I'll vote, I'll vote for you.

SPEAKER_00 23:50

You wanted to talk about copilot versus knowledge flow and our big challenge of trying to show people why knowledge flow is not copilot, why you shouldn't think of them in the same sentence.

SPEAKER_01 24:00

Uh, I I did again, just because of I was reflecting on some of the I've been going back over some of the stuff from Gartner from last week. I had three conversations with customers this week that I wanted to talk about. Um, but uh how do we differentiate ourselves? And this came out in the um uh some of the stuff he put in last night about pricing and and where we take the organization. And our big differentiator is turns out, according to the tool, is you know, explainability and reliability. We can explain why uh a result has come back, where it's come from, and therefore the evidence of why um this this you've got this this response. And um uh that becomes so important in in trust, and the the whole piece about trust is is the new is the new commodity required for for AI uh in our in our enabled world. And the part of the co-pilot challenge is it's very generic and um it but it's good at some things. You know, I I use it not very often, but I do use it. It's in there, it's you know, Microsoft force you press the wrong button and it comes off and you have to use it. So you press the right button and you you have to use it. So it does and it does a bunch of stuff really well. But what it doesn't do is the context layer piece. It doesn't understand you as an organization, it doesn't understand you as an individual. All the things that you've put in, all the kind of, you know, I want it in this style, I want it in this format, I write in this way, I like my PowerPoint slides in green with a purple background, you know, whatever it is. And actually, one of the features that that the team have built in in this last week, brilliant, is that whole memory piece around it understanding you, remembering you, and and and uh uh it being able to think about what it is that you want and need and the style that you want and need it. Um, and you just can't do that with Cold Pilot because it's it's a blanket enterprise-wide tool. So that that that was really interesting to me. But the three customer stories, um uh uh and I'll start with I'll start with BidWriter. Um I was gonna leave it to last, but I'll start with BidWriter because that links to the explainability and reliability piece. Uh one of our customers used BidWriter for it's a procurement, it's a it's a big procurement, it's a multi it's a multi hundreds of million pound contract, and they got on a weighted score a 100 out of 100 by using BidWriter to do the first pass, 80%, and then the the team polished it and put in extra um uh information because what what they did, what we did with them was uh put in the questions, put in all of their information, it pulled out the answers, it told us where the answer was weak, where the answer was strong, what you need to do to fix the weak things, we fixed the weak things, we ran it back through the evaluation score, and we were getting consistently five out of five for every question. And I was thinking that's just too good to be true, anyway. Unless the evaluation team for this competitively tented bid were using something similar to score, which I very much doubt, um, they came back with a hundred out of a hundred, and I've never heard of any competitive bid getting a hundred out of a hundred in that in something like that. So so that would that was that was brilliant, and and hats off to them and and and good luck with that. That's great. But there's been two um I've had two sets of conversations. I'll talk about private sector and public sector. But public sector is really interesting because some of those conversations are now moving from that uh what we could call AI curious piece of I just need one thing, I just need policy body, or I need HR body, or I need bid writer. They're moving now to this whole kind of actually there's a cohort of organizations, and if we just yeah, take an example, you know, in in healthcare, you've got primary care, you've got secondary care, you've got your you've got your everything from GPs, physiotherapists, dentists, hospitals, yeah, all the all the gamut of of different um uh health services that people can touch on. So, how do you get consistency of both knowledge and message? Um, and the conversations have been actually we could use knowledge flow to train our people or and allow them to understand what it is that they can do or where they go for the best source of information. But actually, the uh how do we help our population access the right information and resources? So this is linked to the uh what we did with uh Northeast Lincolnshire and a care plus group for cancer information in in Northeast Lincolnshire and the the whole benefit of it being multilingual and being able to answer any question in any in any in any language is just brilliant. So, really interesting about how the that public sector conversation is moving from the I need a specific tool to do this, actually, how do I change the behavior of a cohort? Sort of fascinating um conversations there. The second uh bit then around the private sector was how does a private sector organization become the hub of uh data and knowledge and information, and indeed trust, how do they make sure that they become the trusted place to go for information? And that comes back to the explainability, reliability. My information came from here. This is the this is the process the AI went through to come up with this answer. And and you you could go in at any point and say, was this right or was this wrong? And that that whole that whole piece is just it we saw it at Ghana, it was time and time again. That linked to the context is king, that contextual there, you can combine those things together. I think we we do have something which is particularly powerful.

SPEAKER_00 29:59

And I think that private sector, the kind of brand public sector too, but that kind of brand trust in the brand. So I think I've talked to a few of our customers about um is is if you have a trusted data source, now that suddenly in a world where we can't really trust anything and you can't even trust videos anymore, then then a world where actually I can trust your brand, I can trust what's coming out of your brand, I think really does become a huge differentiator. Um, but that does rely on having your data in order, getting it well well organized with the right context uh layer, as we know, so that when people ask questions, they get factual factually correct answers, or at least your you know the the answers they're expecting to get. So um, but yeah, I think it's absolutely essential with and that data set I've said for a long time of having your own data set that is not available to AI, because you know, in theory, if it's behind a paywall or indeed on a on-prem or or in your own security, in theory AI hasn't already seen it and eaten it. I suspect that most of it has already but um and I think that idea that you've got something that no one else has in data, and then you might append that with your choice of publicly available data sets to make it a wonderful mix of things that does answer the questions or provides the insights or whatever it is you're trying to do.

SPEAKER_01 31:19

I think that's I think it's really important that that last point. Um that uh augmenting data, your data sources with external data sources. So just come back to your uh education analogy about you know information about that whole stuff about yeah, that and and smart targets and you know augmenting those that that data to to give a much more holistic view of the student and what they are capable of, yeah, rather than just passing an exam is.

SPEAKER_00 31:48

Well that's the thing, and you could in smart targets at the moment, smart targets are a big college thing, um, you know, seems to be making a difference in in achievement if you give clarity, who knew, to each the students and tell them this is what we really need to do now, focus on this. But if you could bring to that more contextual on the student data, you know, we know they like knitting and rugby, a bit odd combination. There you go. Well, let's bring that in. Do a bit less knitting and a bit more attending your class or whatever, you know. But you could speak now to them in a much more personal way, and indeed in their own language if you wanted to, if that helped. So I think there's but the parent report side of it, really interesting because again, in in terms of upending and you know, really transforming, the thing with parent reports is they happen normally once a term, if you're lucky, they are normally fairly generic because the teacher has to write 30 of them in or more than that in one go. I mean, imagine doing that, that's really hard. The job, therefore, is at least half the job is reading the data about that pupil, student, learner, working out what they've done and not done, writing that in words, that stuff AI can smash. Then you add hopefully some personal kind of and you know, I love his contribution in class and he's really funny, or whatever it is you want to add. I think the idea that you could turn that on its head and say, we'll have parent reports not once a term, do them every week, because we can run them so fast from the data. And if you start to collect, and this is the part that you'd need to append things about he was funny in class today, and you know, didn't like his haircut or whatever the things that you add into these, um, then you just literally press a button and the parent can get a report, and then you go one step further where you could give parents a chat with their students' data, so at any time they fancy, they could say, How's little Johnny got on today? How did he do in his spelling test yesterday? And it was long as the data is available and secure, then that is all doable right now.

SPEAKER_01 33:51

How do you think your Louis would like you being able to access information about his college career?

SPEAKER_00 34:00

Well, it's a big question, isn't it? Is how much are people going to want to share? Likewise, with that, if you did assess the whole person, that conversation we were just having, and have this data file that knew everything about you in uh at least your school and education career. What goes through my head is it would be kind of unfair because you don't, as an employee, you'd only pick the creme de la creme. And you think, well, hang on a minute, what's that about? That's a weird thing to think about. To think, well, what's wrong with that? Of saying you're only gonna pick the best person for the job. Well, surely that's a brilliant thing. That's what they're trying to do anyway. Well, exactly. That's what we're trying to do, clumsily, through the through the lens of uh where you've got a seven in English. Okay, good. Does that mean it's gonna be good in in the job? Who knows? Interesting. I um I uh as we draw, I guess, towards a close, I was um at the uh ed tech conference I was at, which was the um FETEC invited me, so I should give them a mention. Um they was there was the CEO from Cognosist, so um she was along the way with slowly to me. Um he was on stage doing his piece. Brilliant, so you did a brilliant job of talking about send white paper, send revelation coming that's going to make it much tougher for educators uh in what they have to be doing, just matter of fact, day to day with a far larger group of people who do not anymore come with EHCPs. The plan is to limit the numbers that get the kind of specialist support and then make educators do uh you know, uh as a more inclusive environment look after everybody. Anyway, that's what his kind of talk and Cognosist, which is his his company, have a platform for measuring neurodiversity and then helping teachers understand how to cope with that, which is brilliant, a really, really good tool. Anyway, he was saying how um his their test for this uh of whether or not it's good or not, he said, My mum was a teacher, he's from he's from Newcastle, and he said, My mum's was a teacher, and she was a great teacher, uh, but not particularly good with technology. So our test is, and he moved the slide on a big quote and said, Can me mum use this? And I loved it. I think it's I think it's tremendous. I think that kind of making it as simple as you can, as intuitive as you can, is important for us all to remember. But so thank you, Chris, from Cosmos.

SPEAKER_01 36:21

Excellent. Well, the fact that he's a Geordie, I appreciate. And that can me mum use this brilliant. I love it.

SPEAKER_00 36:27

That was how he said it.

SPEAKER_01 36:30

I'll be good trying it. Well, that's nice of you not to not to do your uh Birmingham Geordie accent. Thanks, thank God for that. And on that bombshell, you have a fabulous sunny weekend off to your boat. The Birch Boat of Joy.

SPEAKER_00 36:47

Yes, I'll be poodling uh on the River Thames, so give me a sh wave if you're around.

SPEAKER_01 36:53

Excellent, excellent. All right, follow, you enjoy and I'll catch up with you.

SPEAKER_00 36:56

You too. Have a great weekend and uh yeah, see you next week.