WEBVTT
00:00:00.320 --> 00:00:10.880
Welcome to PMI's GEMBAfication, the podcast where we explore the real stories behind performance improvement and bringing improvements to where the work is really done.
00:00:10.880 --> 00:00:16.480
And of course, therefore, we talk about the highs and the lows that come with making improvements.
00:00:16.480 --> 00:00:28.320
I'm your host, Susannah Clarke, Managing Partner at PMI, and today I'm joined by Fraser Dear, Head of AI and Data Innovation at BCN.
00:00:28.320 --> 00:00:36.880
And we're going to talk about some of the problems that he's experienced, how he mobilised, what worked, what didn't, and most importantly, what Fraser learnt.
00:00:36.880 --> 00:00:39.200
So great to have you with me today, Fraser.
00:00:39.200 --> 00:00:40.560
Thank you very much for joining me.
00:00:40.560 --> 00:00:41.600
How are you?
00:00:42.000 --> 00:00:42.960
Very good, thank you.
00:00:42.960 --> 00:00:49.520
Very excited to have a wee chat through uh maybe some career highlights and perhaps some career lowlights.
00:00:49.520 --> 00:00:51.840
Uh so yeah, it should be a good, good fun session.
00:00:52.159 --> 00:00:53.759
I'm looking forward to it.
00:00:53.759 --> 00:01:03.759
Before we get started, I wonder, you know, before you were at BCN, when you and I first met, so you were very much an improvement role at that time.
00:01:03.759 --> 00:01:14.959
And I just wondered if you could just share with us a bit about how you found the transition away from a core improvement role into the kind of roles you're now in.
00:01:15.359 --> 00:01:16.799
Yeah, it's a great question.
00:01:16.799 --> 00:01:22.480
Um, I think when I was in an improvement role, it was in a fast-moving consumer goods business.
00:01:22.480 --> 00:01:27.519
So um high volume, low margin, um, every penny mattered.
00:01:27.519 --> 00:01:37.920
And the the psychology around change was maybe not understood fully, but the requirement for change was understood.
00:01:37.920 --> 00:01:57.359
And I think moving into what I do now, one of the catches is we're very data rich, but in psych potentially poor, because some organizations don't necessarily understand the value of process and optimization of process.
00:01:57.359 --> 00:02:03.120
And so when you start to talk about an improvement cycle, they're like, Well, what do you mean?
00:02:03.120 --> 00:02:04.159
We just do it, right?
00:02:04.159 --> 00:02:19.759
And so I think sometimes that transition between having a real well-structured, organized framework of how to describe the change and how to describe deviations from change is really lost if you're not in that world of improvement.
00:02:20.080 --> 00:02:22.319
Yes, yes, exactly.
00:02:22.319 --> 00:02:23.840
Interesting.
00:02:23.840 --> 00:02:31.919
Well, so on that note, we should start with tell me uh tell us a bit about the problem that you've been trying to solve.
00:02:32.400 --> 00:02:43.759
Well, um, so this is this is a bit of an interesting one because in my role, uh head of AI, uh, and as you can imagine, head of AI at this time of the time of the world is a very busy place.
00:02:43.759 --> 00:02:55.520
And data innovation, it might sound like a bit of a weird combination, but one of the one of the things that we do here at BCN is around products, and one of the products that we have are is something called Easy SPC.
00:02:55.520 --> 00:03:06.960
Now I'm not interested in the Easy SPC product part, but I'm here to talk a little bit about the challenge of bringing together AI with a framework, a tool set that people trust.
00:03:06.960 --> 00:03:20.479
And and the the thing that kind of really interests me here is about you know, organizations don't necessarily have a data problem today, a little bit what I like I said just a moment ago.
00:03:20.479 --> 00:03:29.759
It's having the trust to be able to use AI to be able to interpret the data they've got, and then how to embed that into an organization.
00:03:29.759 --> 00:03:37.039
Now, we see two sides to this, because we've got our business, so our people, and then we've got our clients.
00:03:37.039 --> 00:03:44.879
And so we see the same challenges mirrored, and it's really interesting to see when when we've had the challenge, we can explain the challenge to clients.
00:03:44.879 --> 00:03:49.599
But if we've not had the challenge and the clients getting a challenge, that's where things start to become really interesting.
00:03:49.599 --> 00:03:50.960
Because what did we do differently?
00:03:50.960 --> 00:03:52.000
How did we tackle that?
00:03:52.000 --> 00:03:55.360
And and also when we got it wrong, what did we do about it?
00:03:55.680 --> 00:03:56.159
Yeah.
00:03:56.159 --> 00:03:56.719
Okay.
00:03:56.719 --> 00:04:01.599
Okay, so talk us through a bit how you how do you approach it all in the first place?
00:04:02.159 --> 00:04:02.479
Yeah.
00:04:02.479 --> 00:04:07.199
So if we think about um SPC as a general thing, um, it's well understood.
00:04:09.360 --> 00:04:10.000
Just in case.
00:04:10.000 --> 00:04:10.879
You never know.
00:04:10.879 --> 00:04:15.039
For anyone who doesn't know what SPC is, if you could just explain that, that would be great.
00:04:15.360 --> 00:04:15.680
Perfect.
00:04:15.680 --> 00:04:17.519
So statistical process control.
00:04:17.519 --> 00:04:39.759
Um, a way to look at your system and understand whether or not it is a well-controlled system, whether there are changes which have been introduced, um, whether things have gone outside the control mechanisms, and really understanding how how well controlled your your operations are, your activities, whatever it is that you're using the SPC chart for.
00:04:39.759 --> 00:04:44.800
There's loads of different types of SPC charts, and they're all for different purposes and different reasons.
00:04:44.800 --> 00:04:46.560
That's not the challenge today.
00:04:46.560 --> 00:04:59.439
The challenge today is around saying, well, look, these systems exist, and the the statistical engine inside all of these um calculations is trusted, is well established.
00:04:59.439 --> 00:05:03.279
There may be some individuals who may wish to challenge some of those calculations out again.
00:05:03.279 --> 00:05:18.079
But effectively, if you've got good data, quality data coming into your system, and you have the charts to show what those systems are telling you, i.e., uh, I don't know, let's talk about um let's do a let's do a pretend example.
00:05:18.079 --> 00:05:27.120
We're making, I have a pen in my hand, uh, and the component parts are the pen lid, the the ink, the nib, and the body of the pen.
00:05:27.120 --> 00:05:33.920
If one of those things goes out of control, as in maybe the nib is slightly too large, then you're gonna have a product failure.
00:05:33.920 --> 00:05:42.000
So you would see that in the SPC chart because all of a sudden failures would rise, and all of a sudden you'd be asking questions, say, well, why are we all of a sudden seeing the spike?
00:05:42.000 --> 00:05:47.120
And it's usually underpinning data, which is where you start to find that information.
00:05:47.120 --> 00:06:08.160
And so the question we posed internally was to say, well, could we use AI to help us establish what may be causing the things, the the I'm saying things because there may be multiple different causes, it may be special cause variation or all sorts of different reasons that these SPC charts are showing differences.
00:06:08.160 --> 00:06:19.680
And there's two or three big schools of thought, because one group of people were saying, Oh my god, this is amazing, AI is awesome, and AI can save the world.
00:06:19.680 --> 00:06:20.560
Brilliant.
00:06:20.560 --> 00:06:23.360
So we've got the people at the front that want to do AI for everything.
00:06:23.360 --> 00:06:27.519
Then we've got the people at the other end of the spectrum going, I don't trust AI.
00:06:27.519 --> 00:06:29.120
How can AI help me here?
00:06:29.120 --> 00:06:34.879
Because I've got a real strong framework, I've got a controlled understanding framework, so I don't need anything else to help me.
00:06:34.879 --> 00:06:39.839
And then we've got the kind of people in the middle that are going, well, let's wait and see.
00:06:39.839 --> 00:06:40.720
Let's see what you come up with.
00:06:40.720 --> 00:06:42.399
I'm not going to sit on the fence.
00:06:42.399 --> 00:06:53.360
And I think what was interesting was when we talked about using AI in the control tool, it was very much we don't need it.
00:06:53.360 --> 00:06:56.959
But when we got into the kind of well, what about the understanding piece?
00:06:56.959 --> 00:07:01.279
Because it wasn't about the decision making, that's when it became really interesting.
00:07:01.279 --> 00:07:12.720
Because, you know, improvement teams here, but also with our clients, are kind of under a lot of pressure, more data, more expectations, more value, sometimes less clarity.
00:07:12.720 --> 00:07:28.000
And and it's the getting speed to outcome with perhaps not full data sets where the grey area creeps in with the improvement teams, or certainly in my experience anyway.
00:07:28.000 --> 00:07:30.959
And that's what we saw with our with our internal team too.
00:07:31.839 --> 00:07:32.160
Right.
00:07:32.160 --> 00:07:43.439
Okay, so you've got a theory that you can use AI to help understand what might be causing signals variation on your SPC charts.
00:07:43.439 --> 00:07:50.319
You've got people in three places I don't like it, and I don't want to touch it.
00:07:50.319 --> 00:07:52.399
I love it, I can't wait to get started.
00:07:52.399 --> 00:07:57.120
And those, I'll wait and see, I'm not against it, but I need it to be proven to me.
00:07:57.120 --> 00:08:11.199
You found that the the the trust question has shifted because you're saying, What if we use it to to help understand what where the cause is?
00:08:11.199 --> 00:08:14.160
So so talk a bit more about then.
00:08:14.160 --> 00:08:20.800
So having gone down that path, then you're trying to do two things here, I feel.
00:08:20.800 --> 00:08:22.000
Correct me if I'm wrong.
00:08:22.000 --> 00:08:27.439
One is bring the people with you, and two is confirm whether AI is gonna be useful or not.
00:08:27.439 --> 00:08:30.079
So talk can you talk to me a bit about those?
00:08:30.480 --> 00:08:31.439
Yeah, sure.
00:08:31.439 --> 00:08:42.480
So I think in uh I should also say as a Microsoft partner, so I can only talk about Microsoft AI.
00:08:42.480 --> 00:08:49.679
Um it's we we're in a privileged position because we I have access to all of the cutting edge tools.
00:08:49.679 --> 00:08:55.840
So I can play in my little Fraser playground of joy, um creating things and going, Oh, this is awesome.
00:08:55.840 --> 00:08:59.279
So that's that's one little element to be aware of.
00:08:59.279 --> 00:09:08.399
And and I think from the human aspect, there are people that will say, I don't trust this.
00:09:08.399 --> 00:09:11.279
Well, what grounds do you have to trust it?
00:09:11.279 --> 00:09:12.799
You don't know what it is.
00:09:12.799 --> 00:09:24.080
If if you were to tell someone that this new fangled software that was being introduced to a business, was going to save the world, of course, people would be a bit like, well, I don't believe that's true.
00:09:24.080 --> 00:09:29.120
So let's have maybe some onboarding, let's maybe do some training, let's maybe understand what the platform is.
00:09:29.120 --> 00:09:45.279
And that's one of the key pieces that we initially realized people didn't understand what an LLM was, what all this AI, particularly this new generative AI componentry, what it actually does behind behind the scenes, what happens under the bonnet.
00:09:45.279 --> 00:10:08.720
And there's there's other parts of the Microsoft ecosystem that we've been able to leverage here, things that are controlled and governed and and and secured in a way that when you start to unwrap the layers, you take away the shiny AI front, you know, little agent that's busy yapping away to you, and you start to explain it in a technical perspective to technical people so they can understand it, the trust level starts to come up.
00:10:08.720 --> 00:10:18.399
But it's I mean, I mean I'm saying it like a you know, I feel a little bit like you know, the Emperor's New Clothes, or you know, this is no different to any other chain cycle.
00:10:18.399 --> 00:10:25.440
It's just because this the pace at which AI is moving, that chain cycle needs to be quite rapid.
00:10:25.440 --> 00:10:30.159
Because before we roll through a single chain cycle, it's changed again.
00:10:30.159 --> 00:10:38.879
And so I think that was one of the key learnings for us when we did this project internally was to say, well, you're gonna have to trust this, and you're gonna have to trust it pretty quickly.
00:10:38.879 --> 00:10:41.840
So we're gonna need to do this change loop rapidly.
00:10:41.840 --> 00:10:46.399
So here's some foundational elements, and I'm gonna show you, and then we'll do some more foundation, and we're gonna show you.
00:10:46.399 --> 00:10:51.600
So it was still a change cycle, but it was much, much faster.
00:10:52.159 --> 00:10:56.799
And the change cycle included skills development in order, that's what you said.
00:10:56.799 --> 00:11:04.000
100% created the real change in trust because you were able to generate the understanding as well of what's going on.
00:11:04.639 --> 00:11:15.679
And it's the understanding of saying the answer might be wrong, but I understand how AI got the wrong answer because of the context I provided or because of the question set I asked.
00:11:15.679 --> 00:11:20.799
And I can only do that if I understand how the AI tool actually works.
00:11:20.799 --> 00:11:27.919
Because, you know, um many of our listeners today might be saying, Oh yeah, uh, pick AI tool of your choice.
00:11:27.919 --> 00:11:29.120
I'll say Copilot.
00:11:29.120 --> 00:11:30.240
Uh it's rubbish.
00:11:30.240 --> 00:11:31.759
Well, is it really?
00:11:31.759 --> 00:11:34.879
Or did we just not really understand what the output was?
00:11:34.879 --> 00:11:46.000
Because if you put I mean, I g I got I'm being cliched all day today, you know, rubbish in, rubbish out, and it's kind of the same thing, you know, like the foundational elements are really important to understand.
00:11:46.000 --> 00:11:59.120
I wouldn't expect somebody to go to a DIY store and buy a big fangled DIY tool and expect them to be able to build, you know, the leaning tower of Pisa without maybe some skills and training in the tools that they're using, and it's a kind of the same thing.
00:11:59.440 --> 00:12:01.039
Yes, yes, indeed.
00:12:01.039 --> 00:12:01.759
Indeed.
00:12:01.759 --> 00:12:37.360
I know we've got people within uh with within you know PMI who were very much in that middle group, several a lot of on the fence people, and it was through that iteration of demonstrating and learning from each other we found really beneficial somebody saying, I found a neat thing I can do, I didn't know I could do this, and then sharing that, and and and and by by the sharing, developing skills and will to want to have a go um definitely made a a big difference.
00:12:37.759 --> 00:12:49.200
And and that's it, you know, I can only think of the potential use cases with the tools that we build, but that's when when we take things to clients and they sort of go, Oh, could we do this?
00:12:49.200 --> 00:12:51.519
And it's like, well, give it a go.
00:12:51.519 --> 00:12:57.279
Because you've got the context of your particular business challenge and we've built a framework.
00:12:57.279 --> 00:13:00.159
Use it just like an SPC tool, use the framework.
00:13:00.159 --> 00:13:13.600
If you understand what it's doing and you understand how AI is coming up with these answers, in this particular instance, we're using a data agent inside Fabric, whether that's relevant or not, some of your listeners, but that's how the kind of brain works.
00:13:13.600 --> 00:13:20.559
Well, if you understand Fabric, you understand data, you understand the the logic behind it, which is all to do with querying with SQL.
00:13:20.559 --> 00:13:21.440
Awesome.
00:13:21.440 --> 00:13:29.519
Then the trust level changed dramatically because the hype around AI, oh, it's a agent, it's a this, it's a that.
00:13:29.519 --> 00:13:32.799
Okay, well, that's all lovely to say, but what is it under the bonnet?
00:13:32.799 --> 00:13:40.240
And once they understand that, that was a real shift for us because people were like, oh, this is not this is not Fraser Playtime making things up.
00:13:40.240 --> 00:13:43.200
This is actually built on really strong foundations.
00:13:43.200 --> 00:13:47.440
And from an from a practitioner perspective, it was built around the data.
00:13:47.440 --> 00:13:49.759
The data was the key thing, not the agent.
00:13:49.759 --> 00:13:53.759
The agent was the the surface, it was the database that was the critical thing.
00:13:54.080 --> 00:13:55.840
Okay, fascinating.
00:13:55.840 --> 00:14:04.879
And then through this cycle, um uh uh of change, because even though we're working quite quickly, there's uh been a lot going on, obviously.
00:14:04.879 --> 00:14:13.919
Anything that didn't go so well, anything you thought, oh I wish we hadn't started there, or if I had my time again, I'd do it so differently.
00:14:14.480 --> 00:14:15.519
Yes, 100%.
00:14:15.519 --> 00:14:21.679
Um, and and I was gonna talk through two um two pieces.
00:14:21.679 --> 00:14:26.159
One was the ability to create things.
00:14:26.159 --> 00:14:28.559
We're back in Fraser Sandbox here.
00:14:28.559 --> 00:14:30.799
You can do these things rapidly.
00:14:30.799 --> 00:14:34.399
But the question isn't could AI help us anymore?
00:14:34.399 --> 00:14:40.000
The answer to that question is well, in my experience, nine times out of ten, yes.
00:14:40.000 --> 00:14:44.320
But the the question needs to be, how can AI help?
00:14:44.320 --> 00:14:48.240
Not can AI help, and is it going to add any value?
00:14:48.240 --> 00:14:55.039
So when I started doing some of the work around this, I was like, oh, we could do this and we could do that, and we could do this, and we could do that.
00:14:55.039 --> 00:14:57.440
Hang on a minute, let's boil it right back down.
00:14:57.440 --> 00:15:01.200
What is the key value point that we were trying to solve?
00:15:01.200 --> 00:15:12.799
Because my first iteration of this you know, SPC agent, agentic kind of QA type tool, it did all sorts of things, none of which were useful to man or beast.
00:15:12.799 --> 00:15:25.759
They were helpful to me because I thought this was cool, but this is where this developer AI piece becomes a bit of a danger to organizations because the person that knows how to doesn't necessarily know the right questions.
00:15:25.759 --> 00:15:34.639
And only when you sit with a practitioner who can say, Well, actually, this is the kind of thing I really need, can you really bind the developer to the right outcome?
00:15:34.639 --> 00:15:47.120
Now, me being me, the developer, the mindset, having done some of it before, but it was a little while ago, and I thought I'd probably get maybe 50% of the weight first before I spoke to a practitioner again.
00:15:47.120 --> 00:15:49.440
Yeah, that probably wasn't such a good idea.
00:15:49.440 --> 00:16:00.399
And hey, once again, I'm you know, don't don't run off with your ideas, make sure they they add real value because otherwise, guess what?
00:16:00.399 --> 00:16:02.799
Your adoption rate is gonna be really, really low.
00:16:02.799 --> 00:16:08.399
Because it might be cool, but that's not gonna help your practitioners on the ground on the shop floor.
00:16:08.399 --> 00:16:12.879
Go down and speak to them, go to wherever oh hey, we're talking about gamification.
00:16:12.879 --> 00:16:19.360
Go do the gamba walk, find out what your practitioners need on the shop floor before anybody touches a tool.
00:16:19.360 --> 00:16:22.080
Because if you're talking about tools, it's not gonna work.
00:16:22.080 --> 00:16:28.799
It's the it's not gonna be the right thing to fix the problem that your practitioners actually have at that moment.
00:16:29.039 --> 00:16:31.120
And is it that was the first mistake I made.
00:16:31.120 --> 00:16:38.240
Well, yes, and and I I love the fact that we ended with the it's not gonna fix the problem.
00:16:38.240 --> 00:16:41.360
Because actually that is the that that's the question, isn't it?
00:16:41.360 --> 00:16:48.240
And that's the point of, as you say, going to gemba, meeting the people, talking to them about the work they are doing and what are the problems they face.
00:16:48.240 --> 00:16:54.399
Um so that you're not sort of blanketing AI over them, you are still a problem solver.
00:16:54.799 --> 00:16:55.279
Yeah.
00:16:55.279 --> 00:17:18.559
And and you know, um, I suppose in my years now, I don't want to think about how many, too many years, um, I suppose I've kind of been in that strange position where I've been the I've been the doer, I've been the practitioner, and now I'm kind of I suppose a leader of people, and I love flipping the roles, so I don't go down and say, oh, let's do this, let's do that, let's try this.
00:17:18.559 --> 00:17:21.200
I'm like, what would you like to just to do?
00:17:21.200 --> 00:17:22.640
How can I help you?
00:17:22.640 --> 00:17:25.759
How can I create an environment for you to succeed?
00:17:25.759 --> 00:17:30.640
I might know some of the things, I might make some suggestions, but actually, what is it you need?
00:17:30.640 --> 00:17:33.759
Because then the ownership's with the practitioner, the ownership is with the team.
00:17:33.759 --> 00:17:35.839
Tell me what you need and I'll go rebuild it.
00:17:35.839 --> 00:17:39.359
I'll come back to you with a thing, but then it's your tool to then use.
00:17:39.680 --> 00:17:40.000
Yeah.
00:17:40.480 --> 00:17:46.160
And stop thinking about, well, I have this tool, how can I ram the tool into the problem?
00:17:46.160 --> 00:17:47.440
It's well, what's the problem?
00:17:47.440 --> 00:17:50.000
And then let's try and get the right tools to fix that problem.
00:17:50.000 --> 00:17:58.240
And again, that might sound really basic, but there's just so many it's so quick to spin up AI tools now that we're gonna get into the situation.
00:17:58.240 --> 00:17:58.960
There's so many tools.
00:17:58.960 --> 00:18:00.000
Well, which one am I supposed to use?
00:18:00.000 --> 00:18:00.640
No, no, no.
00:18:01.519 --> 00:18:04.559
I've got one of these and I'm gonna use it no matter what happens.
00:18:04.559 --> 00:18:05.200
Absolutely.
00:18:05.200 --> 00:18:07.039
No, no, I completely agree.
00:18:07.039 --> 00:18:14.720
Now you've you've you've you know you've you've wet my appetite here, because you said that was the first, so that feels like there's at least a second that you want to share.
00:18:15.440 --> 00:18:16.000
Yeah.
00:18:16.000 --> 00:18:24.880
I think I think the second problem we had, oh problem, uh, is um foun foundational knowledge.
00:18:24.880 --> 00:18:58.640
So um I talked a little bit about um developers understanding the practitioner's knowledge of data, but when you talk to non-technical people or non this particular example is about SPC, but it could be about anything, um as soon as we deviate away from a uh a change cycle, a change practitioner, and you start to take these kind of conversations to um non-change enabled board members or senior leaders in an organization, this may not be their you know, their their bag, this may not be their interest.
00:18:58.640 --> 00:19:04.079
Yes, of course, everybody wants to make more profit or make higher margins or whatever it may be.
00:19:04.079 --> 00:19:12.720
But when you start to say, Oh, um I want to have um half a sigma what it doesn't make any sense.
00:19:12.720 --> 00:19:29.680
And so when we started to talk about this particular project internally, again, my f my fault, I thought, oh, hang on a minute, everyone will understand all of this gobbledygook, all of this, you know, even just the word gemba, many people won't know what that means.
00:19:29.680 --> 00:19:42.480
And so trying to talk at the right level and talk with the right kind of language in a mixed audience on a something as specific as AI and SPC was definitely a challenge and not one to be underestimated.
00:19:42.480 --> 00:19:46.480
And so uh I'm sure my mum won't mind me saying this.
00:19:46.480 --> 00:19:50.559
Um I revert to uh does this make sense, mum?
00:19:50.559 --> 00:19:52.000
Could you listen to this?
00:19:52.000 --> 00:19:54.160
And can you understand what I'm trying to say?
00:19:54.160 --> 00:19:57.839
Because sometimes if it doesn't pass the mum test, it doesn't make sense.
00:19:57.839 --> 00:20:15.920
And if it doesn't pass the mum test, then how is somebody who's not um who doesn't have the understanding of of of the value of, in this case, SPC, and of the value of of proper structured, framework to change, um, they're not going to get it.
00:20:15.920 --> 00:20:17.680
And so that was the other mistake I made.
00:20:17.680 --> 00:20:19.200
I ran far too fast.
00:20:19.200 --> 00:20:25.680
Once we had established what was needed, I ran to solution, then came back, tried to explain it to the board who went, what are you doing?
00:20:25.680 --> 00:20:34.640
Um stakeholder management wasn't the best on this particular project, but hey, I'm glad we did it because we we've built a solution quickly.
00:20:34.640 --> 00:20:44.480
We didn't go through a huge um a huge cycle of let's get all the requirements down first, let's build everything out to perfect.
00:20:44.480 --> 00:20:47.200
It was much, much more agile in that respect.
00:20:47.200 --> 00:20:53.599
Um, but the catch was, yeah, I probably should have brought my stakeholders on board a little bit sooner.
00:20:53.599 --> 00:20:55.200
But um, yeah, that was on me.
00:20:56.079 --> 00:21:05.920
I think yes, but I also think that it that the again what you described is not an uncommon challenge, in as much as it's it's hard, isn't it?
00:21:05.920 --> 00:21:10.319
Here we are, we're let's say we're you know we're driving change, we're driving improvements.
00:21:10.319 --> 00:21:13.279
I might have eight stakeholders to work with.
00:21:13.279 --> 00:21:32.799
Um my best chances of success are to step back, look at each of them, work out what's going to drive them, work out where they are, work out what language I need to be using in order to talk to them, and to probably tackle them individually before trying to bring them all together at once.
00:21:32.799 --> 00:21:49.759
But actually, I'm really busy, you know, and and and therefore sometimes taking that step, slowing down slightly in order to speed up, but understanding that actually it's our job to adapt to them and their needs.
00:21:49.759 --> 00:22:05.039
Yes, it's not it, you know, just because we think that um and we might be right that we've got the best solution, um, it doesn't mean that they will even begin to uh understand it, and not because they're stupid, just because we're not talking the same language.
00:22:05.279 --> 00:22:06.160
Yeah, exactly.
00:22:06.160 --> 00:22:18.880
And and you know, we can we can have as much data as we like to prove something or disprove something, but if we can't communicate it in a way that's meaningful for the recipient, then don't communicate it, because it won't land.
00:22:18.880 --> 00:22:22.000
And they may end up with the wrong end of the stick altogether.
00:22:22.000 --> 00:22:24.000
And that happened in this case.
00:22:24.000 --> 00:22:28.240
What happened in this case is someone said, Oh, we don't need easy SPC then.
00:22:28.240 --> 00:22:30.319
And I was like, no, no, no, no.
00:22:30.319 --> 00:22:31.519
Hang on.
00:22:31.519 --> 00:22:38.960
This isn't about replacing what we've already got, it's about augmenting, about enabling more, enabling richer.
00:22:38.960 --> 00:22:40.880
Ah, right, okay, cool.
00:22:40.880 --> 00:22:49.279
So it was using the right language for the recipient, which is difficult, especially if you've got a board or you've got a group of mixed knowledge or mixed understanding.
00:22:49.279 --> 00:22:51.359
That can be a real that can be a real challenge.
00:22:51.680 --> 00:22:52.319
Yes.
00:22:52.319 --> 00:22:53.759
Very interesting.
00:22:53.759 --> 00:22:56.000
Fraser, thank you so much.
00:22:56.000 --> 00:22:58.960
I really appreciate your honesty as well.
00:22:58.960 --> 00:23:01.839
Um, and for sharing your stories with us.
00:23:01.839 --> 00:23:03.279
It's really it's been great.
00:23:03.279 --> 00:23:04.559
Thank you so much.
00:23:04.559 --> 00:23:08.079
Um, to our listeners, I do hope you enjoyed this episode.
00:23:08.079 --> 00:23:15.759
Um, and uh you can find Fraser on LinkedIn if you want to connect with him and talk about AI some more or SPC.
00:23:15.759 --> 00:23:25.680
Um and please don't forget to subscribe, share, leave us a review, because the more this podcast grows, the better everyone's work becomes.
00:23:25.680 --> 00:23:27.680
Look forward to seeing you next time.
00:23:27.680 --> 00:23:28.240
Thank you.
00:00:00.320 --> 00:00:10.880
Welcome to PMI's GEMBAfication, the podcast where we explore the real stories behind performance improvement and bringing improvements to where the work is really done.
00:00:10.880 --> 00:00:16.480
And of course, therefore, we talk about the highs and the lows that come with making improvements.
00:00:16.480 --> 00:00:28.320
I'm your host, Susannah Clarke, Managing Partner at PMI, and today I'm joined by Fraser Dear, Head of AI and Data Innovation at BCN.
00:00:28.320 --> 00:00:36.880
And we're going to talk about some of the problems that he's experienced, how he mobilised, what worked, what didn't, and most importantly, what Fraser learnt.
00:00:36.880 --> 00:00:39.200
So great to have you with me today, Fraser.
00:00:39.200 --> 00:00:40.560
Thank you very much for joining me.
00:00:40.560 --> 00:00:41.600
How are you?
00:00:42.000 --> 00:00:42.960
Very good, thank you.
00:00:42.960 --> 00:00:49.520
Very excited to have a wee chat through uh maybe some career highlights and perhaps some career lowlights.
00:00:49.520 --> 00:00:51.840
Uh so yeah, it should be a good, good fun session.
00:00:52.159 --> 00:00:53.759
I'm looking forward to it.
00:00:53.759 --> 00:01:03.759
Before we get started, I wonder, you know, before you were at BCN, when you and I first met, so you were very much an improvement role at that time.
00:01:03.759 --> 00:01:14.959
And I just wondered if you could just share with us a bit about how you found the transition away from a core improvement role into the kind of roles you're now in.
00:01:15.359 --> 00:01:16.799
Yeah, it's a great question.
00:01:16.799 --> 00:01:22.480
Um, I think when I was in an improvement role, it was in a fast-moving consumer goods business.
00:01:22.480 --> 00:01:27.519
So um high volume, low margin, um, every penny mattered.
00:01:27.519 --> 00:01:37.920
And the the psychology around change was maybe not understood fully, but the requirement for change was understood.
00:01:37.920 --> 00:01:57.359
And I think moving into what I do now, one of the catches is we're very data rich, but in psych potentially poor, because some organizations don't necessarily understand the value of process and optimization of process.
00:01:57.359 --> 00:02:03.120
And so when you start to talk about an improvement cycle, they're like, Well, what do you mean?
00:02:03.120 --> 00:02:04.159
We just do it, right?
00:02:04.159 --> 00:02:19.759
And so I think sometimes that transition between having a real well-structured, organized framework of how to describe the change and how to describe deviations from change is really lost if you're not in that world of improvement.
00:02:20.080 --> 00:02:22.319
Yes, yes, exactly.
00:02:22.319 --> 00:02:23.840
Interesting.
00:02:23.840 --> 00:02:31.919
Well, so on that note, we should start with tell me uh tell us a bit about the problem that you've been trying to solve.
00:02:32.400 --> 00:02:43.759
Well, um, so this is this is a bit of an interesting one because in my role, uh head of AI, uh, and as you can imagine, head of AI at this time of the time of the world is a very busy place.
00:02:43.759 --> 00:02:55.520
And data innovation, it might sound like a bit of a weird combination, but one of the one of the things that we do here at BCN is around products, and one of the products that we have are is something called Easy SPC.
00:02:55.520 --> 00:03:06.960
Now I'm not interested in the Easy SPC product part, but I'm here to talk a little bit about the challenge of bringing together AI with a framework, a tool set that people trust.
00:03:06.960 --> 00:03:20.479
And and the the thing that kind of really interests me here is about you know, organizations don't necessarily have a data problem today, a little bit what I like I said just a moment ago.
00:03:20.479 --> 00:03:29.759
It's having the trust to be able to use AI to be able to interpret the data they've got, and then how to embed that into an organization.
00:03:29.759 --> 00:03:37.039
Now, we see two sides to this, because we've got our business, so our people, and then we've got our clients.
00:03:37.039 --> 00:03:44.879
And so we see the same challenges mirrored, and it's really interesting to see when when we've had the challenge, we can explain the challenge to clients.
00:03:44.879 --> 00:03:49.599
But if we've not had the challenge and the clients getting a challenge, that's where things start to become really interesting.
00:03:49.599 --> 00:03:50.960
Because what did we do differently?
00:03:50.960 --> 00:03:52.000
How did we tackle that?
00:03:52.000 --> 00:03:55.360
And and also when we got it wrong, what did we do about it?
00:03:55.680 --> 00:03:56.159
Yeah.
00:03:56.159 --> 00:03:56.719
Okay.
00:03:56.719 --> 00:04:01.599
Okay, so talk us through a bit how you how do you approach it all in the first place?
00:04:02.159 --> 00:04:02.479
Yeah.
00:04:02.479 --> 00:04:07.199
So if we think about um SPC as a general thing, um, it's well understood.
00:04:09.360 --> 00:04:10.000
Just in case.
00:04:10.000 --> 00:04:10.879
You never know.
00:04:10.879 --> 00:04:15.039
For anyone who doesn't know what SPC is, if you could just explain that, that would be great.
00:04:15.360 --> 00:04:15.680
Perfect.
00:04:15.680 --> 00:04:17.519
So statistical process control.
00:04:17.519 --> 00:04:39.759
Um, a way to look at your system and understand whether or not it is a well-controlled system, whether there are changes which have been introduced, um, whether things have gone outside the control mechanisms, and really understanding how how well controlled your your operations are, your activities, whatever it is that you're using the SPC chart for.
00:04:39.759 --> 00:04:44.800
There's loads of different types of SPC charts, and they're all for different purposes and different reasons.
00:04:44.800 --> 00:04:46.560
That's not the challenge today.
00:04:46.560 --> 00:04:59.439
The challenge today is around saying, well, look, these systems exist, and the the statistical engine inside all of these um calculations is trusted, is well established.
00:04:59.439 --> 00:05:03.279
There may be some individuals who may wish to challenge some of those calculations out again.
00:05:03.279 --> 00:05:18.079
But effectively, if you've got good data, quality data coming into your system, and you have the charts to show what those systems are telling you, i.e., uh, I don't know, let's talk about um let's do a let's do a pretend example.
00:05:18.079 --> 00:05:27.120
We're making, I have a pen in my hand, uh, and the component parts are the pen lid, the the ink, the nib, and the body of the pen.
00:05:27.120 --> 00:05:33.920
If one of those things goes out of control, as in maybe the nib is slightly too large, then you're gonna have a product failure.
00:05:33.920 --> 00:05:42.000
So you would see that in the SPC chart because all of a sudden failures would rise, and all of a sudden you'd be asking questions, say, well, why are we all of a sudden seeing the spike?
00:05:42.000 --> 00:05:47.120
And it's usually underpinning data, which is where you start to find that information.
00:05:47.120 --> 00:06:08.160
And so the question we posed internally was to say, well, could we use AI to help us establish what may be causing the things, the the I'm saying things because there may be multiple different causes, it may be special cause variation or all sorts of different reasons that these SPC charts are showing differences.
00:06:08.160 --> 00:06:19.680
And there's two or three big schools of thought, because one group of people were saying, Oh my god, this is amazing, AI is awesome, and AI can save the world.
00:06:19.680 --> 00:06:20.560
Brilliant.
00:06:20.560 --> 00:06:23.360
So we've got the people at the front that want to do AI for everything.
00:06:23.360 --> 00:06:27.519
Then we've got the people at the other end of the spectrum going, I don't trust AI.
00:06:27.519 --> 00:06:29.120
How can AI help me here?
00:06:29.120 --> 00:06:34.879
Because I've got a real strong framework, I've got a controlled understanding framework, so I don't need anything else to help me.
00:06:34.879 --> 00:06:39.839
And then we've got the kind of people in the middle that are going, well, let's wait and see.
00:06:39.839 --> 00:06:40.720
Let's see what you come up with.
00:06:40.720 --> 00:06:42.399
I'm not going to sit on the fence.
00:06:42.399 --> 00:06:53.360
And I think what was interesting was when we talked about using AI in the control tool, it was very much we don't need it.
00:06:53.360 --> 00:06:56.959
But when we got into the kind of well, what about the understanding piece?
00:06:56.959 --> 00:07:01.279
Because it wasn't about the decision making, that's when it became really interesting.
00:07:01.279 --> 00:07:12.720
Because, you know, improvement teams here, but also with our clients, are kind of under a lot of pressure, more data, more expectations, more value, sometimes less clarity.
00:07:12.720 --> 00:07:28.000
And and it's the getting speed to outcome with perhaps not full data sets where the grey area creeps in with the improvement teams, or certainly in my experience anyway.
00:07:28.000 --> 00:07:30.959
And that's what we saw with our with our internal team too.
00:07:31.839 --> 00:07:32.160
Right.
00:07:32.160 --> 00:07:43.439
Okay, so you've got a theory that you can use AI to help understand what might be causing signals variation on your SPC charts.
00:07:43.439 --> 00:07:50.319
You've got people in three places I don't like it, and I don't want to touch it.
00:07:50.319 --> 00:07:52.399
I love it, I can't wait to get started.
00:07:52.399 --> 00:07:57.120
And those, I'll wait and see, I'm not against it, but I need it to be proven to me.
00:07:57.120 --> 00:08:11.199
You found that the the the trust question has shifted because you're saying, What if we use it to to help understand what where the cause is?
00:08:11.199 --> 00:08:14.160
So so talk a bit more about then.
00:08:14.160 --> 00:08:20.800
So having gone down that path, then you're trying to do two things here, I feel.
00:08:20.800 --> 00:08:22.000
Correct me if I'm wrong.
00:08:22.000 --> 00:08:27.439
One is bring the people with you, and two is confirm whether AI is gonna be useful or not.
00:08:27.439 --> 00:08:30.079
So talk can you talk to me a bit about those?
00:08:30.480 --> 00:08:31.439
Yeah, sure.
00:08:31.439 --> 00:08:42.480
So I think in uh I should also say as a Microsoft partner, so I can only talk about Microsoft AI.
00:08:42.480 --> 00:08:49.679
Um it's we we're in a privileged position because we I have access to all of the cutting edge tools.
00:08:49.679 --> 00:08:55.840
So I can play in my little Fraser playground of joy, um creating things and going, Oh, this is awesome.
00:08:55.840 --> 00:08:59.279
So that's that's one little element to be aware of.
00:08:59.279 --> 00:09:08.399
And and I think from the human aspect, there are people that will say, I don't trust this.
00:09:08.399 --> 00:09:11.279
Well, what grounds do you have to trust it?
00:09:11.279 --> 00:09:12.799
You don't know what it is.
00:09:12.799 --> 00:09:24.080
If if you were to tell someone that this new fangled software that was being introduced to a business, was going to save the world, of course, people would be a bit like, well, I don't believe that's true.
00:09:24.080 --> 00:09:29.120
So let's have maybe some onboarding, let's maybe do some training, let's maybe understand what the platform is.
00:09:29.120 --> 00:09:45.279
And that's one of the key pieces that we initially realized people didn't understand what an LLM was, what all this AI, particularly this new generative AI componentry, what it actually does behind behind the scenes, what happens under the bonnet.
00:09:45.279 --> 00:10:08.720
And there's there's other parts of the Microsoft ecosystem that we've been able to leverage here, things that are controlled and governed and and and secured in a way that when you start to unwrap the layers, you take away the shiny AI front, you know, little agent that's busy yapping away to you, and you start to explain it in a technical perspective to technical people so they can understand it, the trust level starts to come up.
00:10:08.720 --> 00:10:18.399
But it's I mean, I mean I'm saying it like a you know, I feel a little bit like you know, the Emperor's New Clothes, or you know, this is no different to any other chain cycle.
00:10:18.399 --> 00:10:25.440
It's just because this the pace at which AI is moving, that chain cycle needs to be quite rapid.
00:10:25.440 --> 00:10:30.159
Because before we roll through a single chain cycle, it's changed again.
00:10:30.159 --> 00:10:38.879
And so I think that was one of the key learnings for us when we did this project internally was to say, well, you're gonna have to trust this, and you're gonna have to trust it pretty quickly.
00:10:38.879 --> 00:10:41.840
So we're gonna need to do this change loop rapidly.
00:10:41.840 --> 00:10:46.399
So here's some foundational elements, and I'm gonna show you, and then we'll do some more foundation, and we're gonna show you.
00:10:46.399 --> 00:10:51.600
So it was still a change cycle, but it was much, much faster.
00:10:52.159 --> 00:10:56.799
And the change cycle included skills development in order, that's what you said.
00:10:56.799 --> 00:11:04.000
100% created the real change in trust because you were able to generate the understanding as well of what's going on.
00:11:04.639 --> 00:11:15.679
And it's the understanding of saying the answer might be wrong, but I understand how AI got the wrong answer because of the context I provided or because of the question set I asked.
00:11:15.679 --> 00:11:20.799
And I can only do that if I understand how the AI tool actually works.
00:11:20.799 --> 00:11:27.919
Because, you know, um many of our listeners today might be saying, Oh yeah, uh, pick AI tool of your choice.
00:11:27.919 --> 00:11:29.120
I'll say Copilot.
00:11:29.120 --> 00:11:30.240
Uh it's rubbish.
00:11:30.240 --> 00:11:31.759
Well, is it really?
00:11:31.759 --> 00:11:34.879
Or did we just not really understand what the output was?
00:11:34.879 --> 00:11:46.000
Because if you put I mean, I g I got I'm being cliched all day today, you know, rubbish in, rubbish out, and it's kind of the same thing, you know, like the foundational elements are really important to understand.
00:11:46.000 --> 00:11:59.120
I wouldn't expect somebody to go to a DIY store and buy a big fangled DIY tool and expect them to be able to build, you know, the leaning tower of Pisa without maybe some skills and training in the tools that they're using, and it's a kind of the same thing.
00:11:59.440 --> 00:12:01.039
Yes, yes, indeed.
00:12:01.039 --> 00:12:01.759
Indeed.
00:12:01.759 --> 00:12:37.360
I know we've got people within uh with within you know PMI who were very much in that middle group, several a lot of on the fence people, and it was through that iteration of demonstrating and learning from each other we found really beneficial somebody saying, I found a neat thing I can do, I didn't know I could do this, and then sharing that, and and and and by by the sharing, developing skills and will to want to have a go um definitely made a a big difference.
00:12:37.759 --> 00:12:49.200
And and that's it, you know, I can only think of the potential use cases with the tools that we build, but that's when when we take things to clients and they sort of go, Oh, could we do this?
00:12:49.200 --> 00:12:51.519
And it's like, well, give it a go.
00:12:51.519 --> 00:12:57.279
Because you've got the context of your particular business challenge and we've built a framework.
00:12:57.279 --> 00:13:00.159
Use it just like an SPC tool, use the framework.
00:13:00.159 --> 00:13:13.600
If you understand what it's doing and you understand how AI is coming up with these answers, in this particular instance, we're using a data agent inside Fabric, whether that's relevant or not, some of your listeners, but that's how the kind of brain works.
00:13:13.600 --> 00:13:20.559
Well, if you understand Fabric, you understand data, you understand the the logic behind it, which is all to do with querying with SQL.
00:13:20.559 --> 00:13:21.440
Awesome.
00:13:21.440 --> 00:13:29.519
Then the trust level changed dramatically because the hype around AI, oh, it's a agent, it's a this, it's a that.
00:13:29.519 --> 00:13:32.799
Okay, well, that's all lovely to say, but what is it under the bonnet?
00:13:32.799 --> 00:13:40.240
And once they understand that, that was a real shift for us because people were like, oh, this is not this is not Fraser Playtime making things up.
00:13:40.240 --> 00:13:43.200
This is actually built on really strong foundations.
00:13:43.200 --> 00:13:47.440
And from an from a practitioner perspective, it was built around the data.
00:13:47.440 --> 00:13:49.759
The data was the key thing, not the agent.
00:13:49.759 --> 00:13:53.759
The agent was the the surface, it was the database that was the critical thing.
00:13:54.080 --> 00:13:55.840
Okay, fascinating.
00:13:55.840 --> 00:14:04.879
And then through this cycle, um uh uh of change, because even though we're working quite quickly, there's uh been a lot going on, obviously.
00:14:04.879 --> 00:14:13.919
Anything that didn't go so well, anything you thought, oh I wish we hadn't started there, or if I had my time again, I'd do it so differently.
00:14:14.480 --> 00:14:15.519
Yes, 100%.
00:14:15.519 --> 00:14:21.679
Um, and and I was gonna talk through two um two pieces.
00:14:21.679 --> 00:14:26.159
One was the ability to create things.
00:14:26.159 --> 00:14:28.559
We're back in Fraser Sandbox here.
00:14:28.559 --> 00:14:30.799
You can do these things rapidly.
00:14:30.799 --> 00:14:34.399
But the question isn't could AI help us anymore?
00:14:34.399 --> 00:14:40.000
The answer to that question is well, in my experience, nine times out of ten, yes.
00:14:40.000 --> 00:14:44.320
But the the question needs to be, how can AI help?
00:14:44.320 --> 00:14:48.240
Not can AI help, and is it going to add any value?
00:14:48.240 --> 00:14:55.039
So when I started doing some of the work around this, I was like, oh, we could do this and we could do that, and we could do this, and we could do that.
00:14:55.039 --> 00:14:57.440
Hang on a minute, let's boil it right back down.
00:14:57.440 --> 00:15:01.200
What is the key value point that we were trying to solve?
00:15:01.200 --> 00:15:12.799
Because my first iteration of this you know, SPC agent, agentic kind of QA type tool, it did all sorts of things, none of which were useful to man or beast.
00:15:12.799 --> 00:15:25.759
They were helpful to me because I thought this was cool, but this is where this developer AI piece becomes a bit of a danger to organizations because the person that knows how to doesn't necessarily know the right questions.
00:15:25.759 --> 00:15:34.639
And only when you sit with a practitioner who can say, Well, actually, this is the kind of thing I really need, can you really bind the developer to the right outcome?
00:15:34.639 --> 00:15:47.120
Now, me being me, the developer, the mindset, having done some of it before, but it was a little while ago, and I thought I'd probably get maybe 50% of the weight first before I spoke to a practitioner again.
00:15:47.120 --> 00:15:49.440
Yeah, that probably wasn't such a good idea.
00:15:49.440 --> 00:16:00.399
And hey, once again, I'm you know, don't don't run off with your ideas, make sure they they add real value because otherwise, guess what?
00:16:00.399 --> 00:16:02.799
Your adoption rate is gonna be really, really low.
00:16:02.799 --> 00:16:08.399
Because it might be cool, but that's not gonna help your practitioners on the ground on the shop floor.
00:16:08.399 --> 00:16:12.879
Go down and speak to them, go to wherever oh hey, we're talking about gamification.
00:16:12.879 --> 00:16:19.360
Go do the gamba walk, find out what your practitioners need on the shop floor before anybody touches a tool.
00:16:19.360 --> 00:16:22.080
Because if you're talking about tools, it's not gonna work.
00:16:22.080 --> 00:16:28.799
It's the it's not gonna be the right thing to fix the problem that your practitioners actually have at that moment.
00:16:29.039 --> 00:16:31.120
And is it that was the first mistake I made.
00:16:31.120 --> 00:16:38.240
Well, yes, and and I I love the fact that we ended with the it's not gonna fix the problem.
00:16:38.240 --> 00:16:41.360
Because actually that is the that that's the question, isn't it?
00:16:41.360 --> 00:16:48.240
And that's the point of, as you say, going to gemba, meeting the people, talking to them about the work they are doing and what are the problems they face.
00:16:48.240 --> 00:16:54.399
Um so that you're not sort of blanketing AI over them, you are still a problem solver.
00:16:54.799 --> 00:16:55.279
Yeah.
00:16:55.279 --> 00:17:18.559
And and you know, um, I suppose in my years now, I don't want to think about how many, too many years, um, I suppose I've kind of been in that strange position where I've been the I've been the doer, I've been the practitioner, and now I'm kind of I suppose a leader of people, and I love flipping the roles, so I don't go down and say, oh, let's do this, let's do that, let's try this.
00:17:18.559 --> 00:17:21.200
I'm like, what would you like to just to do?
00:17:21.200 --> 00:17:22.640
How can I help you?
00:17:22.640 --> 00:17:25.759
How can I create an environment for you to succeed?
00:17:25.759 --> 00:17:30.640
I might know some of the things, I might make some suggestions, but actually, what is it you need?
00:17:30.640 --> 00:17:33.759
Because then the ownership's with the practitioner, the ownership is with the team.
00:17:33.759 --> 00:17:35.839
Tell me what you need and I'll go rebuild it.
00:17:35.839 --> 00:17:39.359
I'll come back to you with a thing, but then it's your tool to then use.
00:17:39.680 --> 00:17:40.000
Yeah.
00:17:40.480 --> 00:17:46.160
And stop thinking about, well, I have this tool, how can I ram the tool into the problem?
00:17:46.160 --> 00:17:47.440
It's well, what's the problem?
00:17:47.440 --> 00:17:50.000
And then let's try and get the right tools to fix that problem.
00:17:50.000 --> 00:17:58.240
And again, that might sound really basic, but there's just so many it's so quick to spin up AI tools now that we're gonna get into the situation.
00:17:58.240 --> 00:17:58.960
There's so many tools.
00:17:58.960 --> 00:18:00.000
Well, which one am I supposed to use?
00:18:00.000 --> 00:18:00.640
No, no, no.
00:18:01.519 --> 00:18:04.559
I've got one of these and I'm gonna use it no matter what happens.
00:18:04.559 --> 00:18:05.200
Absolutely.
00:18:05.200 --> 00:18:07.039
No, no, I completely agree.
00:18:07.039 --> 00:18:14.720
Now you've you've you've you know you've you've wet my appetite here, because you said that was the first, so that feels like there's at least a second that you want to share.
00:18:15.440 --> 00:18:16.000
Yeah.
00:18:16.000 --> 00:18:24.880
I think I think the second problem we had, oh problem, uh, is um foun foundational knowledge.
00:18:24.880 --> 00:18:58.640
So um I talked a little bit about um developers understanding the practitioner's knowledge of data, but when you talk to non-technical people or non this particular example is about SPC, but it could be about anything, um as soon as we deviate away from a uh a change cycle, a change practitioner, and you start to take these kind of conversations to um non-change enabled board members or senior leaders in an organization, this may not be their you know, their their bag, this may not be their interest.
00:18:58.640 --> 00:19:04.079
Yes, of course, everybody wants to make more profit or make higher margins or whatever it may be.
00:19:04.079 --> 00:19:12.720
But when you start to say, Oh, um I want to have um half a sigma what it doesn't make any sense.
00:19:12.720 --> 00:19:29.680
And so when we started to talk about this particular project internally, again, my f my fault, I thought, oh, hang on a minute, everyone will understand all of this gobbledygook, all of this, you know, even just the word gemba, many people won't know what that means.
00:19:29.680 --> 00:19:42.480
And so trying to talk at the right level and talk with the right kind of language in a mixed audience on a something as specific as AI and SPC was definitely a challenge and not one to be underestimated.
00:19:42.480 --> 00:19:46.480
And so uh I'm sure my mum won't mind me saying this.
00:19:46.480 --> 00:19:50.559
Um I revert to uh does this make sense, mum?
00:19:50.559 --> 00:19:52.000
Could you listen to this?
00:19:52.000 --> 00:19:54.160
And can you understand what I'm trying to say?
00:19:54.160 --> 00:19:57.839
Because sometimes if it doesn't pass the mum test, it doesn't make sense.
00:19:57.839 --> 00:20:15.920
And if it doesn't pass the mum test, then how is somebody who's not um who doesn't have the understanding of of of the value of, in this case, SPC, and of the value of of proper structured, framework to change, um, they're not going to get it.
00:20:15.920 --> 00:20:17.680
And so that was the other mistake I made.
00:20:17.680 --> 00:20:19.200
I ran far too fast.
00:20:19.200 --> 00:20:25.680
Once we had established what was needed, I ran to solution, then came back, tried to explain it to the board who went, what are you doing?
00:20:25.680 --> 00:20:34.640
Um stakeholder management wasn't the best on this particular project, but hey, I'm glad we did it because we we've built a solution quickly.
00:20:34.640 --> 00:20:44.480
We didn't go through a huge um a huge cycle of let's get all the requirements down first, let's build everything out to perfect.
00:20:44.480 --> 00:20:47.200
It was much, much more agile in that respect.
00:20:47.200 --> 00:20:53.599
Um, but the catch was, yeah, I probably should have brought my stakeholders on board a little bit sooner.
00:20:53.599 --> 00:20:55.200
But um, yeah, that was on me.
00:20:56.079 --> 00:21:05.920
I think yes, but I also think that it that the again what you described is not an uncommon challenge, in as much as it's it's hard, isn't it?
00:21:05.920 --> 00:21:10.319
Here we are, we're let's say we're you know we're driving change, we're driving improvements.
00:21:10.319 --> 00:21:13.279
I might have eight stakeholders to work with.
00:21:13.279 --> 00:21:32.799
Um my best chances of success are to step back, look at each of them, work out what's going to drive them, work out where they are, work out what language I need to be using in order to talk to them, and to probably tackle them individually before trying to bring them all together at once.
00:21:32.799 --> 00:21:49.759
But actually, I'm really busy, you know, and and and therefore sometimes taking that step, slowing down slightly in order to speed up, but understanding that actually it's our job to adapt to them and their needs.
00:21:49.759 --> 00:22:05.039
Yes, it's not it, you know, just because we think that um and we might be right that we've got the best solution, um, it doesn't mean that they will even begin to uh understand it, and not because they're stupid, just because we're not talking the same language.
00:22:05.279 --> 00:22:06.160
Yeah, exactly.
00:22:06.160 --> 00:22:18.880
And and you know, we can we can have as much data as we like to prove something or disprove something, but if we can't communicate it in a way that's meaningful for the recipient, then don't communicate it, because it won't land.
00:22:18.880 --> 00:22:22.000
And they may end up with the wrong end of the stick altogether.
00:22:22.000 --> 00:22:24.000
And that happened in this case.
00:22:24.000 --> 00:22:28.240
What happened in this case is someone said, Oh, we don't need easy SPC then.
00:22:28.240 --> 00:22:30.319
And I was like, no, no, no, no.
00:22:30.319 --> 00:22:31.519
Hang on.
00:22:31.519 --> 00:22:38.960
This isn't about replacing what we've already got, it's about augmenting, about enabling more, enabling richer.
00:22:38.960 --> 00:22:40.880
Ah, right, okay, cool.
00:22:40.880 --> 00:22:49.279
So it was using the right language for the recipient, which is difficult, especially if you've got a board or you've got a group of mixed knowledge or mixed understanding.
00:22:49.279 --> 00:22:51.359
That can be a real that can be a real challenge.
00:22:51.680 --> 00:22:52.319
Yes.
00:22:52.319 --> 00:22:53.759
Very interesting.
00:22:53.759 --> 00:22:56.000
Fraser, thank you so much.
00:22:56.000 --> 00:22:58.960
I really appreciate your honesty as well.
00:22:58.960 --> 00:23:01.839
Um, and for sharing your stories with us.
00:23:01.839 --> 00:23:03.279
It's really it's been great.
00:23:03.279 --> 00:23:04.559
Thank you so much.
00:23:04.559 --> 00:23:08.079
Um, to our listeners, I do hope you enjoyed this episode.
00:23:08.079 --> 00:23:15.759
Um, and uh you can find Fraser on LinkedIn if you want to connect with him and talk about AI some more or SPC.
00:23:15.759 --> 00:23:25.680
Um and please don't forget to subscribe, share, leave us a review, because the more this podcast grows, the better everyone's work becomes.
00:23:25.680 --> 00:23:27.680
Look forward to seeing you next time.
00:23:27.680 --> 00:23:28.240
Thank you.
