Ep 23 - a look back and some thoughts on AI strategy

In this episode, Zoe and Kevin discuss their thoughts on the biggest topics of the year in Copilot and what that means for next year before Kevin goes solo to talk about the importance of building a comprehensive AI strategy centered around Microsoft Copilot. They explore the evolution of AI tools, the need for alignment with organizational goals, and the significance of understanding agents and their capabilities. The conversation also delves into measuring success and ROI in AI implementation, the future of AI tools, governance and compliance, and the role of generative AI in business processes. They emphasize the need for a Center of Excellence to support AI initiatives and prepare for the future of work.

Takeaways

• Building an AI strategy is essential for effective implementation.
• Alignment with organizational goals reduces risk and enhances productivity.
• Understanding the capabilities of AI agents is crucial for their effective use.
• Measuring success in AI implementation goes beyond ROI; consider overall impact.
• Democratization of AI tools will make them accessible to more users.
• Governance and compliance are critical in the responsible use of AI.
• Generative AI can enhance business processes but should be used judiciously.
• The hype cycle illustrates the fluctuating expectations of AI tools.
• Preparing for the future of work involves understanding AI's evolving role.
• A Center of Excellence can facilitate the successful adoption of AI.

Key links

• Future of work report 2024
• GitHub Copilot free tier
• Simon Owen talking of CoE and C4E in his blog

Transcript

Kevin McDonnell (00:08)
Hello and welcome to the Copilot Connection.

Zoe Wilson (00:12)
We're here to share with you all the news, insights and capabilities of the Microsoft Co-Pilot ecosystem from across the entire Microsoft stack. I'm Zeyby Wilson and I lead the Co-Pilot Business Transformation Practice across Accenture and Avanade. I'm an MVP for M365, a Regional Director and Viva Explorer.

Kevin McDonnell (00:31)
think that's a new role, but let's let's talk about that in a minute. I'm Kevin MacDonald. I'm an MVP, Viva Explorer and the co-pilot, strategy and modern workplace AI leader, Avanade. We'll be releasing episodes as podcasts and on YouTube with insights from across the community and Microsoft and what the different areas of co-pilot are, the impact to you and your organization, what you need to do to prepare for them or start implementing them now, and in this current age, even more about how you can extend them.

Zoe Wilson (01:02)
So today I'm actually going to run away and hand over to Kevin for the bulk of this because I've got a huge to-do list at work before I come and wrap up for Christmas and start enjoying the festive fun. before, no, not yet, not yet. Yeah, well, let's pick that up in the new year. But before I do hand over to Kevin, we thought we'd just take some time to reflect on some of the changes and progress across the co-pilot ecosystem in the last year. And actually,

Kevin McDonnell (01:13)
Okay, so I can't ask about the new role yet then. We'll save that.

Zoe Wilson (01:31)
I mean, when you think about it, Kevin, what is it like two years and a month now since chat GPT entered the collective consciousness of the world population? Yeah. And, and even when you think back to this time last year, mean, M365 co-pilot had literally only just gone GE. We had lots of announcements, but not all real products that people could use. So it's been a year.

Kevin McDonnell (01:37)
Yeah.

short.

Yeah, and I think it's been nice and I think the, you know, the big moment that was that I think if we're looking certainly the M365 copilot side was obviously the wave two announcements and that the focus there was improving quality and getting more features and kind of addressing some of the, I don't know if I want to say limitations, I might get in trouble, but to me it's limitations. was things like Excel didn't work that well. PowerPoint was a bit disappointing initially.

And I loved how Microsoft addressed that. I think we saw that. I've seen that as a theme through the year of learning quickly, learning what it is that's not quite working and trying to adopt that on there and renaming with it and changing SKUs. Absolutely.

Zoe Wilson (02:41)
Let's not talk about renaming either, we can pick that one up too in the new year.

Kevin McDonnell (02:48)
But I think that's been the big thing. And if we look at other areas, we're still seeing OpenAI releasing new models. We're still seeing on the Azure side, additional models as well as OpenAI being brought into it. So they're kind of broadening the capabilities. Talks of PHY3, now PHY4, and the small language models are kind of bringing more opportunities to there. We're seeing Copilot plus PC really getting landed. you kind of got that option of

Do you go to the clouds for your co-pilots or do you come local? They'll be starting to see a, I don't know if a shift, but I think two different ways that you can adopt from that as well. And I think we've seen that happen through the year.

Zoe Wilson (03:29)
Yeah, I

think the Copilot plus PCs are super interesting as well because it feels like the hardware is actually ahead of the software because we've had these devices that have been available for a little while now.

Kevin McDonnell (03:42)
Yeah.

Zoe Wilson (03:45)
And I mean, it's great that we're starting to see more options for some of the small language models and more of the things that you can do locally. But it feels like the hardware is ahead of where the software is and everything's racing to catch up. And when I think about some of the things that we might see in 2025, I think that will be a really important part of it.

Kevin McDonnell (04:07)
Yeah, yeah, absolutely. And I think with the software, it's people realizing what they can do, what the best way to do it. When do you go in local versus when do you do the cloud? I think coming outside slightly, the Microsoft ecosystem, Apple Intelligence has been there. I am loving a lot of Apple Intelligence because it's summarizing things, but it's doing, I believe, bits of that locally. And then some things are going to the cloud. I know you can kind of ask Siri and it will have chat GPT.

Zoe Wilson (04:35)
How well do you think it works?

How well do you think it works?

Kevin McDonnell (04:39)
I

OK. OK, I'll give you that. was looking through my photos to try and do a family calendar and the number of screenshots I've got of amusing summaries from from ones I think I counted 30. So, yeah, OK. Sometimes it's a little kind of misses the theme, but the number of times I've also seen it go, do I need to look at these messages? No, that's just another WhatsApp spam.

Zoe Wilson (04:51)
Hahaha

Kevin McDonnell (05:06)
group of everyone getting overexcited about something that I can look at later. for me, is it perfect? No. Has it helped me prioritise? Yeah, completely. And I think this is also that big shift towards not just a chat bot for Copilot. It's kind of bringing things and we, you you and I have been lucky enough to see in Outlook the prioritisation of emails. I love that because it's bringing the right emails to the top. I think 95 to 99 percent of the time it's...

Zoe Wilson (05:34)
Mm-mm.

Kevin McDonnell (05:35)
kind of helping me prioritize. And I think there is that shift as well of bringing it into the day to day products too.

Zoe Wilson (05:42)
Yeah, I do. I do have to laugh because I've got like an intelligent, well, not intelligent. I've got like a sensor on my front and back door, which is connected to our wifi. So I get a push notification whenever the doors unlocked or locked. So I can easily tell if, we've forgotten to lock it and if the door status changes a lot.

Kevin McDonnell (06:05)
Can I just say it was very

good you said we and not your other half forgot to lock it. Well done.

Zoe Wilson (06:09)
Yeah, well, could could it could easily

be be either of us, but it's just it's a really nice little feature from a security perspective, especially because we get paranoid about the cats escaping. So it does make me laugh because if I like my other half's just got back from the shop. So I've got an alert to say the doors open and then another one says it's locked. And then the Apple intelligence, if that happens quite a few times and I've not actually looked at the notifications, the Apple intelligence will say.

Kevin McDonnell (06:27)
Yeah, as if by magic.

Zoe Wilson (06:37)
Door state has changed many times. So it's not wrong, but it's not really helpful.

Kevin McDonnell (06:39)
Yes. Yeah.

Yeah, I get that with I have it on the cat flap and I get Smulan and Slinky came in and out quite a lot. It just made me laugh. But it but again, I'd much rather that than a bombard of like 20 different messages. So I love that kind of grouping together of things. So, yeah, I take your point. I kind of like it. So but I think it's that it's that shift in focus that we've seen there that that move away of bringing the.

Zoe Wilson (06:50)
You

Yeah.

Kevin McDonnell (07:10)
the kind of generative AI capabilities in in different ways as well. And, you know, I might talk about this a little bit more later, but I think agents as a kind of broad term for extensibility, I think it's I don't want to eat up your time too much, because I know you're going to run soon. But I think that view of

kind of making it easier for people to understand. think we spent quite a lot of time talking about extensibility. When do you create a custom copilot versus an extension and what is a copilot didn't quite resonate. Whereas now I think the agent view, people kind of get that they think of a travel agent, they think of different types of people and they can kind of see that. I think for those who have got slightly more understanding and thinking.

It's kind of confusing because it's lumped together all these different capabilities in one big word, which is the one challenge. But I think for those who don't, which is the vast majority, I think it has helped things as well.

Zoe Wilson (08:13)
Yeah, it's really interesting, isn't it? Because I see a lot of people just simplifying agents and talking about hyper automation. And I feel like in the same way that for the last 18 months, people have had to spend time understanding what copilot is, regardless of which copilot it is, the fact they've had to understand what copilot is. We've now got a similar journey that people need to go on to actually understand.

Kevin McDonnell (08:21)
Mm.

Zoe Wilson (08:39)
what agents can do and automation will continue to play a role because there will be some things that we can do through extending Copilot or integrating these different parts of the system together and bringing in other bits of data where it is just automation. But actually for people to understand that agents is where we start to introduce reasoning over the top of processes and so to...

Kevin McDonnell (08:55)
Hmm.

Zoe Wilson (09:05)
almost give away a little bit of that agency and the human in the loop obviously will still continue to be important very much in the early stage. think I know people talk about autonomous agents, but I think there's a trust relationship that people need to build with this technology. So understanding all of the different types of ways that we can build agents. And I know this is something you talk about quite a lot, Kevin, but understanding that

Kevin McDonnell (09:22)
Absolutely. Yeah.

Zoe Wilson (09:34)
And then also understanding where that additional reasoning over the top of automation or that complements automation is a journey that we'll need to take people on.

Kevin McDonnell (09:45)
And I think also helping people understand how to get that benefit from agents as well, know, is understanding what they are and then understanding what they matter to you and then how those agents can talk to each other. I think this is for those who are involved in rolling this out and doing this, the reality is going to be these things aren't going to be able to do what you want them to do for a little while now.

There's going to be bits and pieces of that. There's a vision that's there. Six, nine, twelve, eighteen months away as it kind of goes through. And I think we we've got to be aware of that while also selling that kind of exciting potential. And if you don't start doing it now, you're going to get left behind. know, this this is the kind of Netflix blockbuster type moment. There is the opportunity to be very ahead with this and a risk with it as well. So I think that's something that

that people really to kind of think and plan about as well.

Zoe Wilson (10:48)
Yeah, one thing I would

just like to go back to is, you know, talking about the things like the wave two focus on quality and the improvements. I know that we've talked on previous episodes about the reasons why GitHub Co-pilot has been more well received by a greater amount of people and it's because, well, partly it's because it's more mature. It's been around for longer. So some of the...

Kevin McDonnell (11:15)
Is it because they kept

the same name?

Zoe Wilson (11:17)
Well, yeah, that makes it really easier to actually understand as well, doesn't it? But yeah, people have had access to this for longer. It's more mature as a technology, but more importantly, it's in the flow of work. And one of the things that I have started to see much more with the focus on quality and some of the changes that have been brought into the different parts of the co-pilot products is

Kevin McDonnell (11:18)
Yeah.

Zoe Wilson (11:41)
more focus on how to make it more natural to use, how to make it more automatic, instead of people asking for a co-pilot summary, it being the first thing that you see when you go to an email or when you open a document. And that's a really interesting trend, I think, for me, because as we start to see it become more natural and in the flow of productivity work or functional work, it will become easier for people to use. And just to bring that to life a little bit.

Kevin McDonnell (11:51)
Yeah, great.

Zoe Wilson (12:11)
I was talking to people who worked in an internal IT function a few days ago, and they were talking about the usage stats of Co-Pilot, specifically M365, in their organization. And to the point you made about quality in things like Excel and PowerPoint, those things had been trailing in terms of active usage.

But what they've seen over the last few weeks is actually quite a significant jump in the amount of people who are using copilot in those products regularly, which I think is testament to both the improvements in quality and output and what it's actually doing, but also the fact that it's being, that when it appears, when you first interact with a file that's in Excel or in PowerPoint.

Kevin McDonnell (12:39)
really.

Interesting.

Zoe Wilson (12:59)
you're being surfaced with things that are relevant and prompted to do things with it in a different way, which is more intuitive and more in the flow of work.

Kevin McDonnell (13:02)
Hmm.

Yeah. And I think I've told this story a few times. One of the reasons I love GitHub Copilot is exactly because it's just there. And you often don't notice it until it's turned off. And then you're like, help, help, I can't do things. And I do wonder if there's a certain element of that with some of the M365 Copilot and some of the others now is why do we need to turn it off and see people panic and go, hang on, I can't do these things and demand it back to see the kind of true value and the worth of it as well.

Zoe Wilson (13:18)
Hahaha

you

Kevin McDonnell (13:38)
I think that would be good. I know, Ian, I want to let you despair off. So I think one thing, and I'm sure we'll come back to this in either the between times of Christmas and New Year or early in the New Year, I think we'll start seeing what do we think will happen. But the one big thing that we're already starting to see now is that shift to give access to more than the premium side of things with copilot. So we've seen announced

Was it yesterday, day before, that GitHub Copilot is now free within Visual Studio for a kind of limited number of usages, but for the majority of kind day-to-day developers, that will be enough for their usage that we'll be able to get there and we'll share links that's got full details of that. We're seeing in the message center that now the SharePoint agents and agents within M365...

will be available to those without an M365 Copilot license and there will be a, I'm trying not to say free, an included tier for that and then the ability to have pay as you go on top of that. So suddenly you don't need to pay for a license for like your frontline workers who suddenly can get access to these agents that can be built into those frontline worker devices without having to have that full commitment to everyone having that $30 per user per month.

You can start to think and plan against that. Now, there's lots to think about there. There's governance stories. There's how do you manage that? But as a kind of model moving towards there, we're seeing this open up to more people. And I'm really happy about that. I think it's something we've we've been kind of hoping for for a while to see these capabilities get to more people and then make a choice of do I need to get that value of going premium? Do I need do I get enough usage from this? And I think that will be another

It would be another interesting thing to keep an eye on over the next year of how people adjust and work that out,

Zoe Wilson (15:42)
Yeah, definitely. think introducing different models for people to be able to use and consume these technologies will definitely start to democratise access and make it easier for people to use the bits and bits that are relevant for them.

Kevin McDonnell (16:01)
Yeah, and I think it will be interesting to see what disappears as well. I think we've already seen some of the Microsoft things. Viva Goals has had the announcements. I think we saw last night that the M365 search results in Bing is being deprecated as well. So we're starting to see Microsoft

Zoe Wilson (16:24)
Yeah, that one's a particularly

annoying one for me because I use that work results tab in Bing all of the time. you know, I think when that was first announced as something that would be available within Microsoft search, I actually thought that was really valuable. I know a lot of people that use it heavily, but it just shows that a lot of it. I think some of these decisions show.

Kevin McDonnell (16:29)
Yes. Yeah. Yeah. Absolutely.

Mmm.

Zoe Wilson (16:54)
where Microsoft are choosing to invest their money at the moment.

Kevin McDonnell (16:58)
Yeah, I think to my view, and I know we won't see it from Microsoft about the amount of usage, my assumption on that is that they're looking to bring that into, bring people more to M365 homepage. Whereas if you go in Bing search, you don't get to that. So my suspicion is that there will be a push to people to go into, I was about to say office.com, but whatever the URL will be from

Zoe Wilson (17:25)
Hahaha

Kevin McDonnell (17:27)
from that as well. yes.

Zoe Wilson (17:30)
Are you thinking there's another

URL, more URL changes, renames? Are we taking bets on how many we'll see in 2025?

Kevin McDonnell (17:41)
I think we know there's going to be a year old change because they're moving to what's it dot Ms dot cloud. I've certainly forgotten it all dot Ms cloud, isn't it? I think is the you know, there is that aim to move everything towards there. So I'm hesitating and pausing because I can't remember what's being talked about publicly. I know that has been talked about publicly and we know we know there will be shift before everyone panics.

Zoe Wilson (18:00)
Sorry, that was a little... Yeah.

Kevin McDonnell (18:09)
all those redirects will be in place. you know, this is not new news. This is something that was going to be happening within there. And they're not going to suddenly turn things off. For example, I can still go to Yammer.com as a natural habit because I can never remember the engage dot something dot something on there. And that still works. So that that's not going to go away before people start worrying about things.

Zoe Wilson (18:27)
Hahaha

Yeah.

So, so I guess thinking about what, what, what we might see or what, we might come and talk to you all about in 2025. I do think we'll continue to see more product renames, rebranding, repositioning as Microsoft get more customer feedback, more feedback from the market, from their partners, from MVPs. So, so that shouldn't be a surprise to anyone. One of the things I'd

Kevin McDonnell (19:00)
And do feedback.

I'd say this everyone, do feedback if you don't like things, tell Microsoft. It's not a guarantee they're going to change things, but if you don't tell them, they definitely won't.

Zoe Wilson (19:02)
Yeah.

Yeah. I mean, it's an interesting point, isn't it? Because we had Vesa and Wodeck from Microsoft on the Co-pilot fireside chat just earlier this week. And people were asking questions of them about governance and why features are rolled out with the default of being on and why we don't have the governance in. But it's a really interesting Catch-22 situation because we need to get the features out and people need to give that feedback.

back to Microsoft or to their partners who can then feed that back to Microsoft. that they so that Microsoft actually understand whether they've got product market fit, you know, does this thing actually make sense? Is it going to work? Is it helpful for people? And then what are the gaps from a governance perspective? Because if you try to design the governance first, and you end up with a product that doesn't work,

that's a huge amount of investment and it also means that they're actually slower in getting these things to market as well. So I am one of the loudest voices when it comes to things like governance and compliance and controls. But I equally understand that they actually also need to get feature feedback as well.

Kevin McDonnell (20:19)
Yeah. And I think people got to be aware. Don't just grumble. Say why you're grumbling. Say what it matters so that Microsoft can take a decision. Because the reality is, I'm not going to upset people with this. People generally aren't very good with change and new things happening. And it kind of scares people. So the natural thing is, I want to control, I want to this off when I'm, sorry, turn this on when I'm ready. And I get that. I absolutely get that. But

If everyone did that, we wouldn't see the change happening and Microsoft will get left behind by others who will be moving quicker. So there's got to be a balance. You've got to be clear on what the things you want are, not just I don't want to turn this on because that's that's not going to help. Microsoft's not going to take that on board from there as well. Although I do know, you know, I talked earlier about how Microsoft is learning the new pay as you go agents are off by default. Yes.

They're not just turning these things on. They have listened to that feedback. So you have to actively go on and turn this on. And I'm really happy about that. I think that's been a good thing. And I think something that people have missed in some of the messaging.

Zoe Wilson (21:24)
Is that because you need to put your

credit card details in?

Kevin McDonnell (21:29)
I'm just trying to be nice. Well, actually, no, I'm not. I'm not sure it is just that, because if you've got if you've got the syntax pay as you go, it's still not on by default. You still have to go and turn it on. So there is an element of that. I was thinking that. There's a part to it on that, but no, it certainly isn't just that, because if you do have it set up, you still actively have to go and say, yes, I want this. So.

Zoe Wilson (21:29)
Yeah, sorry, sorry, I couldn't resist.

Yeah, no it wasn't a serious comment that was slightly facetious.

Hahaha

Yeah. Yeah. And then other things that, that I'd like to chat about a little bit more in the new year, think have been, I know we've talked about this, already, but I think it would be really good to look at some of the, competitor platforms. So, we know that Salesforce, for example, I see is incredibly vocal about,

Kevin McDonnell (21:57)
It is good news.

Mmm.

Zoe Wilson (22:23)
why their AI is better and all of the shortfalls and shortcomings of co-pilot. Yeah, and I know that a lot of the other big players are also investing in and rolling out their own capability as well. So I think it would be really good to come and have a bit of a chat about that. And then the other thing, if people have not seen it.

Kevin McDonnell (22:28)
But it's a waste of time, according to him.

And so just

quickly on that, sorry to anyone listening, we'd love to know what ones you'd like us to compare and have a look at. mean, in my head, as you say, Salesforce in their service now workday is the big ones. We've got Google and their kind of Gemini and Workspace offering SAP. But and we're probably not talking about necessarily Claude and the different models quite so much. We're more about implementations of those models that kind of fits that copilot worlds as well. Yeah.

Zoe Wilson (22:55)
I say P.

Yeah.

Yeah, scenarios, use cases.

Yeah. And then the other thing that will be really good to talk about probably early in the new year is the new future of work report that Microsoft published this week. It's very lengthy. We did want to talk about this today actually, but it's 52 slides of incredibly detailed research. in all transparency, we haven't got there yet. Yeah.

Kevin McDonnell (23:12)
And I'd love to know what we're missing.

And it's a proper research report, isn't it? You this isn't a fluffy

work lab. Nice. Let's pick a few numbers. There's real detail in this that I think we need to not just throw copilot and pick out the bullet points on there. I mean, we're saying that now, but when it comes to January, we might do, but the aim is to read it and kind of properly digest and take that on, definitely.

Zoe Wilson (23:44)
Yeah.

Hmm.

Brilliant. So with that, Kevin, I think I'm gonna leave you, leave it, yeah.

Kevin McDonnell (23:55)
And I think I did see.

Just very quickly on that, because I'd

love people to dig into it. I think it was Abrahan Jackson or someone put in the comment when it was out and they were saying one of the one of the parts of it is you shouldn't just use generative AI. You know, it was almost kind of we're pushing things too quickly. Don't just take it on board. I thought it was fascinating. There obviously is content in there that might go against some of what parts of Microsoft are pushing. So really interested to to dig into that deeper.

Zoe Wilson (24:23)
Yeah.

Yeah,

so I was actually, was reading it before this call, I was just trying to find the stat. So there's lots of different, obviously we're not going to dig in that, like we said, there's a huge amount of detail, but just this one in particular really stood out to me. And they've looked at lots of different research studies as part of pulling this report together. And one of the early bullet points in this is that among

100,000 workers in Denmark who work across 11 occupations, half reported using chat GPT. And there's a range of adoption across different roles. Younger, less experienced, higher achieving male workers led adoption. And it was the fact that this was young high-performing men who were the ones who were driving. And that was really interesting to me.

Kevin McDonnell (25:12)
You

Zoe Wilson (25:22)
I really need to dig into that a little bit more.

Kevin McDonnell (25:26)
So it's the frat boy takeover.

Zoe Wilson (25:29)
Yeah, but this is one of the things that's potentially worrying because we've already got gender disparity across a lot of industries. will this, and we've already got issues with things like the missing room for younger women who want to get into management roles where it's harder for them. Is this going to actually make it worse by the male led adoption of these tools that will give them that edge over the people that they're working with?

Kevin McDonnell (25:36)
Hmm.

OK, now we need to remember to dig into this more because I'd love to understand why that that feels I don't know. Doesn't feel right. I can't see why that is that apart from maybe how these things being promoted on this. So, yeah, yeah, that's really interesting.

Zoe Wilson (26:10)
Mm.

Yeah, it's interesting though, wasn't it? So definitely go read the report and we'll

have an episode where we really dig into some of the info in it.

Kevin McDonnell (26:23)
Yeah, yeah, definitely. Right. I'm let you run. I'm going to say this at the end of my bit later for everyone to switch off, but I'm going to say this to you as well. Make sure you switch off over Christmas and and take a bit of a break.

Zoe Wilson (26:34)
Believe me, I will. And

that's actually why I'm running now so that I could do the things that I need to do and get them off my mental workload and actually start to shut down. for you, have a fantastic Christmas for everybody listening. hope you, regardless of whether you celebrate Christmas or a different holiday or whether you just enjoy having some downtime, make sure that you have a wonderful end of the year and we'll be back with you very soon.

Kevin McDonnell (26:45)
Yeah, very good advice for everyone.

Thank you.

Kevin McDonnell (27:12)
Well, thank you, Kevin and Zoe, or rather past Kevin and Zoe, because I'm going to take over Solo for a little bit now. And I want to talk to you about how you can build an AI strategy with Copilot. So the aim of this is to, probably guessed on the show, we talk about Copilot quite a lot. But what I want to cover is that it's not just about using Copilot, it's about

How do you tell people what you're using? How do you keep people aligned with that? How does that fit into how you're going to use AI strategy? How do you can use AI within your organization itself? So let's have a little think about this. I was launching along on the video. Yes, I do have some slides. I think it's about the first time we've had slides on the show. there's a first for everything on that.

The kind of first thing to ask is, why is it you can't just have conscious let people use the AI they want? Why can't you just let people use the different tools, grab chat GPT and things like that. It may feel a bit like a consultant trying to push strategy sounds like someone trying to sell things on there. But the reality is that if you don't define a strategy, people will use all the different tools.

And these tools might not have the right evaluation. They might not have the right capabilities to kind of put the appropriate security, the appropriate compliance. You could be at greater risk within that. So what is it that a strategy can help with? Well, the first thing is that alignment. If you've got that strategy and says, this is what we want you to do, this is what we're going to do, then

everyone is in that same game. Everyone's more likely to push around that. I used to speak a lot on OKRs and VIVA goals back in the days before they decided to retire it, which I'm still very sad about. actually joking on that. But one of the superpowers of OKRs is alignment, because if you can get everyone going in the same direction, then you've got people thinking the same ways as a real benefit of that.

And in fact, then if people are doing that same thing, pulling in the same direction, then you've reduced the risk within them. sorry. You haven't reduced the risk by getting everyone the same, but you can look at that same technology, that same way of doing things and analyze the risk there. And you just have to do it that once.

So if you align to Azure OpenAI, for example, then you can understand how Microsoft is using that data. It could be you decide to align to some of the Claude models and then you understand where that data is going to be, how they're using that data to train their next models, where it resides, what are the protections in place to secure your data on there. You understand that risk and you've done it once for the system and it's covered.

If you're taking on all the different models, all the different platforms, from Salesforce Einstein to Microsoft's Co-Pilot to what's it, SAP's Jewel within there. If you've got all these different ones, then you're having to analyze all the different bits of risk. Not saying don't do that, but it's just making sure that you have that understanding that have all those things to do. Because if you do need to do that, then the cost of that increases, the cost of

maintaining these different items, these different tools that are within there as well becomes higher. And actually, if you have that single platform, you can get your negotiations together with that. You can get those conversations happening, that consistent licensing that can help reduce the cost. It is worth saying, if you're only with one platform and you're completely wedded to that.

then you may be stuck there if you're trying to negotiate and have a debate around the best deals, you'll cut a ham's junk because they know that the amount of work it would be to unpick from that. there is a certain amount to consider with with having maybe a couple of options within that. But, you know, in the same way that you may use Microsoft 365, you don't want to be partly using that and partly using Google's workspace because your your cost is going to be

much larger by taking on adoption. Yes, you have that flexibility, but how much you have spent on balancing both of those at the same time kind of outweighs some of that as well. So thinking about how you can focus on one that see that cost reduction is really, really powerful as well. Moving on, the other one is future proofing. So if you

If you have got that single platform, then you can look at understanding what their models are, understanding what's happening next, having that engagement and being able to plan ahead for that. The more different things you do against that, the more painful it can be. And I think also more importantly, with that strategy, you can plan for what's going to happen in the future. You can think about what that need is within there.

If you're just using all these different ones and not putting things in place, then you haven't thought about, yes, this this tool might do it now, but that tool is going to have that capability plus more in the future. So really understanding not just what the here and now is, but what the either the definite plans or the likely trajectory of things will be. Is this organization that you're aligning with constantly improving? Are they going to be first to the game?

Are they going to focus on being stable and giving you the governance, which may be more suitable for you? So planning that strategy can help you be very clear on why you're doing those things. And one thing that's almost got missed on these slides there is capability. Do you have the internal understanding to be able to do these things? Do you have the training or do you have the ability to to ramp up that training? So.

If you're very Microsoft aligned house, then people are probably aligned to using Azure. And if you suddenly decide to bring in maybe plain open AI or some of the Claude services, there's a different way of thinking that you need to train people up and thinking about within there. So working out the strategy for how you're going to get people engaged and not just from the back end point of view, not just looking at those models, but also thinking about how the end user is going to do this as well. So

Think about those different areas. There's a few steps I like to think about when you're building this strategy. the first one is identify the need. Because as someone who loves a good gadget, who loves to bring things in for the sake of it, I do think it's important, especially for a large organ, well, actually for any organization, you need to think about what the need is, because that will help define your strategy.

Are you looking this as a broad brush to improve productivity? Are you wanting to increase innovation? Are you wanting to be seen to be innovative? And I, a kind of techie focused person, this is kind of tricky. The reality is sometimes it's not just what it can do, but how you're perceived to be there in the market. So that that marketing viewpoint, I do think plays into it, can come along to that as well. It's not always everything, but

is certainly some of that needs. Are you perceived as being an AI first organization in there? Or are you perceived as maybe being a little bit more cautious and risk averse within there? Thinking about what the need of what you want to get across is will also steer towards what tools you need. And talking on those tools, given that you have a need, whether it's a very specific one, or as we're looking at kind of strategy for the organization, you're looking at a broader need for that.

What are the tools that can deliver that? And that means you need to get a bit of an understanding of what those options are out there. We mentioned earlier that we're going to try and do a show in the new year where we look beyond Copilot and look at what the alternatives are. And you need to make sure you do that. Whether you are looking at that seriously to say, I don't want to use Copilot or maybe Azure Open AI, a particular element.

I want to use something else or whether you just want to understand what the options are and what the differences are, you should understand what those tools can deliver. And even even looking within the co-pilot space, should you build your own thing with this? You're open. I should you make an agent? Should you use out of the box services as well? So thinking about those different options for what you need is very important. And then going back to that capability I talked about,

identifying the required resources to be able to do that. Are there people in-house that you can train up? Do you need to bring in a partner or hire new people to be able to do this? Thinking about how that aligns to the strategy and that could be, again, for implementing. It could be for training the staff. Friend of the show, Sarah Fenner, is doing a beautiful amount of training around Microsoft's co-pilot and M365 co-pilot and that world.

as organizations look to kind of gear people up towards that. So think about what is the resource that you have, what you need, what's the gap within there, how do you work that out? And that then connects into assessing the value. When you bring these tools in, they're going to cost, whether it's that license cost and the kind of upfront implementation, and whether it's that training, that change in adoption, all these things are going to add

a cost to it. But what is the what is the flip side? What are the benefits? You could say what is the return on investment? I'd always challenge people to look beyond that as well, not to ignore ROI, not to ignore the fact that you need to look at the monetary value of these things. But to say there can sometimes be more than that. What is the impact to your organization? What is the impact to the staff? And yeah, I know you could say if

People are less effective, then there's less money. So it's all down to ROI, but it's kind of working back. And I do think that's a valid point, but it's how do you get to a real way of assessing that and measuring those values on there? I always love that kind of idea of what happens if you take this away from people. I talked about this earlier with GitHub of the idea of turning off and seeing the impact. How would people feel if you took these tools away from them? That really helps to play into that broader value as well.

You need to think, I talked about the risk, but you need to understand the responsible use, the compliant use, how are people going to use it? Are they using it in the right way? Are they trained to know they can use in the right way? Is the tool working towards that? Will it help you kind of rewrite something inappropriate or draw images that are inappropriate? Does it have that right level of control to be able to ensure that? Are you sure you're aware of what compliant use looks like, you know, from things like the

EU AI data, data act or GDPR or even where in financial services and pharmaceuticals you have your own compliance needs. Are you meeting those and does the tool meet it natively as well? Next one is given you to find a strategy, working out how you actually deliver on that. What does it look like? Just coming up with a load of ideas and saying we should do this, we should do that.

doesn't actually do anything. So how do you deliver against that strategy? And I think linked to that, the final one is how do you measure that success? How do you say, this is what we're trying to achieve. Here is a definition that we have achieved that. Saying things like we have increased our productivity by 50%. That's lovely. How did you measure that? How do you define productivity? And I think especially for knowledge workers.

it's kind of tricky to say exactly what that productivity boost is. So you might not be looking pure numbers. You might be looking at a few indicators that suggest you have got to that point as well. So it's thinking about those different steps within there. And given that, how do you decide from all those different factors, from those different things that are going to be in your strategy? These are the things you need to think about. So what are the tools that you're going to bring to that? What are they?

The cool gadgets are cool ideas that you're going to bring forward. And for me, why is that copilot?

So one of the things and I think I've spoken about this on the show before, I know those who see me speak this year, I almost always use a similar slide. And for those who can't see this at the moment on the screen are three concentric circles slightly aligned to the bottom from there in slightly different colours. they're bigger, medium, smaller within there. And next to the big one is a

big sign that says agents for all. And this could be copilot for all. And so thinking about those different types of ways you use that copilot, use that agent. The first one are are those services that could be used by everyone. That's agent like M365 copilot or even just the the Microsoft copilot that comes with the web. I promised I wasn't going to go here. Whatever it is called today.

within there, the artist formally knows of Bing Chat Enterprise within there. These things are available to everyone and everyone can make use of those. But then we have that layer below that agents by role, we have specific roles within there, such as tools like the copilot for Viva Insights that's really geared towards those HR folks, some of the dynamics ones which are kind of aligned either with maybe finance and obviously copilot for finance there.

could be the service help. So you've got the dynamics ones there plus copilot for service, you've got copilot for sales. There are particular roles within an organization that become more suited towards those different ones. It's not for everyone and you wouldn't license it up for everyone. You'd license it for those who would need it. And that kind of fits those agents by role that you can build out. And then that final one, that specific

build out of an agent for a particular scenario. And I do intentionally here just say agent because there's not so much of an opportunity where a co-pilot is just for a particular scenario. I guess and why I'm hesitating slightly, you could argue that co-pilot for security fits that scenario. It's dealing with the

kind of security events and things like that, although I think that's more role than scenario. So you've got that kind of focus there is generally building something out to fit something that you need to do within your organization. And each of these different types can also be supported by looking at from a different perspective of are you creating this genitive AI capability to help your customer experience?

And that's not necessarily building it for customers, but the people like service, like sales, operations that are servicing the customer, that it can help them be more effective. Or are you looking at this to be about employee experience, where you're looking at it more around helping employees be more effective? And then that subset of the employees, the specialist within there. Again, this kind of comes back a little bit more to that.

copilot for security or even your kind of full code developers, your code first developers could be considered within that specialist view from there. And if you layer together some of the copilots, for those who watching the video can see this slide in there, you can see that you've got those different ones that I've kind of talked about primarily within here. You've got that copilot for service that fits that copilot by role.

but is focused on customer service. You've got things like, let's see, we've got Copilot and Fabric, which I've kind of got between the employee experience and specialists because you will get some benefits of like the Power BI, sorry, Copilot and Power BI that can be used by anyone and chat within there with a license, obviously. And then you've got the the kind of capabilities of building your data and working with your data that sits a little bit more in the specialist side of things there.

So do check out the video if you are just listening. I'm not going to try and talk about all these, but I do have that single view on there to think about those different copilots. Now, given those, I talked quite a lot about copilots there, but also touched on the idea of agents. And I think this fits a very important thing to consider is how do you align those tools to what your needs are? Because you can go to Microsoft

You can pay your money for a license or get one of the ones that are included, such as the Microsoft Copilot. And you could use that. And yes, there is a cost for adoption and change that I talked about, making sure people can technically use it. There's probably a little bit of turning things on that you need to do in tech evaluation and certainly some governance to sort out your digital core. But it's about enabling what's there. And that could meet all of your requirements.

But more likely you're to be looking at 80, 90 percent of the capability there will meet what you need. But sometimes you need a bit more. You need to be able to extend it beyond that. And that's where that ability to extend, co-pilot with agents that allows you to bring in your own systems, your own data, whether it's the third party ones like ServiceNow and Salesforce, whether it's bespoke systems within there, you can bring that alignment into

the copilot interfaces themselves. And that's not just M365 copilot that includes the others that come through from that world as well. So some of the dynamics ones that is certainly possible to to bring that capability forward from there and and make those things happen. So you can think about that extension of it. You can also look at your

business processes and automating some of the elements there. So bringing forward your workflows that take place from there and bringing generative AI into this. And when I first started thinking about this, was kind of how can you bring gen AI into Power Automate? We're now seeing those autonomous agents where it's not just about triggering an agent by typing in a chat bot. It's about an event happening where that's an email coming in, whether it's

you dropping a file into a SharePoint list, whether it's an update to your ERP or CRM systems, you can trigger those processes to kick off. And then finally, and this is one for Microsoft to really take on board, especially the M365 Copilot team, there are times that you want to build a completely custom standalone agent that isn't integrated into the different areas on there. So this could be the...

standalone agent. It could be something for your customers that you put out on the website, but it could be there are reasons that you want this not to be connected to M365 within there as well. And that's your custom agent that comes through from there as well. Now, I've started talking about copilots and talking about agents. I guess one of the questions that come up and we did touch a little on this earlier is what is an agent? So to my view, it's something that's autonomous or

possibly semi autonomous, it could still have, I think there is that importance, the human of the loop in the loop, or it could be it triggers one thing, but without generative AI trigger happens. And then there's some AI that works in there with it as well. So it is some element of automation to that and it will use AI and generally, if we talk about agents, we're probably looking mostly at generative AI, but not necessarily.

What it should do is enhance your business processes. Now, whether that's business, meaning your organization as a whole, that is absolutely some of those business processes, but that could be at a team or department level, or it could be your own processes as an individual to make you more efficient, more productive in the way that you work as well. So an agent is there to help with all those different things.

And it's a lovely slide from Microsoft that talks about the spectrum of agents. have agents that are focused on retrieval of grounding data, summarizing, bringing things together for you. You've then got ones that are more task focused, that will actually take action when asked and they can automate flows. And these ones are fantastic for repetitive users. But you also have these autonomous agents. This gets very interesting from here.

These will operate independently, but they will work together with each other. You can have these to orchestrate other agents. can learn. They can know when to bring in other agents as well and to escalate to humans at that point. So this is kind of where a lot of this capability, and again, we talked about this earlier, is starting to move to when we talk about agents.

These autonomous agents is what a lot of people were thinking about initially. The reality is what things can happen right now and not all there as well. So these are something that's starting to evolve and starting to bring some magic from there. Now, talking about that starting to evolve, one of my favorites kind of little squiggly lines is the hype cycle that Gartner has.

And it has that kind of initial innovation trigger, that nice new one. And the hype cycle is a way for God to kind of tell people where they perceive these different areas of technology are in terms of the reality around them. And that initial kind of growth that you see that peak up as it goes up on there is the expectations of what some of these tools can do. And that will hit a peak of inflated expectations, at which point

People are expecting it to be able to do more than it actually can. And at some point, they'll start trying to use them and go, I thought it was going to make a cup of tea for me and give me a nice chocolate hobnob while I'm sitting here. You mean an agent can't do that? No, it's digital only. least the ones we're talking about. I know there are some robot ones, but I'm pretty sure they can't make a good cup of tea and get you a chocolate hobnob. So the

The reality of that expectation starts to drop off that cliff towards that trough of disillusionment in there, where kind of what you expected it to do and what can actually do are different things and that will generally not bottom out to the nothing, but you will see that trough of disillusionment. But as it hits the bottom, you start to go, well, there is value in this and it moves towards that slope of enlightenment where you have the

the reality of what it can do and aligning those things of what it can do with what you with what you need, what can be aligned to your business objectives. And that really is that plateau of productivity where it levels out and goes, right, I know what I wanted. I know it's maybe not perfect, but it's getting me some real value and I'm liking this. Now, for those watching video, you would have seen this already and I hope it's large enough for you to notice.

But you will see from here that generative AI enabled virtual assistants are the ones heading closest on here towards the trough of disillusionment. And this was the 2024 hype cycle for generative AI supplied by Gartner. And they are kind of saying that, actually, the kind of reality of where people is, people are kind of seeing that these can't do everything and it's heading towards that trough of disillusionment. And you may think

This is the perfect time to go. point us investing any time on this. No point looking at this. But I disagree. This is your opportunity to prepare because you want to be able to take those through to that plateau of productivity to get the value of these. So go and find out what those challenges are. Go and sort out things for that and consider what it... Think about those considerations for an agent as well.

And I've got six things I want you to think about for an agent. There are the triggers. I talked earlier with Zoe, we said that the big evolution we're going to see this year is that move away from pure chatbots and seeing how generative AI fits in the day-to-day flow of work. And that means there's triggers. So that could be you looking at something. It could be a document coming in, an email coming in. I don't know what else. And that's what I'm looking forward to hearing.

is what people see this. What are the triggers for agents coming in and doing things from there? Because that will change and that will help bring things into the flow of work.

And alongside that, you need to think about the creators. So who is it that's going to create these agents? If we look back six, nine months, creating an agent was very much a pro-code capability using tools like autogen, semantic kernel. think crudo AI, I've heard a few things about on there. There's others, non-Microsoft ones, looking at that, but very much it was looking at Python, looking at C sharp JavaScript.

within there to create these capabilities. But then Copilot Studio came along with its agents, formerly called bots, formerly called copilots, formerly called PVAs, et cetera. I won't go too far back with that. But suddenly power users had that ability to create things. And yes, there may be some limitations to that. And maybe that it didn't allow you to index your documents to quite the degree you wanted to, but it gave you a really good start.

really powerful way of doing things. And you just needed to understand a bit about the power, power platform tools to be able to make use of that. But what we're seeing an evolution even further now is toward end user agents we've seen with SharePoint agents where you can go in, click a few, a few files or a few folders, up to 20, click on create an agent and you can start talking about the contents of those files and give that out to people.

And that is something that even your gran can do, probably, that can go in there and open these things up. Your granddad could be trained into this. There is that ease for everyone to be able to go and see some benefit. Possibly need a little bit of understanding of how to write a prompt for those, but it opens up to a lot more people. So the creators of those.

of who's creating that agents will evolve as well. And you need to plan ahead for these as well. You need to plan ahead even within that for those different creators to think about how do you keep human in the loop? How do you make sure that there is a boundary to what these things can do so you have control over what these agents do, that there are these pieces in there? And there's a certain element where there is the governance needed for this. There's the central capability. But there is also that focus on

individuals knowing when to do that. When do I want to get a notification and do something about it versus just letting my little agent carry on, it's very little way on there. So thinking about that human in the loop that you have the confidence and trust within for it to continue. Next thing I think about is generative capabilities. Do you actually need GenAI? I was presenting on this recently and there's a big picture of

RegEx is not AI. There are times that you just need something to match a few patterns or say if this, then that, if this, then that. That doesn't need generative AI from within there. That's just a plain flow, a workflow that can pick up from that. So making sure you're using that generative capabilities where you need it. And even if you are using the right level, so putting the right model within there that can do the right thing and

can use the right amount of power. So what I mean by that is that if you use the very latest model, there is a cost around that and there is a higher usage of power within those data centers, which will hit your sustainability. And you need to think about that as well to think about what is the cost, not just to organizations, but to the planet and things on there as well.

I was neatly challenged on this at ESPC. They were speaking a few weeks that I didn't talk about the sustainability impact of these. And I do genuinely think that's something we need to be talking around there. And I think as a simplest level is where you're using this AI, make sure you're using it effectively, because that will use more power and use things unnecessarily. So making sure you are bringing value to your organization and to the planet with the use of generative capability.

...

And thinking about the autonomy, how will this work on its own? What are the boundaries you need to set? This kind of goes hand in hand with human in a loop to a degree. But I think it's more specific to say, when do you want this to do things on its own versus when do you want to put that block within there? When do you want to put that stop saying, now that's enough within there as well. And then the final one when it comes to agents is scale. And

By this, I don't just mean scale in terms of, will this work for one user versus a hundred? And I do mean that to degree. You need to think about what that's going to use from a technical capability. You know, are using SaaS platform as a service that allows it to scale there. But more it's about what happens when one user uses it, when a hundred users, when a thousand use it.

Does it change the way it behaves within there? Will it work from there? And more importantly, when you have one agent versus 10 versus a thousand, is it obvious how these things connect together? Do they talk to each other effectively? Is it obvious that when you have five agents called Salesforce agent that maybe one's more finance focused, one's more CRM focused, one's more HR focused, et cetera, how do you have that understanding of what to use when?

that becomes really important as you scale these different things out from there as well. And also, one of the ones I picked when I was at ESPC, I had a massive screen behind me and I had this nice view of it, trying to talk about this kind of end user created agents and then making use of modules and different blocks. And they look like huge eye of Sauron. So try to have that in your head because in the middle is the end user.

who will use agents or maybe do a little bit of expanding your selecting documents. Then outside that, you've got those power users and they're going to do a little bit more and they'll have different types of knowledge they want to use. They may have different actions they can build into that, have different uses of prompts. And so you don't need those specialists I talk about, necessarily building full agents. They can build those capabilities. They can build those

connectors to those knowledge sources, they can build prompts that other people can use that you can make available to them. And they can bring that capability for this. So think about that scalability of how am going to reuse these things? How am I going to have different agents? But they're different in terms of the instructions they get, not the data they're using, not the actions they take within there. And that will allow the scalability of this because you can test those modules once and know it will work.

the different agents within there as well. you know, I've only just thought about this testing as well within that scale, you need to be able to test these different things, test how they will interact. And again, this fits with the autonomy, making sure you're comfortable with that. So how do you go about doing some of these things? Well, our lovely friend, Simon Owen talks a lot about this. Zoe, if you ever listen to this, we must get him on the show to come and

chat about this, but it's thinking about your center of excellence and your center for enablement. Your center of excellence is where you build that capability. You have people who will help govern your AI focus within there. You have all the different tools within that that will make sure it happens, that people know they can go there and get what they need out of it in terms of making requests for new ones. You've got that central

real speciality within there. Your center for enablement is how you enable people to use those agents to know what is possible. Your adoption change that says, did you know these things are out there? Did you know that capabilities within that? And also then if you're looking at those power users and those end users, how do they know to create those things? How have you enabled them to do things themselves? And that can go all the way to those pro code, code first developers who will

who will want accelerators. They don't want to build everything from scratch to learn from things you've done before. All fit within that centre for enablement. Celebrating success within there. And I'm not going to dive into this too much. Simon Owen has a fantastic blog. If you look up Simon Owen, Microsoft or Simon Owen Avanade, you'll find links to his blog post on there that he talks about. We'll put in the show notes as well in there. So thinking about these two things is something we will certainly come back

to as well. And these are key to not just building that strategy that I talked about way back in the beginning of this kind of chats within there, but also thinking about how you can actually make that strategy real for people, how you can enable people to make use of this, how you can build that capability through a center of excellence to actually deliver on that. Your COE and your CIFRE

those two things that enable that strategy to happen. You build that strategy to work out what it is you want to achieve. You build these to make sure that that strategy can happen together as well.

So no Zoe at this point. So I'm going to say the goodbye. Thank you very much for listening to this episode. Thank you all of you who've listened to Copilot Connection this year, whether it's been through the podcast, whether it's those who've joined a couple of our Copilot Fireside chats. If you go to copilotfireside.chats, you can sign up for those. We have the incredible and beautiful Donna Sarker coming up in January. So really looking forward to that.

And that is not recorded. So sorry if you miss it. You have missed it. But we have that on purpose to make sure that you can actually actually come forward and listen to that. And we can have an honest conversation with them. So sign up for that one. Bring your questions. And if you have questions for us, send them through social media. Contact us directly. We would love to hear from you more about that. What you've liked this year.

a night on there. We'd love to hear that as well. What should you do more of? What should you do less of? Yes, I have been doing a retrospective recently, so that is fresh in the brain for that. So we'd love to hear from you. We'd love you to tell other people if you have enjoyed these episodes, share it with them. Share it in your internal communities in Viva Engage. Share it within your teams, within your teams, teams, within your teams, teams, teams. I think I managed to get to four teams at one point when I was talking to someone in Microsoft, but

Go out and share and let people know about this. And, you know, as it's Christmas, I'd like you to take a break. But one little present, if you want to get us a present, is to go and give us a review. Go on to iTunes, go on to, I think you can do it in Spotify as well, various places where you're listening to your podcast. Give us a rating and just give us a lovely message there to say how much you've enjoyed the show as well. But then switch off, let your brain relax.

Have a fantastic holiday break. Maybe listen to Copilot Connection to let you relax on there. But otherwise we will see you and you'll hear from us in the new year. Have a great time and thank you very much. Bye bye.