Human Insight and AI Adoption

I'm going to set a little bit of context for this talk while Tom sets up. So about eight years ago, next month, I set up a professional services business. And around about the same time, I started studying as a psychotherapist. And those two worlds were very separate at that time.

And the quicker we're on this AI journey and tech adoption journey, the quicker it seems they're mashing together. together.

So the talk we're going to go through here, and we'll reference some of the good work we do with our customers as well, touches on the human element, which I'm really passionate about from a relevance and an adaption perspective, and then the more technical element around how we enable that, but also do that with awareness, so bringing it more into consciousness.

And I'm not going to cover all that in the remaining 13 minutes so I want to give space for Tom as well

and just just while we set up a little bit nod to this community I did see another Tom a Tom Mason present at this event which was absolutely fantastic and I thought you know what what great opportunity we've got within Leeds for one not to be traveling down to London all the bleeding time but to build something for ourselves as a network so I'd encourage anybody that wants to to get up and speak and share their knowledge or even just use it as a format to practice and engage with people come and do it because this is ours to shape and Nick's doing a fantastic job

around bringing people together that the community is only of good much like the AI tools actually the community is only as good as the things that we put into it so everybody's welcome and we can build something good up north in in Yorkshire yeah well the next speaker Andy is from Barnsley so he's got a visa he's got a visa to come into leeds today but yeah he'll put on his best posh

The Business Challenge: Creating More Conversations

bansley accent right so for this for this short presentation that we've got here the problem we were looking to solve um with some of the tech that we're going to show a little bit behind the curtain with essentially as crude as it is we weren't having enough conversations in the business

so we were pushing a lot of messaging out there but the tooling for that and the reason to get people's attention was lacking so really honestly we looked in the market and see well what are the people doing well all we want is a conversation and then the more conversations we were having

that also increased the effort to execute on that so it's not just the time in front of people it's following it up documenting it if i roll the if i roll my career back 15 maybe even pushing 20 years a lot of that was done manually taking notes down in meetings or drawing process flows on whiteboards which I still love and then sending that on for review we can do that a lot quicker

now but I still feel the human element that faces into the interactions with people is critical

The Opportunity Explorer Workflow

again you get out what you put in so this process we're going to step through here is our opportunity explorer which maybe sounds like a bit of a fancy way to say we have a conversation with somebody but how we approach that is we have a conversation either in the room with somebody or on a call and we transcribe it we take the transcripts we probably use facilitator in co -pilot more than anything and then tom's developed a prompt that sits behind that that that we've iterated now over a number of months that pulls out the relevant information.

The key is what I was talking about before is that human element. So it's done through human interaction to actually listening, understanding the problem, knowing the right questions to ask so the information in that transcript is rich enough to get a decent output.

And it's not just packaging it and sending it on blind through, but we just want it out there. We purposely build friction into that process because we review it we look at the output we make the changes and then we seek to talk through it with the customer as well so if nothing else they've had a lovely conversation with a lovely northerner and they've got an output that they can take away doing that with leveraging the AI tooling that

Tom has built with the prompts turns something that would take potentially days to curate and write down and package off into really minutes minutes that could stack up into almost maybe 45 minutes but it's quick which means we can respond quicker which means i can adapt the business quicker does that all make sense i've spoken a lot at you and i'm going to get some

Turning Conversations into Client Stories

water well tom taps through some other bits yes so as sean was mentioning so i've really i've really built a a skill so what what the skill looks to do is as sean mentioned using that transcript that we get from that and out that our call and then it just builds that story that we can tell to the client about the the the problem that or a specific problem in their business that they're looking to solve.

And we do that through measuring the friction, measuring it in a way that's the wheel way and wall, which I'll let Sean cover off a bit later. And then we score against five dimensions and fixed calculations.

So that's, I can't remember what the framework is, and I'm not going to try and say it, but it's alignment, capacity, literacy, something thing and then governance and then we we score that based on the will way and wall so as show mentioning that's the the friction whether we see it as a high risk a low risk or an intermediate risk and then obviously the the the next

steps after that is ident identifying the scope this the the the success and a a route forward so we do that based on and and what the skill does sorry is create a four slide four slide deck so that's the the the initial summary so that's everything on obviously everything on one screen then it digs deeper into the wheelway and wall so that the the framework that i just mentioned and then it goes on to the next step so how we can support that business

How the AI Skill Applies Context and Scoring

going forward that's all making sense so far yeah and where I'm talking about a skill we're giving it additional context so we're giving it the a system prompt going off what Terger was saying we're giving it the context initially so we're saying we want you to to act act as an analyst, to look into the transcript and sift through an hours call which you can

imagine there's a lot of conversations going on in that hours call and what the skill does is just sift through everything, it's essentially like a set of rules, sift through everything, and finds the specifics based on the products we offer so it's got that context as well

and then rank some of the risks based on the wheelway and wall the the friction as we call it the friction read and then moves into the the the success and scope and where we initially started was just the the one page summary the sort of exec summary page that i mentioned earlier

and now because of the amount of the the way it's evolved is because we're getting far far more information as we initially expected out of that our call we've now built that into a four slide deck so the the client can obviously get more more information and more value out of that conversation and this is all done using this using a co -work skill but as we know like with

With skills, it is portable, so we've tested using ChatGPT, we've tested using CoreWork, but the reason we use CoreWork is obviously it's integrated into the Microsoft ecosystem, and we use, as Sean mentioned, Facilitator to transcript the calls.

Guardrails, Uncertainty and Human Review

And I've put this slide on, obviously, because we're at a UK AI event, and we're just talking, like, what does AI do in this workflow? flow it as i've mentioned it reads through the whole conversation so that full hour transcription apply then applies that repeatable method of the will way and wall framework and the scoring rules on each and that's this is all defined in that quite as you can imagine it's quite a big skill

it keeps the uncertainty intact so in in this example so i've actually run a run an example transcript for this session, it says, does not turn Alex's unmeasured 30 % time -saving guess into a claimed benefit. Because he's made the assumption that it's got that 30 % time -saving, where it's not being tested. So what this skill does is calls that out and keeps it in that presentation.

And then the next step of that is it builds a usable first draft so the the human in the loop element is once you've run the initial the the initial skill it's pulled in that transcription it's it's sift through all the conversation it then out it then asks you the question of is this the correct context is this the correct is it is is this the correct way to frame x y and z because one thing we we have with with the first iteration is we

we had a client that obviously has been through in the in -hour call, he said certain things about certain colleagues, for example, and it's took that as literal, so it's included that in. So one of that, going back to the previous side, one of the evolutions was to just update the system context and tell it to not literally take everything that the client is saying as literal and and turn it down into more of more of a business professional language

so i'll let sean go into sort of the human review element yeah i mean the human review it didn't really feel like a decision including this we did it we did it naturally which i i'm predicting is going to get rarer because it's easier not to have it there it's easier to put something in and get it automated back out for the sake of speed but

But from our perspective, it's really important that we keep the human relevance in it for a couple of reasons. I think it's the right thing to do for a better quality output for the customer. But also from a business owner where I've got my own people engaged, I think it's the right thing to do to have their eyes on it, to keep them relevant and their skills up as well and to account for that.

so rather than just pass through as a an automated process which we probably could do maybe 99 % of the time there's still that human factor reading it checking it and then making contact back with another human but we're leveraging the technology in between and and as it goes out there on the

stage it prevents the as i've just mentioned it prevents the five scores back the labels the three what the three zones mean and what the the sort of context is based on the transcription

and obviously the the person that's running this is the one that's been on the call it's the one that that understands what how the conversation flowed so they'll they'll see they'll look at this and and be able to sift through it and tell that yes that's correct no this is not right i need to amend this and then you can just work through and it's just that human review cycle

Building Consistent, Branded Outputs

and one thing we've we've worked through is to create a template that we're happy with and the reason for creating the template was when we when we initially had the the first iteration we didn't have a template we just had the skill and it just defined all the brand colors the slides the general structure of how we want it to look

but obviously as you can tell well as you all probably know when you get it to create a powerpoint it's not always it's not always going to be the same and with our head of marketing the back of the room you can't really get it to just be to to create any sort of branded document it needs to be structured it needs to be the same moving forward and that's the reason we've really looked at design like creating a template that then the the skill sort of defines right this

This context goes here and as you can see this is the initial output summary so that's the first evolution that's still included and it goes through the opportunity statement, it has all the client context at the top, it gives you that initial will, way and while read and you can see at the bottom like agreed scope, success criteria and then you can see the next steps at the bottom.

From Assessment to Next Steps

so one of the things I've done for this call that this not the call this meeting is is I've generated an example transcript for a company called North Bank services and as you can see it's running the opportunity Explorer so this is the initial slide that they'd receive so you can see that opportunity explorers are sort of everything you've just seen on the previous slide and it it'll rank it based on the the wheelway and wall and you can see there you can see the level significant risk watch and manage and then it comes it gives you some agreed scope and success criteria and then offering that next step and that next step is based on the skill and it has

the context of our product backlog and then it obviously understands right the general context context of this conversation was all about AI. It's how AI can improve their business. And you can see that it's pulled that out there, the next step from the transcription.

And then moving on to the next slide, which is sort of the way it's evolved over time. It's got an assessment preview of what the next steps would look like. It's got that that rank in there, you can see that the will is watch and manage. You can see that literacy, infrastructure, governance, capacity and alignment. Or infrastructure, I forgot. Literacy, infrastructure, governance, capacity and alignment. So it gives you a score on each of them, whether it's a significant risk or watch and manage. And it gives you the rationale there. So it helps the client understand where the risks are in a after a simple one -hour conversation what the skill does is obviously build build all of

this from the transcription and then moving on to the final slide it gives you that sort of the structure of what what happens so it's the explore it's the assess it's the blueprint and deliver and this four slide four slide deck well three slide with all the content on realer um it just it's built all this from the transcription it understand we've given it the context we've given it the system the the system prompt so it understands how to run we're giving it the language we want it in we've and obviously we've given it the context of the transcription and then it's built all all of these and it gives them that sort of what we call as

the hook so we obviously the the end goal is to sell one of our services or another product to this to the prospect and it gives you that hook and it's pulled again all of this from the transcript it's understood all the content from the previous so the previous slides so the wheel way and wall and the friction read and then it's built into this hook and it knows what the obviously next steps are based on that transcription and yeah that's everything