Coco: You know, it’s like, uh, I remember when I was rolling out e-learning, I was one of the first people involved with e-learning. And people were saying:

What’s e learning?
How am I supposed to learn through a computer?
That’s just ridiculous, you know?

And this is like 20 years ago, right?

Or later, once e-learning was a necessary evil for many compulsory courses like health and safety they had this idea that e-learning was click, click, click, click, click. You know, obviously there’s many different types of e-learning. We know that now because, you know, it’s more rolled out or normal now. But at that time people really didn’t know what they were dealing with. And there was a lot of negative energy coming at e-learning at that time.

Then of course, there are a few, early adopters as is always the case, a few players that had really lousy stuff who was selling really cheaply and who were saturating the market with complete crap, basically.

Bad for the industry as a whole).  I’m sure this is the case with AI as well, that there are solutions out there that are super cheap or super accessible, or super trendy or super something, but actually there’s not a lot of meat there. There’s not a lot of substance behind it.

QUESTION “So how do I differentiate between a good and a bad tool?”

I mean, obviously it depends on what you need it for. But, you know, there is probably this quality issue. And at the moment there’s probably no control mechanism. Right?

Sarah: Yeah. I think you’ve just gotta really do your research in that solution stage. Right? You know, understanding getting really deep and what these solutions can do, have they solve similar problems to what you’re looking to use them for in the past.

 “Have they completed use cases like this in the past?”
“Have they worked with companies in your industry before?”

Financial services and health care for instance have a lot more complicated regulations and higher data privacy concerns.

You’ve got to make sure that the solution is robust enough to be able to handle those those kinds of regulations and able not only solve your use case, but work effectively within your organization.

Do as much research on these solutions, within your budget.

The resource requirements.
“Am I going to have to train my teams and how to use these solutions?”
“Do they offer training for my team.
“What does the implementation timeline look like?
“How long is it going to take, you know, for the use case to be solved.

TIME LINE
If it’s this going to take five years to implement? That’s not going to work. Right?
So understanding the timelines around implementing those solutions is critical.

Getting absolute clarity on all oy your very specific requirements is so key in that solution stage. …

Coco: But that brings me on to another question actually. Like, um, if I ask an AI consultant like yourself, you know, okay, how long do you think it will take to roll out? You know, you’re probably going to say, well, how long is a piece of string? But once I’ve got you to commit to a time scale.

Is it fairly realistic to actually hit that time scale or what can come up to slow you down?
Are they communication issues or are they different issues?
What could put a spanner in the works basically?

Sarah: Oh my gosh. Lots of things can go wrong, I will tell you that. But by getting very, very specific and by getting all those teams on board you can avoid most hurdles. So data issues can come up, right? Say you thought you had all this great data like, oh yeah, we’ve got it all organized and it’s supposed to be perfect for you to ingest. Like maybe it’s not. Or maybe that ingestion piece was not really well defined yet for instance

“Do you need this information in real time?”
“Do you need it in a batch?”
“Do you need it in a micro batch?”
“Do you (really) need it?

Those things have to be well defined up front to make sure those processes are working for what you need that solution to do, right? Do you need to engage the mobile team and the channel teams – that could be really complicated as well. If you don’t have those pieces in place, and have those teams ready to go – to implement those solutions, that could definitely cause you some issues on your timeline.

With every solution, it depends on the use case, depends on the solution, it depends on your timeline for a proof of concept. But ideally, these proof of concepts should be very, very quick, very small. Very narrowly focused. Like we are only using a very small scope of data. We’re only testing one rule or one communication one model. We’re not testing 17 models. Right? We’re going very, very small. So in theory, these proof of concepts could be could take a week, two weeks, or a month.

It just really it really depends on on the type of solution and having access to everything that you really need to do to make your proof of concept a success. In technology as a whole or agile, you also need people on the other side, you need people in the companies that you’re working with to do keep their part of the bargain.

You simply can’t do everything.
You’re reliant on them for data.
You’re relying on them for communication.
You’re relying on them for maybe educational access to the workshop rooms.

So it’s a bit of give and take, I guess.

Coco: Ok. So we’ve talked a little bit about the data and the implementation. Before we move on to talk about governance and delivery. Let me talk to you a bit specifically about, … I get a bit of a bee in my bonnet about knowledge management. So, um, you know, I, I, I’m in working for many agile teams. I work predominantly with agile teams as a corporate communication coach and business English trainer. What I’m noticing is that in a nearly all of the companies I’m with, they’ve got brilliant guys and women, who are getting to the age where they’re thinking of retiring in the next 3 or 4 years. And yes, they could be asked to stay on as a consultant, but they don’t have to stay on as a consultant. And all of that knowledge is still in their head. Right? They’ve had 20 years to get it out of their head, or maybe 30 even. But it’s still in there. You know, they still haven’t got this information out. And there isn’t always an opportunity to do a handover. Sometimes, you know, somebody might die or they go to another company or they get move to a competitor. So we need to think about knowledge management. I know it’s not sexy, but it is the legacy of the company.

Is there something that we can do with AI to help? Because that seems to be something that nobody wants to talk about. But the shit is going to hit the fan.

Sarah: Absolutely, absolutely. I mean, it makes perfect sense, right? You’ve got these, these, these men and women – their legacy employees. Right? You know, they’ve got a substantial amount of knowledge and information, that we need to have access to across the organization, across the specific teams.

There is actually a really great AI solution called Guide. And it is essentially an AI solution that helps record a daily process. So say, you know, I am a legacy employee. My job when I worked in Treasury would be to create the daily interest rates or create this really big rates report that went across the entire organization. So this was a really convoluted process I had to go to through, you know, use Bloomberg machine. I had to copy rights down from a the Treasury website. I had to put them all in Excel and reformat them. Right. It was a really convoluted process, and there were things I had to look out for whenever I was creating the rates to make sure, the trend lines are looking good and detect any anomalies. I mean, it’s a complicated process that I had to do every single day. And if I wasn’t there one day or sick, you know, who could do that job?

But there’s a solution called Guide and it essentially does a screen share with you. As you walk through your whole process just as you normally do, it documents the entire process.

Go to a website www….,
Put in your your username and password.
Copy and paste …

… it documents the entire process and puts them in a step by step format. That makes it really easy to understand and transfer information. Easier to consume by watching than to reading. If it’s a video, then you get written documentation.

Coco: You can have both?
Sarah: Yeah, it does a video recording and does a step by step transcription. So you can say, okay, go to this website, log in with this login. You know, copy and paste this information, plug it into Excel. So it really breaks it down in a very easy to use format. So there’s definitely tools like that that you can leverage to like get some of that knowledge sharing into one into one place that’s really easy for other teams to understand.

If someone is sick, or goes on leave or retires, then you’ve got access to that information or workflow. That way you can create that knowledge database where anyone who needs to can access and have specific support. If it’s someone on your team and you only want to grant access to a few people, you can control that. So those are some some ways that you can get some of that knowledge transfer information, you know, from those legacy employees into one system. So that really helps with your business continuity. So yeah, there’s lots of good systems like that that you can leverage.

Coco: Cool. Yeah, and that’s not an extra step, right? You could just do your job and then press play, you know, like kind of thing.
Sarah: Exactly.  

Coco: In terms of getting data in one place, if it’s all scattered, do you have some kind of AI readiness tool or questionnaire or something like that that you can give your clients just so that they know how close or how far away they are to really being able to implement AI strategically and not?

Sarah: Yes, absolutely. That’s something I’ve been working on creating for for a while now. Really breaking it down into those questions to get them ready for their AI journey.

Where are you on your AI journey based on whether your objectives are aligned to AI?
Do you have use cases defined?
What’s the state of your data?
Is it in a cloud system?
Is it in a centralized location?
Are you are you still working in different silos?

I’m creating that guide right now so that you are able to really understand really where you are on your journey so that I can make suggestions, like, you know, you really need to spend some more time on your data. Here’s some more resources I can give you that helps you getting your data into a better place to be able to take advantage of AI solutions.

I think the data is certainly what seems to be the crux of the problem for a lot of companies, it makes them kind of a little bit stuck – wondering where do I need to go? Like it’s it’s it’s a very complicated process for sure.

Coco: I’ll tell you what we’ll do is even though it’s not ready yet, I create a web page for every podcast episode. (you’re on it now, so it should show up below) So when you’re ready, just tell me and I’ll put it on the web page and then people will know how to access it. That way they can send you their email or whatever. Listeners who are keen to get their hands on that will be the first to know via my newsletter.

So then, um, I think we wanted to talk a little bit about data security and due diligence, and I’m not sure if that falls into governance. Or if that’s two separate things. You tell me.

Sarah: Yeah. We can we can certainly roll it into into governance. Corinne, as you know, governance is a big thing right? It’s all about being cautious about bias detection, right? You know, that’s a huge, huge thing, right?

It’s important to understand our AI models and checking if there is there any bias in the models. It usually starts with the data. Right? It starts with which data is being fed into the system.

You could have the best model in the world, but if you’re feeding it biased information you’re going to get a biased output, right? So I think when you think of governance, you mustn’t just look at the AI models themselves, you look at everything that’s being fed into the model. You get all of the data.

We set up a governance program at the bank that I worked in, and that was my first AI solution, it was kind of the guinea pig, right. We had never done anything like that before.

Coco: I bet you they were a little bit apprehensive, huh?

Sarah: Oh yeah. It was it was a tough one. Yeah.
But we but we had all hands on deck, right? So we had different perspectives from all across the organization. We had a legal team on board. We had

the Compliance team,
a Model Validation Team,
a Fair Lending Team and
Risk,
we also had
Communications,
Customer Experience and
Data Scientists and
Data Modelers.

So we had everyone coming together and having all eyes on this product and this solution so we could all look at which data elements were being used to feed into this model. Questions that came up included things like:

“Is there any issue with these data sets?”
“Could we maybe eliminate some of these data inputs that are going in.”
“Do we need all of these in here?”

So being very, very conservative about what data is going into these models and how it’s being used, including which decision each model is being used for was critical, right? For credit worthiness and so on you need to exercise some extra caution and be aware of any red flags. We need to make sure we’re not using this model to give someone a low credit worthiness score purely because of some biased information.

You’ve got to be able to detect these things. So we really locked it down. And we were very, very conservative with our approach. And, you know, all of our use cases were more like customer satisfaction, customer service.

But having that’s really what you need to set up a really robust governance organization though, or governance structure of having all hands on deck, all eyes on this, on this one specific focus. Right? And then, you know, with that comes the training with these teams. These legal teams, in compliance teams, the training in AI and getting familiar with how to read these models and how to interpret and understand the inputs and the outputs.

So those are extremely large exercises that we had to do. But then we got we got more comfortable with it, right? And had a regular routine. If we made any updates, if we wanted to introduce something new, we had everything in place – where we would come and say, okay, let’s all all agree, we’re all comfortable with this now. We all sign off and and we’re ready to push this into a production environment. And again, being extremely conservative throughout the entire process.

If we’re not pushing this to millions of customers, we just start with 5 or 10. So so that’s essentially so that the governance process that we created was robust enough.

Coco: Cool. Okay. So before we finish then I’d like to ask you something specific.
So, I’m in the process of planning a program for women. It’s a leadership program. It’s called EmpowerHer_EU. We’re thinking about different ways that we can, uh, you know, provide almost, like a simulation, to give these women an opportunity to prepare very, very precisely for a situation and get feedback. You know, and I’m thinking about things like using different tones of voice to kind of put them on edge, or maybe an accent that makes them, you know, realize, oh, my goodness, I have some kind of biases coming up here or something like that. Is that something I can do with AI?

Sarah: Oh, absolutely. Yes. I think a simulation environment is a really it’s a really unique way of using these tools. And, and there’s some great tools. There’s one called hey Jennie:AI you initially essentially create an AI avatar so you can have a male or female across any race. You know, you can change the tone of voice as well. You can have it sounding happy or have a childlike voice, a very happy youthful voice or something more formal like a news prompter or, someone sounding more aggressive if you wanted to. So that’d be a great way you can use those kind of avatars. You can give it the exact script you want them to read out too.

So if you want a more aggressive tone of voice in a simulation environment, you can do.

Coco: Yeah, exactly. Like some guy being a bit of a macho idiot.
How do you react to him or, you know, somebody who’s maybe sometimes, for example, when I was working in London, I had this girl and she was great, but she had a really timid voice and nobody really took her seriously. So, if we could have used something like this to help her, you know, to react to people when they really boomed at her, this would have been great. So I have to look into that. I might have to get back to you about that Sarah.

Oh, yeah. Those are those are some great tools for sure.

So I think we kind of coming to the end of our process. So we’ve talked about strategic objectives, use cases, solutions, data implementation and governance. So now all we are talking about is probably the last 20%, Right?

The delivery – tell us a bit about that and the communication that drives it or pulls it back.

Oh, yes. Delivery is a is a big moment, right? You’ve built up all of this momentum into finally going live with your AI solution. We’ve started with the proof of concept, starting very, very small. Right. So we’ve we’ve built up all this, we’ve got the data, we’ve done the implementation. So we’re going live now, right?

With the process of going live, make sure you’ve got all those teams involved, you know, if you’re, you know, setting up some data processes in production for a successful ‘go live’, but also making sure that the audience that you’re testing this out in production is on board, right? Again, keeping that very, very small. Like so not let’s not roll this thing out to millions of customers from the beginning. Right? That’s probably not a good thing to do.

Let’s maybe try internal teams first, right? Maybe you select ten people internally to try this out and make sure it works. So that takes us back to that proof of concept starting super super super small. And then you can start to build on top of that. You know that you’ll be making some improvements and then roll it out to 20 internal customers. Maybe you’re feeling good. So you roll it out to 100. Now you might want to transition to actual customers.

Coco: So you’re able to roll it out incrementally?
Sarah: Exactly. Using agile development. Being able to very incrementally build on top of these, because what you’re doing and what you’re able to learn is that feedback loop, right? So whenever you’ve got these models in a production environment, you can see how they really perform in production. So you’re seeing how our customers are interacting. When we send an email, are they opening the email or are they clicking on anything?

Or for those call centre agent conversations – are we getting a good reaction to those. So you’re able to test these out and grow very, very slowly and in a very effective manner in a production environment. So you’re able to measure, using your KPIs at the delivery stage, making sure the solution is doing exactly what you set out to do.

As you accomplish your goals, we are lowering costs, improving efficiency. So you’re really able to get that more accurate measure of success in a production environment. And then, from there you can continue to build and continue to fly with more use cases.

Do you want to expand on the one that we’ve got or select a new use case? Maybe we can tackle another one. So it’s like a process of constant experimentation and optimization.

Coco: Really.
Sarah: Absolutely. Yes, 100%. If you look at it as an experiment. Again, starting as conservatively and as small as possible, then we can build on top of that experiment, we try again and we build something new. So this is a really great way to get your feet wet and really start to build on top of these solutions so that you’re able to see that incremental benefit over time.

So that’s why the delivery stage should really be at the end. And then, once you see some success with that solution, go right back to the beginning and ask which use case to look into next. Maybe that solution that we have already onboarded can work for some other business challenges.

So you can deliver your solution successfully in a safe manner that’s not causing any harm to customers or anything like that. We’re communicating that across your entire business. We’re getting everyone comfortable with the process. So we’re getting to a point where we’re able to get those quick wins working it out internally and incrementally calmly, because we’ve been through the process before. In time, people know what to expect as we continue to iterate.

This is certainly the way that I’ve seen being the most successful.

Coco: Cool. Well, I think that’s been a great interview.

Thank you so much for your time today.
Sarah is available on LinkedIn.
There’s also a YouTube version today.
So you will soon be able to watch us both if you prefer.

Either I can answer any communication issues, Sarah can answer the, AI side of things. We might even jump on a Linkedin live or something like that.

Sarah: That’d be fun. Absolutely.
Coco: Yeah, I’ll have to do that. Okay. Right. Well, thank you so much. I look forward to seeing you folks very soon.

Until then, have a great week.

Be the very best communicator that you can be.

You’ve been listening to Experts! Speak English! with Corinne Wilhelm and Sarah Cornett.

Thank you so much. Goodbye.