Catherine Reed, Senior Marketing Director at SAP and PhD candidate in bias and AI for marketing, shares her expertise on using AI to enhance marketing strategies. Together with Dom Hawes, Catherine discusses the practical benefits of AI in customer scoring, its role in account-based marketing (ABM), and strategies to identify and manage bias in AI tools.

What you'll learn:

• How to make the most of AI-powered tools in marketing.
• Common pitfalls of bias in AI systems and how to mitigate them.
• Simple ways to start integrating AI into your marketing processes.
• The importance of collaboration between sales and marketing using technology.

Listen now to discover practical steps for improving marketing precision with AI.

About Catherine Reed

Catherine is a passionate and results-driven marketing leader with 12+ years of experience in the B2B software & technology industry. Her career spans across multiple disciplines in the marketing ecosystem and has led her to her current leadership role at SAP in Global Program Management.

She is currently completing a PhD in Artificial Intelligence at the University of Gloucestershire, with her research focusing on mitigating the propagation of bias through AI use in marketing. Her interests lie in a modern MarTech stack, next generation marketing technologies and how this translates into a better customer experience. She is inspired by possibility, driven by challenge...

Links 

Full show notes: Unicorny.co.uk

Watch the episode: https://youtu.be/h39dsfVnl9g

LinkedIn: Catherine Reed | Dom Hawes

Websites: SAP

Sponsor: Selbey Anderson

Other items referenced in this episode:

NotebookLM

Bishop, S. (2018). Anxiety, panic and self-optimization: Inequalities and the YouTube algorithm. Convergence

Luhang Sun, et al. (2024). Smiling women pitching down: auditing representational and presentational gender biases in image-generative AI. Journal of Computer-Mediated Communication

EU Artificial Intelligence Act

Focus all your attention on your scored contacts who want to talk to you rather than just trying to fire all your marketing out everywhere and hoping that it'll land well.

There's a very lesser known bias that is very strong. It's called Yorkshire bias.

Okay.

Which is I know what I like and I like what I know.

The more you prompt an AI, the better you'll get at it and the more it will learn from you what you're looking for. Because obviously every AI that you interact with on your account will be learning from you. It's a good thing.

Sometimes it could be a bad thing as well, depending on what you're prompting.

Hello, you, I'm Dom Hawes and this is the Unicorny Marketing Show. Welcome. Now today we're speaking to Catherine Reed.

She is from SAP and she is also currently studying a PhD in bias in digital marketing and AI specifically. So she's a really interesting person to talk to as technology slowly creeps into our daily lives.

Now, I'm not going to tell you too much more about her right now because she's, she's about to introduce herself. Let's go straight to the studio.

I'm Catherine Reed. I am a senior marketing director, SAP and I live in Windsor.

In Windsor?

Yeah, in Windsor.

Ooh, with the king.

Not with the king, but very close to his castle.

So you've worked in B2B now for over a decade?

Yes.

A lot's changed in that time. What do you think's changed for the better?

Definitely. You hear me talk about the Martech stack a lot today.

Okay.

The evolution of the Martech stack, you know, being able to have better technology for contact information, targeting, you know, your target accounts.

That evolution has been really exciting to see over the last decade, especially as, you know, some companies had that Martech stack capability already, but it's becoming obviously a lot broader now.

Yeah. So it's easier to get email addresses, but that also means it's easier for people to get our email addresses.

It's a bit of a double edged sword and one of the things I find.

So I've tried quite a few of the platforms of the Apollos or the cognizance and those people that are trying to provide you with direct contact, contact details. One of the challenges I have is I'm.

Because I'm a dinosaur, I quite like sending people things in the post and finding people's address these days is really hard.

Yes.

Are you using a mix or are you almost exclusively digital these days?

It depends. SAP, it still obviously does Does a mixture.

But because I'm in global marketing, we focus solely on digital and our field marketing colleagues will look after the account based marketing, the events, you know, the in person contact, which would include direct mail mailer. Sorry, but to a. Yeah, okay, now that makes sense.

I guess if you're global, it's going to take a very long time for things to get where they need to go, by the way. Also great for the local teams to be able to do the local activation stuff themselves.

Yes, definitely, definitely. They're a lot closer to the customers. They can adapt central marketing efforts to their needs.

They can plug and play customer references for their regions and market units and they're obviously a lot more closer to sales.

That's a good thing to have marketing close to sales. We like that. So one of the things I think we're going to talk about, we may come back to talk about that.

Specifically with reference to AI today is account scoring.

Yes.

Because it sits right at the heart of the ability to use technology to market better to our customers. Not everyone that's listening to this podcast is going to understand account scoring. Could you give me like a 101?

Talk to me about what account scoring is and how it works and why it's important.

Yeah, it's very important.

So every time an individual interacts with your company, and that can be through a level of interest or different pieces of content, perhaps they attend a webinar, they're given a score based on what they're interacting with.

If they're on your target account list and you can cross reference and match that they're on your target list, they can be given a higher score as well. It just means that you can be automatically scoring those customers until they hit a threshold.

And once they've hit a threshold, you know that they are truly engaged.

They're actually consuming your content and your information and you can start to follow up on them to directly focus all your attention on your scored contacts who want to talk to you rather than, you know, just trying to fire all your marketing out everywhere and hoping that it'll land.

It's an effectiveness tool that means that you are going to put more effort into people that have a higher score on the basis they're more interested in you already.

Yes. Yeah.

Okay. And how do you build, how do you weight or build scores for each of the activities?

So you have, you can have different, different scores based on the different type of content. So if someone's just reading an ebook, it have a lower score than if someone attends a Webcast, for example, different product groups.

You could have different scores for different product groups.

You know, if you're, if you've got one or two products that are your kind of winning products for the year and they have to succeed, you know, you would score them slightly higher than maybe legacy products that you're hoping will just, you know, start to. Start to fade away.

And so account scoring sitting right at the heart of pretty much every digital marketing machine, because that's how you prioritize who you're going to be selling what to.

Yes. Yeah, it's, you know, you're obviously, if you're doing your marketing correctly, you're targeting your right customers.

You are targeting those companies and job titles that are in your target list. But obviously not all of them are going to be ready to actually interact with your marketing.

But the ones that are interacting with your marketing are the ones that are going to start getting scores based on that.

They're looking at different pieces of content, and then they're the ones that you're going to want to begin to interact with once they've hit a scoring threshold.

Can people's scores go down as well as up? Like, if they start disengaging with things.

It'S up to the company. They can set how long, but I believe it's about six months.

After six months, if there's been no interactions, then you can start targeting them with kind of like a recycle marketing motion where you try and reengage them. Your messaging obviously changes.

I was thinking with context of the 95.5rule that everyone's talking about. You know, any 5% of your market, any one given time is in market for 95% of the time they're not in market. You're still communicating with them.

You're still. That your brand marketing is still working. So they're still building a score.

Yeah, but I'm assuming that the moment they come into market, there's a change in behavior. Will, like the rate of change of score be measured? Like, do you notice a massive uptick in someone's account score when they come in market?

So I'm not sure I understand what you mean.

That's the joy of being me. So I'm going to rephrase that, actually, because I said it in such a clumsy way.

I hope this doesn't get edited out, actually, because I think this is an important point.

We know that only a small number of people at any one time are actually actively looking to buy something, but we know that they do all their information gathering before they make the decision to enter market.

Yeah.

While they're in that bucket of 95% people, it's really important that they're consuming your content and that you're scoring them. Now my point was, and it may, it may, maybe it doesn't work like this, I don't know because I don't, I don't sit where you sit day to day.

But I would like to think that when someone actively comes in market and starts looking for things to buy that their engagement score would rock it. It's kind of sense that it'll be maybe plateauing and then suddenly it's going to spike up.

Precisely.

That was what I was asking is like how do you identify, can you identify when someone's kind of, from the data, when someone's not in market or whether they are in market.

That's a great point actually. So you would look at a customer journey from awareness obviously all the way through to consideration purchase and then post sale.

So those people that are 95% just having a browse, they sit firmly in interactions right at the top of the funnel and they're going to be consuming different types of content. They're going to be consuming like early awareness, articles, blogs, things like that, organic, organic, social.

And once they start actively viewing and interacting with your content, obviously their content needs are going to change. So they kind of, they're looking for more thought, leadership, knowledge.

Maybe they're looking at customer references because they want to know, you know, what customers have succeeded. Then they start to request demos, they maybe want to do a free trial and they attend webcasts. All of that have a lot higher scores.

And because you, you know, most of the time that consideration content is also gated, it's on your website. You start gathering cookies on them so you can start attributing a score.

But the majority that sits right at the top and the beginning of the customer journey, you can't really score because it's just organic and easily, you know, consumed.

So the thing I like about score also, I mean, you know, we're talking about, you know, the customer journey and we know that, look, it's not a linear journey. We know people hop about on them but you know, if you meet, if you average it out, there's a journey that you can see.

But by scoring this it allows you like, you don't have to have them in buckets. It's all about the score. Right. Presumably they're self choosing the content they're consuming.

Yeah, you do get some funneling.

So you know, once they're starting to interact with your content, the Martech tool, whichever, whichever one is being, is being used to score, they will start putting like, product codes against it so you can see which products the customer's actually looking for. And then from that especially Nurture email, they will kind of funnel them down a journey on based on the software that they're looking at.

Wow. It's no wonder sales and marketing are besties.

Yes.

Like that is. That's great information to be able to.

Give a sales team precisely so much more information than you ever, ever would have received before.

You've got a whole journey in, you know, in whichever tool it is, whether it's, you know, marketo or cr, you can then look and see exactly what content they've looked at, interacted with what they've attended. You know, whether they directly contacted us as a company or whether they got to a score a score and were contacted by us as a company.

You can give all of that information to the salesperson.

I've joked about, you know, sales and marketing investors in many companies, of course they're not.

Yeah.

But that kind of collaboration, that affinity between sales and marketing is really important, isn't it?

It's just so important. You have to have the same charter and goal.

You know, you need to know what you're selling and who you're selling to, and the core message and both sales and marketing have to be aligned on that from the very, very beginning. And all your planning for the year ahead should be based on that collaboration together.

Because if you don't, you're going to give mixed messages to your customers.

You're going to have a fractured customer experience because they're going to be getting mixed, mixed messages and you're going to waste money on areas that are conflicting.

But if there are marketers out there, or indeed salespeople listening to this podcast who want a close relationship, actually the account scoring we've been talking about might be that common ground they can use to get a bit closer.

Oh, 100%. Yeah.

You can go armed to your planning discussion or your meet with sales or with your sales team and show them you think you know your customer inside and out. Sure, you do. You spend a lot of time with them. But did you know that they've been interacting with all of this?

Here's the stuff they like.

Yeah, here's the stuff they like. You thought that they are interacting with this certain product.

Well, actually they are, but they're also interested in a secondary product and through, you know, arming Your sales team with a lot more contact and client knowledge, you'll gain their trust.

When we met to discuss this show online, you said something that hit me immediately. You said, many businesses don't know as much about their customers as they think they do. Can you elaborate on that for me?

What I meant by that was that not allowing for other buyers in accounts.

So some companies will have the rule in place that if there is a deal going through with a customer, marketing are not allowed to touch that customer. So it could be that you're having a deal with, I know, the analytics director and the analytics in that company.

But if you're trying to sell, you know, complimentary HR software, for example, you're not allowed to touch that customer. Even though the analytics team would have. No, you're trying to hit the, you know, HR director.

The analytics team would have no idea that you're in contact with the HR director. But there's almost like a blanket ban on talking to them. And being able to use AI in your martech stack can actually help you stop that suppression.

You can put a bit more rules in place.

Well, it's almost as though you knew where I was going to take this show because you've very naturally helped me segue into your specialist subject, which is artificial intelligence, which of course is being increasingly used in marketing, either built into the tools that we use or there's a plethora of SaaS AI out there at the moment. Let's start kind of broad before we narrow down.

How do you see AI fitting into technology, particularly for the marketing ecosystem, just really simplifying it.

If we have a look at it by traditional and then generative AI. So traditional AI obviously, or rule based, algorithm predictive, you can have that within your martech stack.

You know, whatever tools you're using, those companies will already be including AI within them. Like Microsoft Teams. They now got Copilot, which pretty much removes any type of admin that you ever used to have to do, which is fantastic.

And then generative of AI, obviously, creation of text, images, videos, even audio as well.

Yeah, I've been playing with some really wild stuff. I mean, I claim my voice, which I use on the podcast.

Sometimes you need to do a podcast all just through AI, like generating your voice and see if anyone notices.

Hang on, are you saying I'm a robot? Where do we get this one from? Nicola, It's a good point. I did actually go to Google Notebook.

Google Notebook, the language model has got a function on the movie where you can upload a bunch of training material. Press a button and it will generate a conversation for you. So I uploaded the previous editions of Unicorny Marketing show.

Press the button and we have a 15 minute podcast with two random voices. But it was a good conversation, actually.

It was interesting when you listen to it to start with because it's completely generated and they've built all the human kind of frailties and foibles and the ums and ahs. It's all built into the algorithm.

Yep.

But when you get about five minutes in, you kind of, if you are a kind of pattern recognizer, you start to see what they're doing is basically they've taken a series of shows. It's a bit like a listicle. It's a list of things.

They've got one item per show that they discuss and then they create a bit of randomness about that one item and then they move on to the next one. So it's not actually a very interesting show.

Okay.

It sounds really human and really plausible.

Yep.

But it's just not very interesting.

No interest. Doesn't have that human nuance to bring out the comedy or anything in it.

Exactly.

But just impressive for what it can.

And it's not a fail. Right. Obviously, like me in front of the camera, it's not going to fail like that. So, anyway, it's interesting.

I thought we were talking about customer scoring and customer interactions a minute ago. How does it work and why is it so effective?

It really is AI that has enhanced that scoring capability.

Oh, I see. Okay. So the scoring itself is done by. So AI is working out how much. How to score each piece of content you put in.

You input the rules into it to say, this type of content has this score, this type Of. Of this other piece of content has this score. If they're on our target account list, add an extra, I don't know, five onto their score.

But it's the AI that makes sure that's applied and make sure that that is, you know, it just enhances the whole of the journey, rather than people, rather than marketers having to go in and manually score.

Presumably. Also, you could look at Accounts Won and Accounts Lost. So you could look at those who worked and those who didn't.

You know what their interaction history is like.

Yes.

So presumably you could get AI to amend and adjust how you score different types of content.

Yes.

To make your predictions more accurate as you go.

100. Yep. Yep. You could. You could ask it to. To do that.

Yeah.

So, I mean, we had a unit that was working with a Financial services company, they wanted to, they wanted to score their customer interactions, but they didn't have any data upon which to base that to start with.

Yep.

So rather than do something that was arbitrary, we were trying to examine all of their previous case histories to try and give them at least some kind of logical basis to start with.

Because like the problem of being arbitrary is I always say that, you know, the, there's one thing worse than having no data and that's having bad data because you make decisions thinking it's based on fact and it's not, it's based on errors. What about with more traditional marketing approaches? I'm thinking about things like account based marketing here.

How does AI help in those scenarios?

Yeah, so it just makes it all a lot more efficient. A lot more personalization can be done.

Your content can be personalized, obviously for those accounts that you're targeting, you can do additional retargeting of those accounts as well. It just makes everything a bit more efficient.

And then you can, as we were just discussing, I suppose you can use AI to do predictive modeling to see what content those accounts are actually interacting with and maybe predicting the trends and where your, where your marketing efforts are best placed.

Now your special area, which I'm going to come on to in a minute is ethics and bias. We're going to come back to that in a minute because I think it's really interesting.

But one of the things I'm worried about at the moment is that some of our listeners who are not very technologically advanced might feel like the whole world of AI is completely inaccessible. Not something they can get involved in. But actually I don't think that is the case.

Could you, you give advice maybe to those people who feel uncomfortable about using the technology, about how they might start to get some tools into their activities?

Yeah, definitely. So I suppose with any revolutionary technology, which AI is, it is natural to be nervous about using it, especially, you know, for the first time.

But I would encourage every marketeer to start embracing it and using it. There's a few kind of core ways that you can use it, obviously through the Martech stack that you have available to you.

So if you're using project management tool like Asana, they have AI prompting reminders, everything you know, built in within that they can reschedule your calendar.

If you will need to put in a strategy, you can ask them to actually say, I need to deliver this event for by this day, what are the steps I need to start over the next eight weeks to achieve it or break it down for you. It's not going to be 100% perfect, but it's a great way of relieving some of your time to start, you know, you know, to start freeing up.

Yeah. Productivity is a great place to start, isn't it?

Little, yes.

Little sort of small, small impact if it goes wrong. But get confident with the technology.

Yeah.

And then start pushing it a little bit more maybe.

Yeah.

Learning how to prompt is a skill, you know, and the more you prompt an AI, the better you'll get at it and the more it will learn from you what you're looking for. Because obviously every AI that you inter with on your account will be learning from you, which is a good thing.

Sometimes it could be a bad thing as well, depending on what you're prompting.

Is that it?

Yeah, skill. But yeah, you've got Microsoft Teams copilot. That's absolutely fantastic. You were saying that you use your own software, that you've.

Yeah, well, custom GPT is part of ChatGpts. Yeah. So we've been building and deploying custom very, very, very, very application specific GPTs which, which are great.

If someone can't like prompting, you say people need to learn how to prompt. I know you can ask the GPT itself to teach you to prompt. Yes, but that's a little bit meta for me. How do people learn how to prompt?

Obviously you can have a, have a quick Google like what is. How do I learn how to prompt? How to prompt in an effective way. But always doing is better than reading about it, isn't it? And just starting.

I know when looking back through my prompting history, I'm like, whoa, what, why, why did I ask that? Obviously it's never going to understand broad open questions.

You have to be very specific and you know, anything that is, that comes back to you is not quite perfect. You're like, well, why would they answer like that?

And then you have to go back and look at your prompt and think, ah, well, yes, if I, I need to ask in a slightly different way. And then just as you practice, you start to retain, you know, the, the knowledge of how to prompt from better.

And the investment is definitely worth the time, I think. I mean, I know a lot of people who haven't even touched it, they're not going to do it because they don't think it's worth the time.

But, but actually when you do get to a level of proficiency in it, and I won't say by any means I'm an expert, but it's open on my desktop every Day, all day. And there's virtually nothing I do now that I don't think can it help me do it faster, quicker, cheaper.

The latest releases are so, so much more superior than, you know, like when ChatGPT3 came out and everyone started trying using it.

Google Bard, they were not perfect, you know, they repeated themselves, they would give false information sometimes it was just something that happened.

But the more people are using them and they're learning off of, you know, the public using them and they are obviously enhancing the data and the coding that goes into these generative AI tools. I mean the latest release of Chat GPT, it's just incredible.

You can load in, only do this if you're allowed to obviously or use your enterprise level generative AI tool. But you can put in whole email trails and say can you summarize this? And then craft a reply? And it is pretty much perfect. Now the reply.

Yeah, I have to say it's been. I was let down by yesterday.

I made a LinkedIn post actually just yesterday about how cross I was that my new best friend and let me down because I wanted to run a pace and it was basically shit. I had to rewrite the whole thing and then I got over emotional about it turned into a bit of a rant.

It was, it was to Leo Burgers at where you were sitting, he's a product marketer. We had a really good conversation. I was trying to, trying to write something for him and normally this. I've trained a GPT to write posts for me.

Yeah.

And I normally just have to edit them a little bit. This time it was absolute garbage. Absolutely.

Because the problem with GPT that I have, and I'm looking at alternative language models at the moment is because it's training data set is so wide, it's basically the whole of the Internet and most people are muppets, most of the training material is muppet like and therefore it's got some very strange beliefs about things. Obviously most people aren't muppets, I just said that for a fact. But, but it has some very strange beliefs.

You get polar opposites on the Internet, don't you? And it's consuming that. So yeah, there is going to be, there is going to be that.

Which brings us on to bias because we're talking about, you know, training data and things like that. And this is your specialist subject because you are indeed in the middle of studying for a PhD.

I am, yes. At the University of Gloucester. Yeah.

So my thesis is on exploring the bias in AI, but the actual usage and prompting of it by marketers and how we can limit it because you're not going to get rid of it. But how do we, how do we actually, you know, put things in place to, to manage it?

Should we say how does that bias manifest itself as a marketer? How do you notice or what might be the issues if there is bias in the system?

There's three ways that bias can obviously get within AI and that's both traditional, generative. Any other AIs that you have, that's in the data, you've got data bias.

So by the large language, sorry, models that you get, you've got the coding bias, the person that's actually coded the AI put in their rules and then you've also got the prompting. There's quite a lot of research already in place for number one and number two.

So the actual data and the coding, that's well researched, that's well established. There is bias within that. There is very little around the prompting.

What risks do you think there are to marketers if they're using AI and they're not aware of the bias in it at any of those three levels?

Yeah.

So if we don't even think about the prompting and you're just sitting down to use the AI, you might not be aware that it already comes with a lot of bias or in it.

There was some great research done on YouTube and obviously as marketeers we use YouTube a lot and obviously Unicorn Eve podcast goes on YouTube and the YouTube algorithm actually penalizes content creators that don't make gender specific content. So if you're a female content creator and you make what is considered male content, the algorithm will penalize you.

Really?

And the opposite way around. Yes. Correct. Yeah. I can send you the research.

That's barking.

Yeah, yeah, barking. So you will be penalized if you don't make gender specific content. So you wouldn't even know that as a marketeer.

No, you wouldn't. There's no, I have no, no clue about that.

And then there's also well documented in Google images.

So you've, what you've got in Google images is you've got, if you're looking in what is considered predominantly female roles, the images will actually show women with down pitched head in a submissive way. Whereas if you are looking at a male dominated industry, you obviously have strong male characteristics within those images.

So if you're using an image generator or something like that, and part of its large language models might come from Google, obviously not all of it, but some of it would do you could be getting biased imagery coming through.

That's extraordinary.

Yeah.

Because Google's got really well publicized guardrails.

Yes.

Google, normally you would expect to go completely in the opposite direction.

Yep.

To find that there's biased content sitting out there. That's amazing. Presumably they've done something to correct that when this goes live.

Yeah.

If anyone goes to try and replicate that and finds, you know, strong, upward looking, inspirational images of women, hopefully they know that it's been solved.

Yes. Yeah. So obviously as things get publicized or research is published about it, those, those things are addressed.

But you, you know, as some things are addressed, obviously other, other areas do open up. You might have obviously every time the language models get upgraded, more data is inputted. Where's the source of that data coming from?

I think that's one of the challenges. We just don't know.

Yep. And there's no global law to regulate it. You just have to be as good a human as you can be.

Think when I'm prompting this, make sure that I'm trying not to put my own biases in, get it double checked and just use it as conscientiously as you can do.

So data, I mean data we've known for a long time because some of those horrific studies that came out and found out that it was more expensive for women to do something based on the fact that they were women, which is the data set had been white men, all that kind of stuff. I thought Google Gemini showed us very clearly how coders can introduce bias into systems, which I think they've resolved with a. Resolve.

Now, prompting I hadn't thought of necessarily because you only know what you know and the prompting is the. Everything else is an informed bit. Right.

So the people who are helping provide the training, data expert, the people who are doing the coding are the experts. The amateurs are us. Like they're sitting in the chat prompt. I hadn't even thought about how we actually build bias into our own bloody.

That's quite dangerous. Especially if you're like a self confirming person like me.

If you are. Yeah.

And you're not, you're not willing, willing to ask someone, can you check over what I've just generated or would you check over the rules that I've just inputted into this piece of software? It's especially true what I'm seeing in my research so far is. It's especially true if you think, you know, I know my customer.

They're going to be this age, they're going to be this gender, they're going to Be here. And this is what they're going to be interested about. Without actually looking at the data, but also reviewing those outliers within the data.

You know, if you're getting pockets of customer characteristics that aren't what you, you think is going to be your customer, you can start doing something called hyper personalization, where you're only focusing on what you think and the people you think are correct. And there is eu. We got the EU AI Act. That is the only legally binding kind of law on AI in the world.

And so if that's, you know, it's good for Europe, you have to, you know, your, your company has to adhere to some AI rules, but that doesn't mean that it's the same for the rest of the world. And so you've got different rules, you got conflicting advice.

It's very hard to, to manage that when you're just an individual thinking, how do I use this?

Does that put the EU at a disadvantage, presumably, if they're having their development throttled, or do you think it's an advantage?

I think it's an advantage in the long run, just because while it does stop the creativity and the output, shall we say, actually it does limit the output to some extent. I think being able to just consistently do the output without any real review of the ethical moral guidelines behind it is not good.

It's not good for humanity. Whenever anyone does anything like that.

No, because somewhere in the middle of Palo Alto, Dr. Evil is planning to overthrow humanity. Luckily, the Belgians are going to rescue us because they made a law. I think that's the way it works.

What about the role marketers play themselves? You talked about.

You know, Ritson always says in his course, the first thing he says is, you are not your customers, you know, and you do not know your customer. But instinctively, many people think, of course, as you've already said, that they do know who their customer is and they're.

And they're actually looking for data to confirm what they already know rather than look for things that are new.

But I think comes at things from maybe a slightly different approach, like so, because it's the AI that will hopefully be interpreting the data for us, not us interpreting the data.

Yeah.

Marketers who are determined to perpetuate bias, presumably can manipulate AI, can they?

Yeah, you could do if you think, oh, no, this is a very specific gender, an age that I have to target. Yeah, you can overwrite it.

But that's where my research that I'm currently looking at looks like we've got a Good framework of people, process and infrastructure to help limit that being able to override. So you have the process obviously, which is, you know, any, any internal business process of what you should be doing to use the AI correctly.

And that will be an output of my research people education. So making sure that there's educational material available as to why maybe you know how to use it correctly and then infrastructure.

And that's your Martech stack. You know, your martech stack should flag if you are trying to put in rules or manipulate the data to your own end.

I mean that sounds like a leadership responsibility to make sure the right frameworks are in place. It doesn't seem there's a lot of guidance out there at the moment for people who are trying to do this.

I mean, I know that many companies have backed land use of the OpenAI tools because they're concerned they're not secure.

And also I think as you rightly raised, that there is no consistent legislation for AI in the same way as there is GDPR is not perfect, but at least everyone understands it now and there are similarish things going on in the States. Whereas with AI there's nothing.

If you're in a marketing department, if you're a marketer and you want to embrace the kind of tools that you've been talking about, how do you make sure you're not falling foul of kind of any company or corporate guideline or that you're not laying your company open to risk.

The only thing that you really can do is make sure that you have the time to have it reviewed by someone, whether that's someone within your branding team. If you've created something generative that needs to be on brand.

Obviously if you have done any kind of traditional AI routing rules, you've coded something that way you should check, you know, with like your analytics leaders and they have a duty of care to make sure that what you've done, you know, adheres to, you know, to that company's rules.

But you should at the moment have your anything generative or traditional AI that you have created that's going to go to a customer, you really should have it double checked.

Double checked, yeah. So this is what we do in agency. Four eyes, you call it always four, always four individualized.

I don't suppose that belongs to two people, could be many more. But you always got to have more than one person checking content before it leaves the building.

The other thing I think is quite useful, which we, which I've been trying to do here with certainly with my team is if you do use an LLM to prompt to help you generate content, you put a copy of the prompt so everyone can see what created the content itself.

Yeah, that's a great idea.

Which is. Yeah.

So partly, I think it helps explain the results you've got, but also we've started now building a big list of prompts that we've worked with that work really well for us.

That's fantastic.

So it's like a little template that everyone can use. That's been amazing. So when are you finishing your PhD?

Oh, it depends on how quickly I can get it done. Yeah. So either towards the end of next year or the year after.

And that's going to be published, presumably?

That will be published, yeah. I've got a paper coming out in December for initial findings and then.

Yeah, and then it will be published with a recommendation of Framework for companies within. You know, any marketing department can take it, big or small.

That's fantastic.

Yeah.

Well, what we're going to do, we always do. Assuming it's okay with you, we'll put a link to your LinkedIn profile on the show notes.

But is there any other way people, if they wanted to get in touch with you, could or should we just shield. Should we shield you through LinkedIn?

Yeah, LinkedIn sounds good. Yeah. Through LinkedIn.

Don't give out your email address. It's terrible. Some people want their email addresses given out. I can never understand why.

Oh, goodness.

Because I get enough spam already. So we'll put your LinkedIn details on there. Catherine, thank you very much.

It's been a really interesting discussion and as you get closer to publishing, I'd love you to come back and tell us about any new findings you have or how this thing is taking shape.

I would be honored. Yeah, it's really exciting. There's three crucial parts to it.

It's the Martech stack, the customer journey, and the interactions between the different stages of the customer journey and the Martech stack. And then obviously the people, process and infrastructure I'm reviewing. I've got 112 interactions that have all been mapped.

We know how the AI, how the different types of AI are interacting across those interactions. And then we're not delineating by the different types of bias because there are so many of them.

We are just through the literature, we can confidently go into the research gathering knowing that there is bias within it and just accepting that that is what it is.

I was just wondering, is there ever a time when bias is helpful? I mean, I guess it may be sometimes or at least being able to know it's there maybe is helpful rather than the bias itself being helpful.

Like as a human being, like recency, bias is really helpful. You know, the things that you've just done being more memorable than the things that you haven't just done probably helps you shortcut quite a lot.

I'm not. But given that AI is a facsimile of the way that we think, I just wonder whether there are any.

I mean, maybe this is a whole different podcast, but I just wonder whether there are any of those sorts of categories of bias actually that are helpful to AI.

Yeah, there was a great lecture that I attended by Someone called Dr. Foster and he was saying that bias is in everything you do, from who your friends are, what you want to eat when you wake up in the morning.

So, yes, in some ways bias is good because I don't like that for breakfast. I'm actually going to choose something that I do like. So that's, I mean, that's a really simplistic way of looking at it.

But yeah, bias can help keep you safe in situations that you might be unfamiliar with or in places that you're not familiar with. You know, it can help you keep safe. Sometimes bias has negative connotations and rightfully it should do.

Because when the bias is so polarized, you know, and it starts to marginalize whole swathes of our society, then yes, it's wrong.

I mean, there's a very lesser known bias that is very strong. It's called Yorkshire bias, which is I know what I like and I like what I know.

I'll include that in my research somewhere.

And I've just been cancelled in Yorkshire. I'm joking. I love Yorkshire. It's a very beautiful county.

It certainly is.

It is. Catherine, thank you very much indeed. We could NASA for hours, but unfortunately we run out of time. Thanks so much for coming on to the show.

Oh, it was a pleasure. Thank you so much for having me. Thank you.

Well, there you have it. I want to say a huge thank you to Katherine Reed for coming and spending time with us in the studio. A fascinating story.

This is something that we are all going to need to get our heads around as AI slowly starts taking over our lives. Anyway, thank you for Catherine and thank you, by the way, to you too.

We know that you have a choice of what you watch or listen and that your time is precious and we grant greatly appreciate you spending it time with us. Now, if you enjoyed today's show, I would love it. Please, if you'd give us a thumbs up, a like or maybe even subscribe.

That way you know you're never going to miss another unicorny marketing show.

But also, I'd be really grateful if you do like the show, if you'd recommend it to a friend or colleague, because word of mouth is the most successful way shows like ours grow. That's all we've got time for today. Thanks you so much for joining us. Till next time.