A hammer for every nail: Exploration of AI at Cannes

[00:01] Recorded live in Cannes at the can Lions international festival of creativity it’s the MM+M Podcast in partnership with IQVIA, and CMI Media Group told these tools. [00:12] Are only as good [00:14] As the data feed into them. It’s going to this idea of you know garbage in garbage out. You knew what the prompt engineer was 18 months ago. [00:24] 18 months of experience there is nobody with more than 18 months of experience. [00:29] Hi everyone and welcome to the MM+M Podcast this is a special edition coming to you from Cannes in the South of France we’re here for the Lions Festival and my guests here today are from IQVIA, and the CMI Media Group and we’ll be talking about an exploration of AI at camp believe it or not. AI is a very hot Topic here this year. [00:49] My guests from cmi is Oleg Korenfeld. [00:53] The technology officer for the CMI Media Group and Compas [00:57] and from IQVIA, Edith Hodkinson head of digital strategy for IQVIA digital engagement welcome. [01:03] Thank you. Thank you. [01:05] Okay, so we’re going to talk about AI the hottest topic in the world. [01:09] And I think the first thing. [01:11] That comes to mind here in Cam when we talk about AI is the diversity. [01:16] Of it only can you explain what the different types of AI are that are there being used in Healthcare marketing? [01:23] Sure, so first of all. I think it’s also good to just make that statement. There’s no one type of AI that exists that covers everything. [01:31] There’s different ways of using machine learning. [01:34] And large machine learning models to apply to different algorithms and out of that we get. [01:42] in my view kind of three general buckets [01:44] of artificial intelligence one is applied to test automation. [01:48] This is when you take. [01:50] manual tasks [01:51] You write rules around it and you basically automate them through bots you trained bytes on the tasks that used to be done by human beings. [01:58] And that creates a lot of opportunity for speed of Execution to accuracy of work. [02:04] The other one is predictive AI where you take large amounts of data. [02:09] And you predict you you do forecasting into US results can analysis. [02:14] and the last one is generative AI which is probably the one that people know the most about in the last year and a half because of [02:21] companies like GPT open AI that produce products like chege GPT [02:25] And Dolly and stuff those are the journey of AI types of applications. So those are the three types of technologies that are available to us. [02:32] To apply to different parts of what we do in Healthcare marketing. [02:36] And those are tools right either. [02:38] I would say that they are applications of AI as opposed to specific tools now. When all it mentions large language models and different ways of processing and analysing data. [02:49] Yes, there are tools but I think the application is where we’re talking about here the three probably most common ways to use AI [02:59] I also think it’s great to start this conversation by talking about. [03:03] the difference between gen AI and AI [03:06] ai has been around for more than 20 years in Healthcare marketing specifically. [03:12] Jenn AI has been around for 18 months and so [03:15] j’nai is [03:17] different because you can generate new content. [03:20] When all I can I were chatting we were talking about the concept of it. You know simple concept of a cat and a dog right you can. [03:28] Process fast amounts of data with AI to understand if this is a cat or if it’s a dog. [03:34] But with Jen AI you create a new animal you know that’s why it’s used so often. It’s the this dog and cat has wings and they’re a combination right so I think it’s [03:44] it’s super important and even when we hear a lot of the stalks alright, you don’t hear that distinction between jennai and AI [03:53] So we have we have these applications. [03:55] And the tools that come from them. [03:58] And it’s one thing to have them and it’s another thing to talk about them, but what we really want to do is adopt them. [04:03] How do you take these AI products? [04:05] And put them in the work and healthcare marketing. [04:08] right the one [04:09] common thread that I see across these kind of three buckets that which has discussed. [04:14] is [04:15] the data that feeds [04:18] the algorithms [04:19] And the best place to start. [04:22] Is making sure? [04:23] That the data quality. [04:25] that you feeding the algorithms with [04:27] is as good and is validated as possible. [04:30] because [04:31] to me [04:32] and [04:33] all of these [04:34] algorithms all these tools. [04:37] Are only as good. [04:38] As the data feed into them. It’s going to this idea of you know garbage in garbage out mentality. [04:43] these applications don’t think [04:45] they perform based on the data. Set that you provide to them. [04:48] So the better the quality of data. [04:50] The better the output you will get whether you’re trying to automate a task. What do you try to do some kind of analysis with the data set or whether you try to invent this? [04:59] hybrid between a cat and a dog [05:03] It’s such a great example. [05:04] I think also you know the the models that you’re using if you’re using llms you have to use. [05:11] specific Healthcare llms [05:14] we have specific llms for [05:16] Medical Science for example so if we look at MSL notes, there’s a different large language model that we will use. [05:25] Then when we’re analysing hcp research behaviour across a medical website or a published insight. [05:32] So, it’s super important to have the right large language model when you’re using genai as well. [05:38] I guess another important piece kind to follow the data thread is again in it. [05:44] Obviously very important to our vertical and health. [05:47] is privacy and just [05:50] understanding that the data says that you will. [05:54] insert in these tools [05:57] is protected [05:58] And you can’t just drop it into an open projected as in private. [06:04] Private it means to be controlled by you and owned by you. [06:08] As as a marketer. [06:10] Um because you don’t want to dump your information or your clients information into. [06:17] an open AI butt [06:19] And you don’t know where they’re going to store it and how they’re going to use that data set. [06:22] You need to be in full control of that. [06:24] And but how do you do that? How do you how do you fulfill? [06:28] The promise of this and still protect as much privacy that goes back to understanding what kind of tools that are out there. [06:34] And what kind of protections that provide you so before you decide to use to? [06:40] GPT or [06:42] co-pilot or whatever other big platform that are available out there today. [06:47] Make sure that the version of the platform that you use. [06:51] Gives you that level of protection where you get the story data in your own. [06:55] cloud [06:56] where it’s not allowed to leave your walls. [06:59] And I think one other point I’d like to before we move on talk about is what else is needed for adoption is. [07:06] The Invention and defining of new roles and positions within organisation I mean that that I think we’re sort of passed the point of people being afraid that AI is coming for their jobs. [07:19] Alright, it’s sort of it’s modifying people’s jobs actually that’s an interesting one and it’s not my quote so I’m not going to take credit for it and I’m probably going to butcher it here but I want to I’ve heard somebody describe it this way. [07:30] AI is not going to [07:32] take your job. [07:34] But the person who knows AI better than you will take your job right. [07:38] Which is a good way to can understand it and I tend to apply to the way we thought about email. [07:45] And I’m old enough to remember when email. [07:47] Came into our everyday lives. [07:50] and how that [07:51] tool [07:53] changed the way we operate and communicate. [07:56] I see AI being a similar way, so however understands the field most comfortable using this new set of tools. [08:01] will be [08:02] in much better place. [08:04] So eat it what do you tell clients? [08:06] about this you know if [08:09] in order to be able to take advantage of AI you need new positions within your organisation. What are you telling them? [08:15] I think a great example is prompt engineers. [08:18] Who knew what a prompt engineer was 18 months ago. [08:21] When we recruit and hire a new prompt engineers, they only have 18 months of experience. There is nobody with more than 18 months of experience. [08:29] So I think it’s understanding. [08:32] The the need for prompt engineers as one example as well as hiring and creating within your organisation. [08:39] space for innovation and titles for innovation around AI [08:44] And you need a different person for each. [08:46] platform or can people multitask across platforms [08:51] I think people generally will be able to multitask, but I go back to different types of AI [08:57] applied to different. [08:59] Types of work you still need that human intelligence human expertise in certain work. [09:05] And whatever tools I- built. [09:07] support that work [09:09] you need to be an expert in that. [09:11] Alright we all use. [09:13] outlook [09:15] for many different purposes in all our jobs and we’ll have a very different jobs, but we use that tool. [09:20] And to make our work. [09:22] More than dynamic, let’s say alright. [09:24] same will apply to [09:26] Any kind of tasks in marketing okay? [09:29] Let’s get specific. Let’s let’s talk about some some specific examples. [09:34] where you’ve [09:36] where your views a type of AI to implement a health marketing strategy? [09:41] so it can take the first one right so task code automation and here’s an example at cmi that we’ve [09:47] been active with for about a year now, so we have plenty of [09:52] results now to [09:54] talk confidently and [09:55] See that it’s actually working with taken a process. [09:59] for trafficking campaigns into ad servers [10:02] I’ve been in this industry for Close 25 years. I’ve been dealing with that servers and setting up campaigns. [10:08] This entire time it’s been one of the most manual. [10:11] And time consuming processes. [10:13] and it never improved in the last 25 26 years because [10:18] it requires armies of basically data entries. [10:21] and people to [10:23] plug [10:24] creatives and rules to targeting into ad servers [10:28] and checking and rechecking that this work is done. [10:31] Correctly and there’s no mistakes because if you launch a campaign with the wrong targeting. [10:35] That not only costs money it can cost a lot of create a lot of problems. [10:40] with taking that process [10:42] and we will automated it. [10:44] This process of following the rules of setting up campaigns in that servers. [10:50] Was automated and I’m not saying 100% I don’t think it’s possible to order me anything 100% you still need. [10:56] Again like I said human intelligence to oversee the process. [10:59] But we’ve been able to automate. [11:01] 70 to 80% of that process today and what that did was [11:06] That up are time to market. [11:08] bye [11:09] half so if before it would take us on average. [11:12] about [11:13] two weeks [11:14] to set up a campaign make sure it’s live now. It takes about 5 days. [11:18] the other very [11:19] important [11:20] point to make is in the year that we’ve been running this across all our clients. [11:25] We have zero errors errors. [11:27] And that’s the biggest one because again when you have a human process. [11:31] No matter, how hard you try and the reason why you used to take two weeks is because it took that much time to check and recheck the work before it went live. [11:38] Where now we don’t have to worry about that process because we know. [11:41] about [11:42] the tired [11:44] there’s not miss appear does not miss a dot somewhere it does not add an extra zero cutting and pasting something out of Excel [11:50] that [11:52] went away. [11:54] And that because of that now a year into it. [11:58] To me, it’s just it’s mind-boggling after 25 years of seeing this process and not finding a way to make it better within the year. [12:05] with transformed as completely [12:07] okay, so that’s a great example about tasks automation. What about predictive AI it is? [12:12] I can give a great example building on the example that oh like just mentioned about making a campaign live. [12:20] So we worked with our friends at cmi and the Weather Channel once the campaign was live. [12:25] And so a for a brand once the programmatic campaign was active. [12:30] We were able to develop add pieces of creative of you know different versions of a specific ad based on color hayti 80 based on color based on imagery now remember this had to go through the Amalur process. [12:47] But if you take that task and we can talk about AI and in the use of our Jen AI in. [12:53] In the mlr process, but let’s just focus on once this piece of content came out right and the 80 versions of the content. [13:02] we were able to use predictive AI [13:05] once the campaign went live to understand an hcps preferences. [13:10] for colour for imagery [13:13] and so now we were able to say talking about at the individual level like at the end of the individual level. [13:19] So now you have somebody who’s going to be served and add and this. [13:23] This model is learning and it’s learning that. [13:26] This HTTP is more likely to click on a banner, ad that’s a particular colour or using specific imagery. [13:34] um, I think that’s a great example of once the campaign goes live how we can continue to learn and optimise and use AI to to deliver the best engagement, so let’s let’s unpack that a little bit basically what you’re able to do as [13:48] ads run them through mlr. [13:51] And and then target it to the individuals preferences. [13:56] That that’s exactly right so before it was a bit of a guessing game when a creative agency would design a campaign. [14:03] And they will go like well. We think the green creative with the [14:06] klic throw over here. [14:08] On the message and running through let’s say mobile channel vs. [14:12] Desktop versus they even digital out of home. [14:15] All of that was a lot of guessing work. [14:17] now [14:18] to eat this point we don’t need to make all those gases. [14:23] We have all those options that we put into market and very quickly. [14:27] the algorithms and algorithm understands which variation [14:31] of these messages and types of messages. [14:34] works best for different types of http [14:37] in Market by channel, I mean this is as omni channels it gets. [14:41] So that’s predictive. What about generative AI [14:43] So that’s actually very nice bridge from predictive to generative because this is where we start to blend. [14:49] different types of tools [14:51] for [14:52] a particular process [14:53] some of these processes that we follow in planning and buying a fairly complicated and there’s different steps to it different stages of the process and we found that there’s different types of [15:03] AI [15:04] that could be used at [15:05] and stitched together at different types of process, so let’s take media plan. [15:09] planning for example [15:10] again [15:11] historically manual process [15:13] you have to take many different data sets. [15:16] you need to [15:17] organise them and figure out to figure out. What set of plans you can build what kind of [15:23] suppliers you want to use what goes you have. [15:26] So if you take that process this is where? [15:29] predictive AI helps understand the data [15:33] and process the data to recommend. [15:36] How much money to spend across any channel let’s say? [15:39] and then once that’s done a planner can use generative AI to ask very specific questions instead of before where they had to pull millions of reports dump them into Excel [15:50] and [15:51] do a Pivot Tables to come back with the results now, they just ask a question. [15:55] Which set of suppliers work best for this campaign? [15:58] Wow, it’s an open question. [16:00] and [16:01] the algorithms come back with an answer. [16:04] So now we’ve tied two different types of AI in the same planning process. [16:10] I think I can add to that oleg because we use another type of gen AI to at the beginning of that campaign and so now the campaign is live. [16:20] where we’re looking and optimising [16:24] but what is the data? That’s feeding the personalization in that campaign? [16:29] And at iqvia, we have behavioural insights on individual hcp’s where we understand their research behaviour across 5,600 medical websites. [16:40] And so we’ve been using Jenna I to better understand. [16:44] And hcps research behaviour, how can we create a summary of what’s important to that hcp so when we start to engage and create that media plan that is personalised. [16:55] We have this Foundation of need so this hcp has a clinical care gap. [17:01] Or they have an information need around a specific therapeutic area or even a brand. [17:06] And so Jedi is helping us understand that need. [17:09] Creates summaries for that need and to all explained across multiple channels right, can we feed these insights not just a digital media but to the sales force as well? [17:21] And so we’re seeing some really great results both in summarising need. [17:26] But also understanding content affinity of hcps. This is a format of the content. [17:33] They might prefer a podcast like the listeners today or they may prefer to read a clinical article. [17:41] So it’s super super interesting and it’s so early right. It’s we’re just at the beginning. That’s such a great point. We are just at the beginning and [17:50] if I were to use any kind of analogy [17:53] again another analogy her recently what inning of a baseball game away in right now. [17:58] we’re not at [17:59] even stepping up to the [18:01] plate yet for the first day warming up still warming up. I think that the amount. [18:05] of information [18:07] that we get right now amount of confusion around what is available to us will shake out. [18:14] And the more confident we get going back to comfort and confidence of using these tools. [18:18] I think the better we will know in six months and 12 months. [18:21] Where true opportunities long term will be for us we definitely see. [18:26] amazing results already [18:28] but again we’re just scratching the surface. [18:30] So is it a fool’s errand to take out a crystal ball and ask you to to see where we’re going with this or or are you feeling bold and you’d like to make some predictions? [18:39] I’m feeling bold and saying that nobody knows and however says that they know are either lying to themselves or lying to everybody else. [18:48] Like and completely agree with you. [18:51] All right well. [18:52] That’s the net that’s that then. I think that’s a I think it’s also the little that I know about the topic. I would agree with you because I think it’s just it’s way too early to try and predict anything about it except that it’s here to stay. [19:05] And we should learn how to use it and we should embrace it rather than push it away. [19:10] This has been a really great talk. I really appreciate your time. I know it’s very busy here it can only corn felt from cmi. [19:18] And it is hard consent from iqvia. Thanks very much for coming by [19:22] Great talk, thank you Steve thank you so much.