Like many (many!), I have played with Chat GPT to see how if it lives up to the AI promise and test if it can contribute to better marketing strategies. In this article I share my experience with Chat GPT when it comes to information gathering.
Information is a key component in strategy : like in battles, knowing the terrain and opponents’ moves helps understand the easiest ways to win and the strengths to build on. Over the years, I have have used countless sources, from existing to created information. In this journey, I have acquired a taste for short but meaningful data – to avoid “death by power point”. But above all, I have learned to challenge sources. So, I asked 2 types of things to Chat GPT : market data and brand profiles. Here’s what I learned.
1- Challenge Chat GPT answers.
I once wanted to know about the top 10 spirits companies in the world. Nothing very complicated or confidential, you’ll tell me. The machine spat the usual suspects (Diageo, Pernod…) but I wondered why Chinese companies (e.g. Kweichow Moutai, Wuliangye Yibin…) did not appear in the ranking. To which Chat GPT replied it only took into account groups that had diversified international businesses vs. domestic mainly (like the Chinese companies). OK, understood, even if arguable.
Then, while examining the list, I noticed players like Proximo were also missing asked Chat GPT the reason of this omission, to which it replied that it only produced a ranking with companies that were publicly traded. I then indicated Chat GPT that even if Proximo is privately owned, a 1.5 bio USD turnover surely would make it worth including among the largest companies, to which Chat GPT agreed, politely.
2- Keep your critical eye when it comes to information, anyway.
But wait a second… challenging data should be the same thing in real life. Don’t we all double check data? Actually… no, it is easy to trust answers which come from “authorities” and probing answers takes a lot of work, making one look like a painful sceptic. Everyone seems shocked at the AI power, but if you think of the GPS navigation systems, it already is a way most of us trust to find one’s way, and all this system does is to suggest routes based on a question (I want to go from A to B) and constrains (roads available, traffic…).
So we live in a world were we already trust blindly data produced by machines, because their complexity leaves no other choice. And this reliance on technology can be worrying, especially if the technology breaks or disappears.
3- Refine your questions and feedback. It’s all about context.
So it seems that Chat GPT is “happy” to learn from a conversation. It also shows that the information provided is not flawless, or let’s put it the other way : the quality of the answer depends on the data the machine finds and the way the prompt is formulated.
So, Chat GPT seems to be a “super Google”, and one should always check the answers and the way the questions are formulated. However the prospects of humans being only good at making good questions (or “prompts”) with no critical analysis is a bit worrying. But if anything, the exepriece confirms the proverb “good questions yield good answers”.
4- Use Chat GPT to open investigation trails
The second information type I asked Chat GPT to provide were things like “what can you tell me about brand XYZ?”. Here the machine, tended to paint rather positive pictures about the brands (a brand which is famous…), in a very politically correct fashion, no matter the state or size of the brand. Almost as if it was afraid to be sued. So obviously the machine went to pick the positive stories about brands. If actually one wants to dig our more, then prompts need to be more specific and open conversations about controversies or particular episodes.
But as the machine learns from the conversation thread, I’ve read with interest the answer to prompts such as describing the main archetypes for brand XYZ, and then asking the machine what in the brand product or communication would make it describe the brands as a “Jester” or “Dreamer”. So, from a qualitative point of view, the answers were interesting as they either corroborated some observations and conclusions I already had made, or provided some routes to further consider to reinforce a case.
5- Use Chat GPT to complete other information
In both cases, Chat GPT felt like a “super google” because it allowed to formulate questions in a more natural way vs. Google which will tend to look for the exact terms. Futhermore, the model seems to learn from a conversation, and this allows to finetune answers. However, it did feel sneaky as the machine did not want to list its imformation sources.
So… Do we have trust Chat GPT blindly? No way. Because it is a machine and there are many discussions about “it” taking human jobs, I think it makes us more demanding about the answers’ quality. But hold on, when it comes to data, we should also have a healthy dose of skepticism, no matter where the information comes. Need we remind the consequences of “fake news” ?