How to do an AI implementation for your marketing team<\/em><\/strong><\/a><\/p>\nWhat we needed to do was, instead of just giving it to them as a blank slate, only have it do a few things.\u00a0<\/p>\n
One was to provide already-generated prompts that they could use, but those prompts needed to be relevant to that marketer themselves. We had to create personalized prompts, which meant we had to engineer it so that it takes user data and knows who the user is, what their job title is, their previous campaigns and their current campaigns. Taking all that information and generating a prompt that is relevant to them.\u00a0<\/p>\n
It also gives another view that enables them to explore insights the AI has already found. Then when they click on one of those insights, it recommends prompts that they could use to explore further. So rather than just giving them a blank slate, they needed some guidance on what to ask and what to do next.\u00a0<\/p>\n
Q: <\/strong>It\u2019s kind of like a two-dimensional person being presented with a three-dimensional world. You have to think about the data now in other directions. How should companies be thinking about training people around this? I run into this assumption from some places where they think, \u201cOh, it\u2019s natural language, you don\u2019t need training.\u201d Should there be more training of people around the prompts and thinking about the data?\u00a0<\/p>\nA: <\/strong>Yes. Yes, absolutely there should be. That\u2019s the only way we saw it working. Otherwise, the adoption rate was just low. When we started to train people on basic prompt engineering, we then started to see an uptick in the adoption rate.<\/p>\nFor the adoption of these tools, it\u2019s essential that marketers are trained on how to prompt them. You must also ensure that those tools are guiding the marketers as well, that they already have existing prompts or templates that they can click on.<\/p>\n
We rolled out some prompt engineering courses to the marketers so that they know how to effectively prompt these AI chatbots to get the answers they need.<\/p>\n
We have an AI trainer and he comes in and trains our clients on how to prompt within their specific area whether it\u2019s marketing or within marketing, whether it\u2019s analytical marketing or the operational side and so on. Having a trainer coming in to do a more tailored way, we found that more effective than just a general sort of prompt engineering course.\u00a0<\/p>\n
We also create prompt templates as well so that people don\u2019t only have to rely on training. They can actually pull from a prompt library. So we\u2019ve organized a prompt library for each type of team. If it\u2019s a team that focuses on partner marketing, for example, we have a prompts library specific to that team and we are continuously building on that prompt library updating it to make sure it has the most relevant prompts there.<\/p>\n
There are collaboration elements as well, where people contribute to the prompt library, create their own prompt templates and upload them into the library to say, \u201cHey, I\u2019ve got this new prompt for that.\u201d And we\u2019re seeing that\u2019s organically developed as well as people are using and adopting these tools more.<\/p>\n
Q: <\/strong>And that gets them to learn and explore?<\/p>\nA: <\/strong>Sometimes. After every prompt, if it gives recommendations on what the next prompt could be, you actually find that the marketer doesn\u2019t act. When we look at the user stats and how people are using the tools, we find that 60% of the time they don\u2019t actually type in anything to the prompts. They\u2019re just selecting the recommended prompts from one recommended prompt to the next recommended prompt to the next one, to get those insights. That\u2019s pretty important as well. It\u2019s like people don\u2019t want to do any work, they just want to be guided by the right prompts.<\/p>\nQ:<\/strong> Can you run into the problem of people saying, \u201cOh, that\u2019s what the AI said to do,\u201d and going with that?\u00a0<\/p>\nA: <\/strong>Yes, if there are no controls in place, yes. If you don\u2019t have a policy of validating these AI results, then you run into a risk of your team just lazily relying on the AI\u2019s recommendations and just going with that. You need to have an AI validation team in place so that if you\u2019re following an AI recommendation it needs to go through some sort of validation. You need to have somebody manually go through that to ensure that this is the only option or what are the other options that the AI did not see.<\/p>\n<\/p>\n