Using data and AI in direct marketing

without being creepy or selling your soul

A few weeks back I made my way to sunny Brisbane to speak at a seminar on the opportunities and challenges of incorporating data and AI into your personalised marketing strategy. It’s not a new topic of course, but analytical marketing professionals, like all the other folk who work in and around data are taking a fresh look at what they do today, what they could do tomorrow and what fits with both their expertise and their brand promise.

I had expected to enjoy a warm morning run along the river without the need for gloves and a beanie (for non-Aussie readers, sub-tropical Brisbane is A LOT more than my home town of Melbourne at that time of year). I hadn’t realised what a warm reception the topic would get from over 100 folks who were up bright and early and with a lot of great questions, significant concerns and a strong interest in compiling a list of questions to test out the many, many vendors spring up in this space.

Here’s what we came up with.

🤔 Has all the data on your platform been obtained with explicit consent?

🤔 How do you assess your models for bias?

🤔 Is the data in your model inclusive of my target audience?

🤔 Will the data I share get incorporated into your models and rulesets?

🤔 How do I incorporate my brand guidelines into the generated copy?

🤔 How easily can I integrate this into my exisiting workflow?

And finally remember that if you can't measure the impact of whatever fancy new AI you've rolled out, it can end up being a bit of an exercise in futility.

Unfortunately, I think a lot of vendors, both new and established will struggle to answer all of these questions. For some use cases, you may not be particularly concerned but for others, embedded bias, lack of consent for secondary use of the data underlying the model and data that doesn’t reflect your audience can be substantial hurdles.

There’s an apparent tension between whether your data is incorporated into future vendor models vs the need to ensure brand guidelines are followed in generated copy but there are solutions that savvy vendors should be able to work through with you. It’s not unreasonable and already accepted practice to sign contracts that exclude your query data (which includes any prompt engineering) from being incorporated into new models. Yes, this will put more of the burden of customisation on you. There’s no free lunch.

A short post this week as I’m away on holiday enjoying the stunning scenery of New Zealand’s Southern Alps.