There is a period of all of our lives where hormones cause us to do crazy things usually because we were not thinking. In the frenzy of creating exciting AI prototypes, there is a similar affect where our data hormones can get the best of us as well. But to truly operationalize and scale an AI-based digital platform credible data practices need to be used or you could face angry customers, angry regulators, and angry internal policy makers. Here’s four tips that will help ensure your AI prototype gets a chance at a scaled implementation.
1) Scraping: ’Just say no’
It’s time to stop using all the free stuff you can get your hands on and ensure you can get the real data needed to scale that prototype. Otherwise you put your whole AI project at risk. If you’re a data scientist go to your Chief Data Officer and let them know what you need, see if they can help you get it. They may even have other projects that could use it and therefore they could apportion costs to each group and lessen your cost burden. Once you go to external facing product from your internal facing Minimal Viable Product (MVP) you will be creating data ethics nightmares if you are scraping data. Because your MVP could then be trained to your scraped data and your scraped data may not have a legit equivalent, you could find yourself having to start all over again.
2) Data Brokers: Vet their data sourcing and collection methods
We have all learned from the Facebook and Cambridge Analytica situation. Not everyone who sells or prepares data is going to be ethical about it. If you are making an AI app that you want to bet your career on, don’t you think you better vet where they got this info from? Do a bit of data lineage research. You could again recruit help from CISO, Chief Data Officer, CIO so the burden doesn’t fall completely on you. No matter what you just want to ensure that the data was collected in a way that wouldn’t make headlines for being creepy.
3) Business Partner data: Check, validate, and provide customers options
Just because you know someone who is gullible about data swapping in the Marketing department at a perfectly legit business partner of yours does not mean you should get them to send you the data. In data science, there is always a tendency to want more data. I know it is heretical not to want more data. I get it. BUT, there are times that getting exactly what you want may do you in. Your clients will backlash if they think you know more about them than they know about themselves. So I ask very carefully how it would be perceived for you to have this information? If you have to question that, then let your customers decide. Tell them your idea for creating value for them and let them Opt In or Out of participation.
Let me give an example:
If you’re in healthcare insurance your customers’ worst fears are that you might use their genetic predispositions against them or their family and either opt them out or make the premiums so high it would opt them out. If you then announce a new AI-based app that tailors preventative care ideas directly to them and have something like this in the app “You may want to start upping antioxidants in your diet. Eating berries can reduce the affects of breast cancer”. Then this could clue them in that you have access to their worst fears, their genetic information.
Now this could be considered very helpful if your client has opted in to this program, and given full permission to you to have this data, and also that they know in advance they have the breast cancer gene. However, if they don’t, then you will need to first obtain permissions to link data together and make assurances that clients cannot be dropped from care by providing the information.
4) Check your data policies before bartering data
Another easy way to get data could be to barter for it. I’ll give you some of what I have, if you will give me some of what you have. Even if you do this inside your own company and not with external business partners, just ensure that you have permission from your Chief Information Security Officer, Chief Privacy Officer, Ethics and Compliance Officer, Legal Officer and/or Data Officer to use the data for the purpose you will use it. While you may think you are just making a quick swap with a colleague, privacy terms for specific uses for data and GDPR can be violated easily. So it’s always good to just make absolutely sure you are okay to proceed with data that may not be in your normal jurisdiction or that individuals inside your company are supposed to have access to it.