The AI wave has come on faster than most (including me) expected. Over the last decade, companies have slowly but surely started deploying AI. Return on Investment (ROI) was beginning to turn positive. Then ChatGPT arrived, and suddenly AI was everywhere. If a company thought AI was critical to its future before, it has now acquired a new sense of urgency far beyond what it previously had.
That said, just because something sounds urgent (or is, in fact urgent) does not make it any easier for an organization. Some factors (such as the release of more powerful tools) make adoption easier, but the urgency does not translate into a similar ease of implementability. In particular, if an organization does not have a solid data practice or data governance policy – can it do AI? This is a challenging question. On one hand, the easy answer is data has to come first. AI relies on data, and bad data leads to bad AI or worse, legal and other problems if bad data is used to make problematic AIs. On the other hand, at the pace at which AI is moving, can an organization really say no AI at all till they have achieved acceptable data practice and data governance across every aspect of their business? What would the opportunity cost be? And when would the data strategy be acceptable enough? As AI changes, will the needs of the data strategy also change?
As in many cases, the easy answers are all insufficient (at least in my view). Good data practice cannot be ignored, and AI cannot wait. So what is to be done? Some ideas follow
Start with the problem.
This is good practice regardless of where your data strategy and data readiness are at. AI that delivers good ROI starts with a focus on the problem, not the AI. The questions you should ask yourself are:
- Why do I believe AI will generate ROI and solve this problem better than it is solved in my business today?
- Does the answer to this depend on the data I have in-house? Often the answer is yes but sometimes the answer is no. For example – if you license an AI tool to help your employees write better reports, that may not require any or much custom data on your part. If your data strategy needs work, you may benefit from prioritizing work on your data strategy while leveraging AI options that add value to your business but do not require your in-house data.
- If the answer depends on the data I have in-house, what is the state of the data? What do I need to do (for this data) to get it into a state that can support AI to success? What best practices can I put in place along the way (and data practice investments I can make) for this data that I can also leverage elsewhere?
- If I am convinced of the above, does my organization understand that the data practice and the AI practice have to evolve together? Is it clear to my stakeholders that putting the AI ahead of the data is likely to fail, but if properly executed, all data for all use cases does not have to come ahead of AI for a particular use case? Do I have an organization in place that can orchestrate this (arguably more complex) workflow?
A few things that are critical to get right
- If you have any doubts about the quality of your data, fix those before investing in AI. Bad data always leads to bad AI.
- Do not let the AI create any data-related ethics violations, such as bias or privacy issues. Follow good data practices as you develop and deploy your AI. Some example guidelines can be found here.
- Make sure the data you use to build your AI is protected from theft and loss. Once an AI is built, you should expect to train it again and again. Loss or damage to the data will then damage a launched product or AI development investment.
The pace of AI evolution is fierce. There is unfortunately not much opportunity to wait till all dependencies are fully met. However, organizations that do not have their data practices in good shape will lose out on AI in the long run, and organizations that have solid data practices will have an advantage with AI. If possible, thread the needle and get both done.