Many organizations are deep into the process of digitally transforming their business, often with varying degrees of success. In many cases, businesses implement artificial intelligence (AI) as part of the technology mix, especially in areas where speed or scale are challenging for traditional methods. Yet in some cases, early investment in AI fails to live up to the hype. What can you do to make sure your project is different?
Remember, It’s All About The Data
The job of AI is to mimic the cognitive behavior of humans — perhaps faster and perhaps better, but ultimately any AI functions to process data and produce results. To do so, AI needs large amounts of high-quality data. In many business applications I’ve seen, poor data hygiene practices are the root cause of why AIs fail to deliver the right results. Yet there are other reasons AI projects can produce less-than-stellar results. As a business leader, there are a few options you can leverage to prevent costly failures. Here are five that I’ve found to be the most critical as the VP of products at an AI solutions company:
1. Define The Right Problem
Before implementing AI technology, business leaders should take a step back and ensure they are crystal clear on the problem they need to solve. Too often, organizations apply a very generalized set of technology, such as AI, only to realize that they need to take a more specific approach to problem-solving. The reality is that, much like blockchain, the level of hype around this kind of emerging technology can make it difficult to know how to apply AI optimally. If you don’t have a clear statement about the problem, you may apply technology investments to the wrong area and solve the wrong set of problems. This can cause your organization to associate AI with failure and make business leaders reluctant to invest further.
2. Determine What Success Means To Your Business
As Yogi Berra is credited with saying: “If you don’t know where you’re going, you might end up somewhere else.” Even when it’s correctly applied, AI technology may not return the desired results if a company defines success poorly. AI, by definition, can optimize decision making performance based on a feedback loop that adjusts the underlying models to “learn.” However, without a clear perspective on what success looks like, the AI can quickly begin to optimize for something completely different. Simply put, the technology can learn the wrong lessons and reinforce that incorrect model over time. Examples abound in image recognition systems where seemingly intelligent AIs can be fooled by even minor image changes or simply inserting unexpected elements the system was not trained to ignore. Have a very clear, definable measure of success when you’re training an AI to optimize itself around specific metrics.
3. Ask What Data Is Needed And Who Has It
One of the defining aspects of AI is its appetite for data. A good model often requires an incredible amount of data, although there’s a lot of focus today on building approaches that are less data-hungry (paywall), especially for tasks where large data sets are difficult or impractical to obtain.
Regardless of how much data is consumed, I believe the critical question is “is this the right data?” In other words, does the data align with the goal you set previously? (Whether the data is “good” is another question we’ll touch on shortly.)
It’s important to gain clarity on what data the AI technology needs to run, as well as a broad level of agreement on who has it, where it’s located, and if it’s usable. For example, privacy issues are a major hurdle that you should consider and that can impact a project if you don’t.
Before starting out on a project to utilize AI technology, it’s a good idea to make sure you know who has the required data and confirm access to that data. Otherwise, your AI project could starve long before it can produce meaningful results.
4. Fix The Data
If there’s one area where I’ve seen businesses struggle to implement AI, it’s making sure the data is good. Rarely are businesses able to magically produce clean, consistent, and relevant data when and where they need it. This is why it’s critical to do the pre-work by fixing the data before you send it to the AI for analysis. While the adage of “garbage in, garbage out” has been a fixture of the IT world since its inception, I’ve found that this is especially true of AI systems, as poor data hygiene can yield even poorer results or no meaningful results at all.
For example, is the same kind of data, such as dates and times, formatted similarly so that the AI can consume it? Are there gaps in your data, and if so, do you know how to treat them consistently (ignore, extrapolate, average, etc.)? Simply getting the data ready and being clear on the steps you take to fix bad data are important parts of building the right model — and ultimately getting the correct answer out of the process.
5. Consider Governance, Care And Feeding
Now that you finally have an AI system up, running and producing meaningful results, it’s time to transition it from a pet project to part of your core functionality. AI technology can produce incredible results, but you also need to provide it with regular care and feeding. Managing the technology to track metrics, tune the process and keep good data flowing are steps you can’t afford to skip if you want to reap the kind of transformational benefits AI can offer. Make sure you keep your infrastructure and ops teams aligned, your CTO informed, and your senior stakeholders on board.
Ultimately, AI can transform any business. It’s a complex technology that can make a remarkable impact, but to leverage that power to foster success, businesses need to make an investment in processes and planning to fully unleash its potential.
By keeping these five tips in mind, leaders and organizations can hopefully ensure successful implementation of AI technology today and lay the groundwork for the future, too.