Despite billions poured into AI, only 25% of Fortune 500 projects hit their ROI target—data challenges are one of the biggest hurdles.
Taming the data monster is notoriously difficult. Because there is such a tight coupling between data and outcomes in adopting AI, it's tempting to think that we can't get anywhere without perfect data.
The reality is that it's always a dance across data, outcomes, and algorithms. Undertaking a lengthy and costly data exercise is often difficult to justify without a concrete link to ROI and, more importantly, it's often difficult to get right without understanding what you are looking to accomplish and how you plan to get there.
Start with Business Objectives, Not Data Perfection
While open ended research and capacity building is important to understand the art of the possible, any innovation projects that you ever hope to take to production should start from concrete business objectives. We can use early prototyping and POC phases to explore technical options including what data is critical to delivering on those objectives and then undertake focused data cleansing, enhancement, and acquisition activities to ensure the best possible foundation for our project.
The Role of AI in Data Processing
AI tools themselves can revolutionize this effort. The ability to process unstructured data, image data, video data … is progressing every week. Equally important, however, is our ability to recognize and design where we need strong QA/QC and human involvement to ensure quality output.
The Bottom Line
Perfect data is a myth. Start with business goals and let your understanding of the data sharpen through doing, not delay.