As we start the new year it may feel overwhelming to deal with the challenges in front of us. From regulations changes and how they should be enforced, including with quantum technology advancements; the pace of AI technology changes introducing new concepts, sometimes in a matter of months, from Machine Learning (ML) to Large Language models (LLM) , vector databases and Retrieval-Augmented Generation (RAG) , to Model Control Protocol (MCP) , Agentic AI, what’s next? AI by itself comes with not only fast-paced technology updates but brings uncertainty if not anxiety as what will our industry become in maybe just a few years from now. While generative AI has produced amazing outcomes, it is also still making significant errors, requires costly infrastructure and struggles to show return on investment. Numbers are known, showing between 70% to 90% of AI projects are failing and some reports citing 95% of generative AI pilots failing to show measurable ROI.
Why are so many AI projects failing?
One of the causes often identified is poor data quality. So, what do we mean by this? Is this unweighted data accuracy going into LLMs? Biased data? Privacy settings preventing access to only relevant data and not just public samples? From a database portfolio experience point of view, this is not the first time we have heard about this issue. The same data integration issues existed already to provide accurate analytics. Data silos continue to exist, not because of technology barriers (we’re beyond the limitations of ETL (Extract-Transform-Load) data integration) but because of data governance. The point here is not to convince everyone we need to provide quality data; We all agree on that; the point is are we able to solve data governance issues as they are uncovered in AI projects?
When Time-to-Market rules all projects
This is where another factor comes into play which is “time” and that I would propose as a second reason why most AI projects are failing. Our world has become very averse to slow time.
Projects are from the start under the Time-to-Market constraint, governed by the budget we’re willing to invest, but also that we can’t grasp anymore that it may take a year or two to implement something new. This is great news for AI technology. AI appears to be able to deliver immediately, especially gen AI, with millions of examples of vibe coding, generating MCP servers in a matter of hours, truly amazing speed of delivery and increased productivity. But let’s go back to the key reason for failing AI projects, data quality. How quickly are we able to put together the correct dataset together to feed into AI? Is the data under our line of business governance? If not, are the other parties willing to exchange data? Is the data static or dynamic? On what terms do we have access to the latest data? What if the data required is under strict data privacy rules? As we’ve heard many times, data is the new currency, so yes data has value and it also means that you may not be able to get it for free! This is maybe the most important shift visible in AI technology. Source code used to be where the Intellectual Property (IP) was the value. With AI this is shifting to the datasets and getting access to a dataset is not as immediate if not a dead-end that causes your project to fail.
What’s in it for Nonstop?
My proposal would be that any AI project should enforce 2 fundamental work streams. One on the AI technology and one to discuss the scope of the data that is required, resolve governance questions early on, this may require new types of agreements and value exchange. This is especially essential if you want to use multiple disparate datasets, coming from different institutions with strict privacy rules such as in human trafficking or fraud detection use cases. Nonstop is one platform that has high quality data as long as if it is used in real time, with the correct data governance and augmented with strict data privacy enforcements such as anonymizing, masking or encrypting data where it should be. We have the technology that is required, there is no showstopper there for Nonstop. Data created on the Nonstop are secure transactions that include valued, single version of the truth, data that IT can monetize for AI projects and surely increase the chances of your AI projects succeeding.
Final Thoughts
AI will continue to accelerate, but the organizations that succeed will be those that master data governance. With NonStop, we are already equipped with the quality and security foundations needed to thrive in this next wave.
I wish everyone a happy new year!
Roland Lemoine
HPE NonStop Product Manager
Be the first to comment