Issues Affecting AI ROI
Issues Affecting AI ROI
1) Data; 2) failure to define the problem you are trying to solve; 3) trying to apply AI where it is not the right tool; 4) silos; and, 5) insufficient skillsets.
Join the DZone community and get the full member experience.Join For Free
Insight for I&O leaders on deploying AIOps platforms to enhance performance monitoring today. Read the Guide.
To gather insights on the state of artificial intelligence (AI), and all of its sub-segments — machine learning (ML), natural language processing (NLP), deep learning (DL), robotic process automation (RPA), regression, et al, we talked to 21 executives who are implementing AI in their own organization and helping others understand how AI can help their business. We began by asking, "What are the most common issues you see preventing companies from realizing the benefits of AI?" Here's what they told us:
- It is just about a cliché now, but finding good AI-training data is always a fundamental challenge. Besides that, one of the biggest issues I have seen is a disconnect between the business side and AI research teams. Building products that are both technically feasible and that meet market demand is the challenge. On the research side, we can see the value of this really cool technology, but it can be difficult to monetize if it doesn't solve a business problem. That is why a strong collaboration between product managers, business leadership, and research is essential to new technologies being successful. We are nimble enough to allow leadership and AI to collaborate directly. If I am unsure of a particular feature that I am building, I can walk down the office and ask the product manager or even the CEO for clarification about business needs.
- Not having a handle on their data across silos. When the data house isn’t in order. Can’t succeed until the data is in order. Simple to say but so much data in so many formats sheer volume it’s overwhelming and complex and becomes a barrier without the right AI platform to handle this immense type of data.
- Not understanding the true state of their data. Enhanced data curation, quality, and readiness assessment. Lots of data but they cannot get their arms around what is usable and what isn’t.
- One of the main barriers is the training time for AI platforms. AI requires tremendous data set and time spent to digest and train itself based on the information it was fed for it to offer a robust platform with high accuracy.
- The biggest issue we see is the need for valid, accurate data. Data is the new oil, so to speak, and it is imperative that companies have valid data in order to be successful in any new technology implementation, especially AI.
Defining the Problem
- The ability to clearly see the right business problem to address and having the skill to do that in a good way. AI requires a different programmer. Need to be the one that can see the big picture and see where the right place and way to use AI.
- Going back to your question earlier about a successful AI strategy, many companies do not have an adequate understanding of AI and have not answered those basic questions, which leads to unrealistic expectations. This and poor planning are perhaps the most common issues seen. Beyond those, there are many technical and business issues, chief among them are cost and maturity of the technology. Skills and talent are also in short supply. Talent poaching is a common way of acquiring the needed skill sets; however, this impedes progress as employers seek to replace lost talent. Use cases continue to emerge, each requiring a different algorithm. These algorithms have yet to be developed and trained. Training data sets require vast amounts of raw data to produce accurate results. However, not just any data will do. Data must be relevant and adequately cleaned for the results be very accurate. The axiom garbage in, garbage out still applies. One final point is that many organizations fail to plan for scale. Infrastructure needs are non-trivial; organizations need to plan for data stores to grow very quickly to Petascale. Likewise, infrastructure must be scalable to grow in lock-step or costs will quickly get out of hand as systems are re-architected to keep pace.
- It’s important than an organization approaches AI from the starting point of, “what problem do we need to solve?” rather than “let’s do something with AI.” And it should be the right problem for which AI can really have an impact. AI and Deep Learning are best suited for tasks where you need to find patterns and useful information in data. If data does not exist or labeling it is not possible, you can’t train the systems for the task at hand. It’s also important for organizations to understand that AI is just one component of the solution: you need to create workflows, integrate with backend systems, take usability aspects into account, test, train users etc. In other words, AI projects are also “normal” IT projects in a sense. Enthusiastic young startups and their customers often overlook these aspects. It is easy to build an impressive demo, but a lot harder to build real systems that solve real problems.
AI Use Cases
- Lack of data. If you can’t measure it, you can’t learn by example. There are cases where you have to learn it the first time. There are cases that are too horrific you can’t await data (plane crash, terrorist attack). People attempting to use AI/ML where it’s not needed. Distracted that ML has to be used everywhere. The best way to get ML value is to understand what’s really going on. The biggest benefits are the people who understand their business at the most granular level. Get the data scientists out there to see where the value is really coming from. Sprint stands for Southern Pacific Railway Intercontinental Network of Internal Communications. It’s a right of way company we can push trains or bits down. Reconceptualizing the value of the business. Graphics business goes from static signs to changing sign, to the web to print — small steps became a much bigger and more profitable business based on where the value was. What made a difference?
- AI is the solution to a lot, but businesses need to understand that not every problem is an AI related problem. If applied correctly, AI has huge potential to automate and grow. The focus should be on using AI algorithms to get the foundation right for technological advancement within a company.
- One of the main challenges is breaking down the perception and identifying what is true AI. Because we are so early on, many executives and leaders don’t yet fully appreciate how AI and machine learning can be applied within their organization, so they’ve underinvested because they just don’t know yet what the possibilities are. With those that have made significant investments in AI, it is critical to define where it will have the most impact and ask the questions on what high scale problems it can help solve.
- For this area, we focus on recognizing this as a transformation for the technical side and the organization. Collaboration is important because ML is so technical you have a strong disconnect between operations and data science. Most who didn’t study data science do not understand the algorithms. When the algorithms change behavior in production, you need to understand the algorithm. How are companies creating a people practice around operationalization? Some have applied AI teams. Some have virtual teams where two days out of the week the data scientists are in operations. The analogy to DBAs needs to train a person who understands databases and operations. And databases have just become more complex over the past 10 years. We help the two skillsets collaborate.
- Anywhere between FUD and challenges of organization and culture just like DevOps.
- We think successful delivery of AI in any enterprise requires involvement and awareness across organization towers and stakeholders. Besides that, mindset plays a key role. Implementation of AI depends on successful technology migration which has significant business risks without proper planning, design which requires tools to do so.
- In general, it’s a new area. Get statisticians to do more research. It's a skill set problem. SMEs understand data sets and the problem but don’t know how to manage and process data. Don’t know how to extract from Hadoop and run EDL. Hindering more return on AI investment. Might need multiple teams collaborating to fill knowledge gaps. 2) Run code on any hardware. AI is run on GPUs which is new technology. Acquire GPUs and new part of the platform still being adopted. Not as varied as CPU instances more maturity needs to happen.
- Lack of quantitative expertise in-house. Agencies are no better, so we end up with a blind leading the blind situation. We need more effort around executive education. Arm execs with the right questions to ask. AI solutions provider need to be able to help with execution.
- While we’ve seen an incredible surge in the automation space, some of the business applications are clouded by the noise like an autonomous van in an accident on its first day, or a smart suitcase fleeing its handler. Such headlines are leaving people wondering, “does AI even work today?” We know that we left 2017 with AI as a new buzzword, but is there any AI application actively improving workflow and efficiencies for the enterprise? There is; you just have to look for it in the IT department. IT infrastructure already generates huge amounts of event data, and all of those events are already, by definition, formatted to be easily machine-readable. Additionally, once an AI solution has made it through this initial skepticism, an enterprise may be overwhelmed by the daunting task of ripping out their legacy systems for a new model. To resolve this, we integrate easily with the systems that teams use and cherish and enhances them — providing an opportunity for overall growth, not simply destroying what has worked in the past in hopes that this new model will resolve everything at once.
Here's who we spoke to:
Opinions expressed by DZone contributors are their own.