Earlier this year, I was on my way home from a conference when I learned my flight from Seattle to Denver was delayed. That by itself was no big deal \u2013 happens all the time. Unfortunately, it was the first leg of a two-flight trip, and the two-hour delay meant there was no way I was going to make my connecting flight out of Denver to Dallas. As frequent travelers know all too well, that\u2019s when things start to get messy.\nThankfully, I thought, I was a card-carrying, first-tier member of the airline\u2019s premium mileage club (company and club shall remain nameless). Surely, a quick phone call to the exclusive phone number I\u2019d been given would connect me directly to a friendly ticketing agent, who would verify my exalted priority status and work some magic to get me home at a reasonable hour. Alas, it was not to be. The phone was answered by a machine, and the computerized voice assured me that while they were experiencing unusually long hold times, my call was very important to them. After what felt like an eternity, I did speak to a real person, who spent 20 minutes asking me questions, combing through various flight schedules, connecting airports and seating availability, and trying (to her credit) desperately to stitch together a workable solution. Eventually, we decided the best bet was to cancel my flight and rebook \u2013 at the airline\u2019s expense \u2013 with a different carrier that could meet my needs. I finally made it home in the wee hours, and the next day was a blur.\nThroughout the ordeal, I couldn\u2019t help but thinking: here was a job (the job of making complex travel reservations) that could have been handled immediately, and almost instantly, by artificial intelligence. It was a matter of sifting and sorting through mountains of data to satisfy a specific set of requirements \u2013 something humans aren\u2019t particularly well equipped for, but which is right in a computer\u2019s wheelhouse. Theoretically, when I called, AI would\u2019ve answered with no hold time, and could have run through every airline and airport database in seconds to present me with options. Even better, it could have already known that my flight was delayed and contacted me proactively to ask if I wanted to change flights! In that alternate universe, I might have gotten home faster with less stress and frustration, and the airline could have salvaged some customer satisfaction with no human intervention required.\nSo the question is, why isn\u2019t the airline already doing this?\nWhat\u2019s the hold up?\nMy airline snafu, which happens to thousands of travelers every day, is just one example of how companies could be using AI to operate more efficiently and improve the customer experience \u2013 but they\u2019re not. We know the technology is out there and that some organizations are putting it to work in similar, and even more impressive, use cases. But many more companies, despite making massive investments in AI, are still stuck in the concept phases and are months or years away from reaping the returns.\nWhy is that? Truth be told, it\u2019s because plugging AI into long-established, complex systems and integrating the technology amongst human counterparts is no easy task.\nThis conundrum \u2013 the struggle to derive real business value from AI \u2013 was at the center of a recent panel discussion in which I participated. It was IP Soft\u2019s AI Pioneers Forum, where some of the leading thinkers on the subject pondered how companies can put AI to its highest use and achieve measurable performance increases in less time. Here are a few key points of discussion from those sessions, and some thoughts on how businesses can work through the sticking points.\nClarify your objectives\nInterestingly, panelists agreed that the nebulous nature of AI, and the lack of a clear definition of what AI is and isn\u2019t, may be holding the technology back. In other words, many companies seem to have reached the conclusion that they need to move forward with AI, but key players in the organization have yet to agree on exactly what the technology can and should do.\nMy advice to these companies: instead of trying to understand the technology, start by clearly defining the business outcome you\u2019re trying to achieve. Be specific. For example, go beyond the generic \u201cimprove the customer experience\u201d to create a detailed wish list of how it would play out in real life. It may turn out that your problem could be solved by something other than true \u201cAI.\u201d\nIf you\u2019re convinced that AI can bring the vision to life, recognize the fact that AI isn\u2019t an all-powerful solution. I like to think of AI as a brain. In humans, the brain does the processing and then sends instructions to other parts of the body to carry out an action. It\u2019s similar in technology, in that AI must work in conjunction with \u201carms and legs\u201d that can carry out its orders. Those other \u201cbody parts\u201d could include resources you already have in place such as robotic process automation, conversational agents and, in some cases, even people.\nAssess your maturity\nAnother sticking point the panel discussed was the issue of maturity. That is, organizations have to ask themselves whether they truly have the ability to define, develop and manage their AI investments in a way that will create value. After all, AI isn\u2019t some piece of plug-and-play software you can just flip on and start using. There are significant process changes that need to occur, in technology systems and human employees alike. Security should also be of chief concern. AI\u2019s impact on security can be profound, which means you must determine what controls and protections will be necessary from the very beginning to ensure your sensitive data (sources and outcomes) remain secure.\nWhen there\u2019s confusion and disagreement over how to proceed, it can lead to a case of analysis paralysis. So before charging full steam ahead with AI, companies should realistically assess their own readiness to do so. Thankfully, the IP Soft AI Pioneers Forum is now working to develop a universal AI maturity model that may be helpful to companies in these cases.\nDecipher your data\nFinally, there\u2019s a major challenge in data access and integration. Larger, older companies often have huge stores of customer and transactional data collected over many years. But, as valuable as all that information would be to an AI system, they aren\u2019t able to access it because those legacy systems weren\u2019t built with data mining in mind.\nEarlier this year, my employer hosted a CxO forum where, for example, three companies reported having over 400 legacy systems that needed to be accessed to get the necessary data to feed to their AI engine. They clearly had their work cut out for them.\nGetting back to the airline example, to realize my dream AI scenario they\u2019d need their intelligent bot to be able to access and cross-reference customer records and personal preferences, flight status, reservation systems, airport maps, and more. Figuring out how to merge it all together could unlock powerful AI capabilities, but for now that wealth of information sits largely untapped.\nThere may be no silver bullet answer to this problem, but some out-of-the-box thinking could get things started. For instance, look into the next generation of \u201cdata lakes\u201d (perhaps \u201cdata oceans\u201d) as a way to house those massive volumes of raw legacy data, without the costs of transforming\/restructuring it all into a polished data warehouse format. When it comes to accessing that data, it wouldn\u2019t necessarily require a complete systems integration in the traditional sense. Instead, RPA-powered bots could be sent to fetch bits of legacy data in real time and courier them back to the AI \u201cbrain.\u201d\nFinally, keep in mind that you may be able to leverage your existing ecosystem to make the job easier. Many of the leading vendors for IT, ERP and CRM systems are building AI into their offerings, which may unburden your company from having to commission its own solutions.\nTake baby steps\nWith so many factors in play, deciding to go all-in on AI can become overwhelming and, consequently, stagnating. Instead, approach it through iterations. Take small steps to demonstrate value early in the process. As with any innovation, providing proof of early wins will build support, and that support drives the funding to reach the next level.