Many companies are still not seeing significant impact from their AI efforts. Some experts say this may be because they\u2019re not embracing something called \u201corganizational learning.\u201d\nIt\u2019s not enough to use AI to optimize a business process \u2014 for example, to make better predictions or automate a manual task. Enterprises need to go one step further, to take lessons learned from their AI projects and use them to transform their organizations.\n\n[ Cut through the hype with our practical guide to machine learning in business and find out the10 signs you\u2019re ready for AI \u2014 but might not succeed. | Get the latest insights with our CIO Daily newsletter. ]\n\nWhile most, if not all, organizations would say they learn from their successes and their failures, few have formal processes to embrace these learnings and promulgate them throughout the enterprise \u2014 especially when it comes to the use of AI. As a result, only 11% percent of companies saw significant benefits from their AI initiatives in 2020, according to a recent report from MIT Sloan Management Review conducted in collaboration with Boston Consulting Group.\nTake, for example, scoring loan applications, much of which involves tedious data entry done manually by loan officers. Using AI or machine learning can dramatically optimize the process, reducing costs and the need for as many loan officers on staff. But enterprises can only save so much money, and employees are reluctant to get behind projects that could cost them their jobs.\nMeanwhile, AI can also be used to glean new insights from that same loan application data. A bank could discover underserved customer segments, for example, that could lead to a dramatic expansion of business. Or a bank could discover that people were afraid to apply for loans because of worries about damaging their credit ratings, says Sam Ransbotham, professor of information systems at Boston College\u2019s Carroll School of Management and coauthor of the MIT Sloan report. Offering them an opportunity for a no-risk assessment that doesn\u2019t affect their credit rating could change that.\n\u201cThat\u2019s not just automating the loan process; that\u2019s fundamentally changing the loan process,\u201d he says.\nThere\u2019s no limit to growth potential, and employees can get behind a new technology that offers more opportunities for interesting work.\nThat\u2019s an angle that\u2019s important for CIOs to be aware of, Ransbotham says, since they typically pay more attention to efficiency. \u201cSome CIOs are more service-oriented,\u201d he says. \u201cThey\u2019re about getting the costs down for their IT operations. There can be a tendency to automate what we\u2019re already doing, versus doing something different.\u201d\nIn their survey of more than 3,000 respondents, MIT Sloan and BCG identified several factors that helped companies move into the 11% who reported \u201csignificant financial benefits,\u201d including sharing knowledge between humans and AI, incorporating AI into the overall business strategy, moving beyond using AI for simple automation, and finding ways for humans and AI to work together so that AI augments human work and human work augments AI.\n\u201cWhat we found is that when people do these steps that are organizational learning oriented, they can increase the likelihood of being in that 11% group by almost 80%.\u201d\nBetting big on AI\nIn late January, Johnson & Johnson announced its COVID-19 vaccine, a single-dose vaccine that requires normal refrigeration, not a deep freeze. According to J&J, its vaccine is 66% effective overall, but 85% effective in preventing severe cases, and 100% effective in preventing death.\nThe vaccine, says J&J CIO Jim Swanson, wouldn\u2019t have been possible without AI. Eight or nine months ago, it took two weeks to make a batch of the vaccine, he says. Now, it\u2019s two batches in one week \u2014 a fourfold improvement.\n\u201cWe used AI to improve everything from our fermentation process, to our flow of yield,\u201d he says. \u201cThere are a bunch of insights, and all the pieces added up to the outcome.\u201d\nCooperation across several domains of expertise also accelerated the process, he says. \u201cWe pushed hard on this idea of the bilingual data scientist. One who really understands research and development or the supply chain.\u201d\nBut while leveraging AI to accelerate COVID-19 development is what\u2019s making headlines right now, J&J is also using AI to create entirely new business opportunities. For example, it is using AI and ML to examine retinal scans to determine whether a patient has glaucoma.\nThe company also makes surgical robots. \u201cYou get higher precision and better procedures,\u201d Swanson says.\nAnd it goes even further than that. The end goal is to improve patient outcomes, so the company is also looking at pre-op and post-op processes. \u201cYou can use AI so that the right patient gets the right procedure, and so that you can best support their recovery,\u201d he says. \u201cNow you have an end-to-end view of the patient \u2014 that creates a whole new set of opportunities.\u201d\nThe same approach is being deployed across the board at J&J, Swanson says. Take its Avena skin care line. AI allows consumers to take a picture of their skin to get a personalized product recommendation.\nAnd then organizational learning kicks in, as J&J uses those images to find out what skin issues people face. \u201cNow all of a sudden you\u2019ve accelerated your product offering,\u201d he says. \u201cYou have a data feedback loop that continues to create relevant products.\u201d\nThat feedback loop is dependent on having the right data infrastructure in place, one that supports privacy and security, so that data can be democratized throughout the company. \u201cIf you can\u2019t share data securely, you can\u2019t share it,\u201d Swanson says, which might mean de-identifying data, for example, to focus on phenotypes \u2014 age groups with certain conditions.\nThe final piece of J&J\u2019s organizational learning strategy involves the growth of collective AI expertise. \u201cIf you\u2019re not comfortable with using data, you can\u2019t leverage it,\u201d says Swanson. \u201cSo we have forward-thinking R&D scientists, commercial people, supply chain people. We\u2019ve set up a data science council that I co-sponsor with the head of R&D and we\u2019ve made the decision to decentralize AI into our business.\u201d\nMore importantly, J&J\u2019s AI strategy has sponsorship from the top. \u201cWe\u2019re making AI and technology the core of our company \u2014 it\u2019s not something you can do on the side,\u201d he says.\nSpreading the gospel\nLike J&J, companies that are the most successful with their AI initiatives don\u2019t limit AI to small groups, says Anand Rao, partner and global AI leader at PricewaterhouseCoopers. Instead, they embed AI throughout the company so it is used even by those employees who don\u2019t have a technical or analytical background.\n\u201cThe challenge most of the time \u2014 and this is when companies don\u2019t get the ROI \u2014 is because they\u2019re not trained appropriately, coached, and managed,\u201d he says. \u201cYou don\u2019t want just the one individual or small group learning from this, but the entire organization learning from this.\u201d\nIt helps to have people who are \u201cmultilingual,\u201d he says, meaning those who understand the business side, subject matter domain, software, and the AI algorithms. \u201cOr find a team that can work together and do this.\u201d That\u2019s one of the biggest challenges, he adds. \u201cIt\u2019s difficult to get people with different mindsets to work together.\u201d\nHuman-machine collaboration\nAnother company that\u2019s taking the organizational learning principle to heart is Genpact, a global professional services firm that started out as the business process unit of GE. Genpact, which spun out in 2005, now has nearly 100,000 employees and reported $3.5 billion in revenue in 2019.\nWhen the pandemic hit, revenue dropped significantly and the company was looking at possibly having to lay off 10,000, as many of Genpact\u2019s clients were in hard-hit sectors, says Gianni Giacomelli, the company\u2019s chief innovation officer.\u00a0\n\u201cInstead, we were able to match them with new demand and retrain them in real time,\u201d says Giacomelli, who is also the company\u2019s head of learning and development. \u201cSometimes it took just a couple of weeks to get them retrained and into new jobs. We actually managed to grow compared to our peers, even during COVID-19.\u201d\nThe retraining effort was made possible by two separate uses of AI at the company. First, Genpact uses process mining, natural language processing (NLP), and network analysis to figure how things got done, to identify exceptions, and to figure out who at the company had what skills and what domain expertise.\nThis information helped the company place employees in new jobs, and once an employee began a new role, AI systems enabled them to get up to speed quickly by explaining the process for specific tasks, or by connecting them to relevant experts.\n\u201cThis enabled us to react a lot faster to the conditions that were thrown at us,\u201d Giacomelli says.\nIn the past, knowledge management had its share of difficulties. Five years ago, the failure rate of these programs was about 50%, according to the KM Institute. But due to significant improvements in NLP and other AI technologies, the situation has changed dramatically.\n\u201cOver the past two, three years, the quality of the creation of ontologies that machines can do by themselves is so much more accurate,\u201d Giacomelli says. \u201cAnd what you get back is much more precise.\u201d\nAI can find organizational knowledge that is located in documents, in business processes, and with people.\nAt Genpact, AI isn\u2019t the sole domain of the IT department. And that\u2019s a key difference between companies that are able to see significant ROI from AI, and those who don\u2019t, says Kathleen Featheringham, director of AI strategy and training at Booz Allen Hamilton.\n\u201cAI is the fourth industrial revolution,\u201d she says. \u201cIt wholesale changes the game. It\u2019s not just an IT problem \u2014 all the roles are evolving.\u201d\nAI-powered enterprise transformation involves re-evaluation of performance goals, of training objectives, she says. And it needs to be connected to an organization\u2019s vision and mission. \u201cIn my experience, when AI wasn\u2019t tied to the vision and mission, people actually became very resentful,\u201d she says.\nCreating new lines of business\nOne of the principles of organizational learning is that AI is used to augment employees, to work alongside them and complement their skills.\n\u201cIf you have a collaboration between what machines do well and what human intuition and knowledge does well, that\u2019s where you get a huge business benefit,\u201d says Judith Hurwitz, president and founder at Hurwitz and Associates and author of ten books on leadership, technology, and analytics.\nSoftware development company Globant is doing just that with its AI-powered augmented coding. It uses NLP to enable developers to search for code by functionality, allowing for shorter learning curves and faster, more accurate development. The system also automatically generates documentation and auto-completes code from context.\n\u201cIt\u2019s not going to replace the importance of developers,\u201d says Nicol\u00e1s \u00c1vila, Globant\u2019s CTO of North America. \u201cThe technology isn\u2019t there yet, but it also isn\u2019t what we stand for.\u201d\nInstead, the augmented coding technology does a lot of the heavy lifting around routine work. \u201cIt gives [the developers] a baseline to start with that\u2019s contextual to their particular problem and their particular client. We\u2019re just using their time more effectively,\u201d he says.\nGlobant initially invested in AI capabilities five years ago, \u00c1vila says, and made the training universal \u2014 not just for developers, but even, to some degree, for people in HR or purchasing or other departments. \u201cYou had to have an idea of the capabilities of AI, at least at a high level, so that every employee can be identifying opportunities.\u201d\nThis has evolved into AI-powered applications in other areas of the company, including in recruitment and retention.\nAutomated coding grew out of experiments applying NLP to programming languages in early 2019. That effort blossomed into a set of internal development tools \u2014 and then into a commercial product.\n\u201cWe definitely see this as a growing business opportunity,\u201d says \u00c1vila.