A mid-sized agricultural technology company was at a loss: They had a team of agronomists across the globe helping farmers optimize their crop cultivation, but the teams were siloed and stagnant.
Agronomists in one region would often ask questions about issues that their colleagues in other regions could address, such as, “How will this unexpected weather pattern in my region affect weed growth?” But unless the agronomist knew the right person personally, it took a lot of time to get the question answered. Most people just didn't care.
For a while, AI, especially its newest form, generative AI, was touted as an easy fix. But simply introducing AI into the enterprise by asking employees to use OpenAI's ChatGPT or Microsoft Copilot “out of the box,” as most companies have done in the past, won't work.
While generative AI can answer a wide range of human-generated questions, it can't answer questions that require deep expertise, nor can it connect people to the necessary in-house expertise. Agronomists are still out of luck, unless their colleagues have actually written and distributed papers about how certain weather patterns affect weed growth.
Generative AI has captured the hearts of many in the business world. As the initial excitement for 2022 fades, executives and entrepreneurs continue to explore how their companies can realize the vast potential of this groundbreaking technology. But as the agritech example shows, leaders must also address structural barriers to AI adoption.
Three Key Stages of AI Adoption
Our comprehensive research on AI adoption and experience from executives who have implemented AI systems across a range of companies leads us to believe that a piecemeal approach is likely to fail. Based on our research and experience, we recommend that organizations such as the agritech companies listed above should follow three key phases of AI adoption to reap the full benefits of AI:
First, it's important to understand the current capabilities of generative AI. ChatGPT can understand natural human language and also has some visual analytics capabilities. Instead of typically using software or other constructs to transform queries into language that a machine can understand, users can ask questions or give instructions in natural language and get (usually) reasonable answers in the same language within seconds. This is phase 1, which is important, but not sufficient to address the most pressing business problems: these problems require accurate, concise, and consistent answers that current AI just can't deliver yet.
Second, in a business context, it's important to understand the questioner's intent. What does the person typing the question really want to know? What is the context of the question? What would be most useful to the questioner? This is phase two, where we need to go beyond current generative AI models and inject some expert knowledge in the form of business-relevant rules.
Some examples of such composite AI systems are beginning to emerge. As an example, consider a car rental business, where managers ask for a comparison of gasoline and electric cars, not just for knowledge purposes, but to evaluate the relative merits of these vehicles in terms of a likely purchase.
The AI first checks if the car purchase is a genuine intent, then suggests buying a hybrid car, taking into account the low density of nearby charging stations (or, as Hertz too late discovered, the higher-than-expected maintenance costs of electric cars.) To do this, it requires layers of algorithms that reason over current data of large language models about user activities and interactions.
The third stage for companies to realize the full value of AI requires senior executives to drive AI adoption and alignment across the organization. Executives have long talked about breaking down silos, but most organizations are still divided into largely siloed groups. Leaders have long lamented, “If only we knew what we know.” AI alone won't solve this problem; it will only make each silo more efficient.
To realize the full potential of AI, companies need to rethink their organizational structure. They need to envision a central AI system that reaches every part of the organization, taking unstructured data from across the company and sending it to the one group that needs it now. Instead of just talking about broad collaboration, companies can put AI at the center. This idea is now beginning to gain widespread agreement among executives at leading companies such as Snowflake and Accenture.
So AI will not only make existing collaboration more efficient (for example, reducing the need for cumbersome company-wide announcements or large knowledge-digging meetings), but it will also foster collaboration on issues that most people don't currently discuss with their colleagues at all. However, there are some caveats.
AI has great potential, but it won't work well for every interaction. Employees remain well-suited to handle interactions that build trust, relationships, and judgment, rather than transactional demands for knowledge. AI may play a supporting role, but only humans can build true trust.
Second, as companies establish AI, they will need to rethink how they measure and reward individual performance: Companies will need to incentivize people to make their emerging knowledge available to central AI platforms.
Thought leaders describe AI as a revolutionary change comparable to electricity. While news about AI's miraculous feats flood our newsfeeds every day, businesses struggle to adopt AI to create real, meaningful value for their organizations. We believe a systematic, phased approach to AI adoption that starts from an information systems-centric approach and gradually drills down to the underlying organizational structure can accelerate true AI-driven value creation in organizations.
Ram Bala Samvid AI is an associate professor of AI and analytics at Santa Clara University, co-founder and chief AI scientist at Samvid AI, a researcher, leadership coach, and entrepreneur. Over the past 20 years, he has focused on applying AI to pricing, marketplace design, and supply chain management.
Natarajan Balasubramanian Professor of AI Strategy at Syracuse University Professor and leading researcher on the adoption of AI and machine learning in organizations.
Arun Rao CEO and Co-Founder of Samvid AI. An entrepreneur and veteran Silicon Valley executive, Samvid previously held technology leadership roles at The Gap, CH Robinson and Flex.
Amit Joshi Professor of AI, Analytics, and Marketing Strategy An award-winning professor and researcher at IMD, he has extensive experience in AI and analytics-driven transformation in industries including banking, fintech, retail, automotive, telecommunications and pharmaceuticals.
