Total Information Mastery: Enterprise AI Strategy, Risks, and Road Map

Think back to the first time you set foot in a library. You may have been overcome by the immense knowledge at your fingertips and a sense of everything you could learn, create, and achieve. Such availability of knowledge is power and potential energy.

Now, expand this access to the world’s expert knowledge and integrate it with that of your own organization; it supercharges the ability to create new knowledge. We call this combinatorial superpower “Total Information Mastery,” akin to kinetic energy, or energy in motion and application. To transform from the potential to the kinetic, however, requires force and, until recently, the world did not have the required force to enable this shift.

For the first time, artificial intelligence (AI) has brought force to the global economy, driving a state change across markets. In the context of such tectonic shifts, leaders should rethink how their organizations win. Meeting this challenge requires understanding the new value-creation opportunities available from the emerging technology and addressing strategic risks that previously did not exist.

Emerging Strategic Risks

There are many new risks that organizational leaders should consider as they select which AI platforms to use and how they wish to incorporate the technology into their organizations. In particular, leaders should understand that just as AI provides new value creation and productivity opportunities, it also represents some entirely new risks which organizations have not needed to consider previously—as such, many of these may go unmonitored without ongoing consideration of how AI continues to evolve.

Sector dynamics. The AI sector is very young, and there are three emerging risks leaders should consider as they select partners and invest, which include the following:

• Changing vendor terms and conditions. Multiple AI providers have unilaterally changed their rules and terms and conditions to claim the right to use user content to train their models without providing customers the ability to opt out.
• Overestimation of consumer AI capabilities. Important limitations remain across all AI technologies. Leaders should understand these and measure them against their intended return on investment.
• Consolidation across the AI sector. There is a strong possibility that large language model performance will approach commoditization, given that they are all largely trained on the same public web content. This, combined with the increasing number of companies building their AI solutions on a few large platforms, means that some sector consolidation is inevitable.

AI models. There are limitations present in each AI model that organizational leaders should be aware of to position their usage accordingly, including the following:

• Problems memorialized in code. Errors and sleeper-cell backdoors may be embedded within an AI model if the training data contain such flaws. Resulting problems are difficult to correct once the model is deployed and can potentially perpetuate historical inaccuracies or biases or enable cyberattacks.
• Monolithic large language model architectures. Large language models are often designed as monolithic entities that are not easily adaptable to specific tasks or domains, limiting their flexibility and increasing the time and resources required to apply them to enterprise-specific use cases.
• Model collapse. AI models can experience model collapse, where they suddenly fail to continue learning or start generating less diverse or lower quality outputs. In certain cases, models may even consume their own AI-generated outputs, leading to a state of virtual mad cow disease where reliability craters.

Security. The structure of AI models presents security risks that were less urgent with earlier software, such as the following:

• Universal and transferable adversarial attacks. Adversarial attacks involve making subtle changes to inputs processed by AI models to cause them to make errors. These attacks are particularly concerning because they can often be applied across different models, undermining the reliability of AI systems more broadly.
• Poisoned training data. Introducing subtly corrupted data into the training set can lead to a model that behaves normally on most inputs but fails catastrophically or behaves maliciously under specific conditions.
• Complete enterprise information systems compromise. Traditional cyber threats target specific systems, but AI-generated attacks have the potential to attack far broader swaths of enterprise infrastructure and shut down critical operations. Organizations will need to plan for full-enterprise backups to address such existential threats.

A New Strategic Opportunity: Total Information Mastery

Directors should define a strategy for Total Information Mastery to help address these risks and drive kinetic change in their organization’s ability to use data to create value. Total Information Mastery includes the ability to sift through massive amounts of data across sources, end-to-end knowledge integration, advanced decision-making support, and the generation of new knowledge, products, and intellectual property (IP). Capturing this value requires a new strategy and road map that incorporates bold goal setting, enterprise AI strategy, organizational design, and investment.

Strategy and goal setting. Directors should work with management to develop a portfolio of strategic and tactical targets for the company’s AI investments, ranging from productivity and efficiency targets to setting bold innovation and transformation goals that consider the largest problems facing the company’s markets. Consumer and functional AI tools offer productivity and automation benefits which, while important, may not be sufficient if competitors succeed in commercializing new AI-powered business models.

Enterprise AI strategy. An organization’s AI strategy should be designed to deliver total information awareness from the start. This includes sourcing and integrating information across multiple external content types, and multiple content sources as relevant (e.g., public and unverified, public and trusted, published and licensed, proprietary, and personal). Integral to this design is a focus on data accuracy: To a large degree, leaders should expect that the public Internet will become less reliable as more AI-generated content pours into it. Organizations will need to validate data quality directly.

These external content sources should be integrated with the organization’s internal data available across enterprise systems, such as those provided by third-party vendors and internal cloud, to enable new knowledge and processing capabilities. Finally, an organization’s AI architecture should be able to access multiple infrastructure types, including on-premises, edge, mobile, personal computers, servers, solution providers, and public and private cloud.

Business leaders should challenge their technology leaders to deploy an evolving portfolio of AI models that can deliver different benefits. Large and small language models, image processing, recurrent and convolutional neural networks, sequential data, reinforcement learning from human feedback, and joint-embedding predictive architectures will all power enterprise applications. In particular, retrieval-augmented generation holds promise in the context of total information awareness as it can integrate, optimize, and generate across multiple sources.

The most successful organizations will be those that enable end-to-end AI operability across the enterprise as they will have the ability to deploy advanced models through more extensive and reliable data sets and have near real-time, full-system auditability for risk management. This total information awareness will be difficult, if not impossible, to accomplish if the enterprise brings in myriad disconnected and component AI parts as there is no way to assure quality across them.

Organization and Culture

The purpose of an organization has always been to optimize productivity across human and asset portfolios. Total Information Mastery presents a next-generation opportunity to maximize the contributions of a company’s talent’s expertise and to engage employees in the creation of these next-generation capabilities. New organizational design is required. To achieve this, companies can do the following:

• Develop models that allocate more weight to the contributions of internal experts, as the product contributions of lead engineers carry more authority than unverified public information.
• Develop smart incentives that can evolve to accommodate the massive productivity gains already occurring across functions. Productivity gains ranging from 30 to 300 percent are already possible in some jobs, heralding an augmented workplace capability in which talent, enabled by new tools, are expected to significantly outperform historical norms.
• Design incentive structures and support tools to make AI usage across the enterprise non-adversarial, such as by incentivizing the development and sharing of new ideas—similar to the residuals model from the entertainment industry—which better reward those who contribute to enterprise value. Create digital twins for these individuals to reduce their cognitive load and enable them to focus on higher-value work.
• Design new job types, such as virtual analyst roles, that apply AI to the organization’s financial and public disclosures, brand campaigns, and other external engagement to anticipate and optimize the market’s response.