Effective AI development requires a top-down approach, while vibrant innovation demands the opposite. How can companies resolve this inherent tension?
In recent years, artificial intelligence has gone from a much-touted pharmaceutical industry buzzword to an absolute prerequisite for any company with serious aspirations for the future. Around the globe, companies are seeking to mine data for insights not easily extracted through traditional research methods. And there is good reason for this focus: With the help of AI, data from a wide variety of sources can be used to achieve more precise diagnoses, to improve the development of new substances, and even to enhance the treatment approval process. Used intelligently, AI can identify mechanisms of action from conventional patient data that could make costly approval studies unnecessary in the future.
At every level of the industry, the pressure to make AI investments reflects the promise of this still largely untapped field. Investors, employees, clients and other stakeholders expect drug firms to implement AI in their strategy. So, AI abstinence is no longer an option for pharmaceutical organizations. But company leaders must confront a problem: The best practices they have adopted to encourage widespread organizational innovation are not well-suited to the world of AI.
Opposing Forces: Top-Down vs Grassroot
In any organization, effective AI projects must draw on a massive data lake that provides the information needed for useful machine learning.
Top down
At first glance, this calls for a top-down approach that leverages the size and reach of an organization to create a data lake with a broad enough scope to power research and analytics projects in every corner of the company. Building a data lake of this size requires coordination among those gathering, inputting and utilizing the data; if different teams gather different types of information or use incompatible code, practitioners can end up with small, disconnected “data puddles” rather than a true data lake. But when a large organization announces a centralized innovation initiative, there is much that can go wrong.
Top-down initiatives tend to be bureaucratic and slow. They are also often out of touch with the real needs of stakeholders on the ground at the local level — an issue that’s particularly troublesome when it comes to AI. For example, a top-down effort that relies on data gathered in the U.S. may prove useless for a company’s offices overseas, where medical practitioners may rely on different measures or metrics in their selection of a course of treatment.
Grassroot initiatives
While AI calls out for a centralized approach, true ongoing innovation often requires the opposite. In our innovation work with companies throughout the pharmaceutical and life sciences industry, we have found that empowering rank-and-file staffers and supporting them in their own grassroots innovation efforts is often the best way for an organization to instill true adaptability and inspire game-changing innovations.
When this approach to innovation is applied to AI, however, there are immediate roadblocks. Employees experimenting with their own AI innovations often hide these efforts from organizational leaders out of a reluctance to get bogged down in requirements, approval processes and bureaucracy. These small projects generally can’t be scaled for wider use, because the underlying data, architecture or parameters of one “data puddle” is often incompatible with those being created by other teams at other locations.
In an attempt to circumvent these difficulties, some organizations seek to insource AI innovation by acquiring well-developed AI startups. The advantages in this approach are obvious: The development risk is outsourced, and the assets of the acquired companies are verifiable and assessable. However, such acquisitions inevitably encounter classic M&A drawbacks. The takeover candidates must be integrated into the core business, which entails costs and personnel losses. The activities of the acquired companies often remain marginal; true organizational transformation is limited; and it often leaves company employees eager to get involved with AI themselves feeling dissatisfied.
A Hybrid Approach
Despite the difficulties inherent in a grassroots approach to AI, there are also some upsides. When an employee detects a need and independently develops a solution, they usually are doing so with a close understanding of their local market and its needs. These projects are usually conceived by tech-savvy experts on-site who know the local regulations, requirements and codes of ethics, who collect their data where it is relevant, and who continuously subject their concepts to a reality check.
The grassroots pioneers who create such projects often experience deep satisfaction, a sense of their own agency within the company, and a belief that they are having an impact on the business. This is a significant benefit, especially given the tight competition for technically skilled employees.
How Companies can support
So how can companies support grassroots innovation while creating machine learning projects that can impact the business at scale? The answer will be different for every organization, but this is what we would suggest as a starting point:
- Be aware of the tension between AI requirements and innovation best practices. Only through awareness can you seek out and adopt strategies that work.
- Have a discourse about this dilemma within your organization. Spread awareness and invite discussion that could yield new ideas and solutions.
- Let employees throughout your organization know that they don’t need approval to embark on an AI project.
- Let them know that they are, however, expected to follow a centralized framework of technical standards. These ground rules should be easy to use and should create consistency in data gathering and architecture, so that any one “data puddle” can be connected to another.
- Request that all AI projects offer company employees full access to their data structure. This will allow practitioners to save time by using previously established practices while also keeping their data compatible.
- Create incentives for employees to experiment with AI within the framework that you’ve laid out. These incentives could include company resources, recognition in the shape of a new title, increased interaction with AI experts, professional development programs, and mentorship programs.
Although it sounds a bit un-techie: successful AI innovation requires wise organizational setups.
