Key Considerations for a Successful AI Strategy

Blackford Analysis October 24, 2021

The first step in designing an AI strategy is to identify the most prolific and impactful problems and priorities in your organization, and then align potential AI initiatives accordingly. In the AI evaluation process, it can be appealing to seek out algorithms with distinct and specific use-cases that neatly align with singular departmental goals. For example, improving clinical report consistency within the radiology department is an explicit goal with a clear outcome that can be addressed with a singular AI algorithm. However, if you were to zoom out and look at the big picture, your organization’s overarching goal might be to grow referrals through increased patient and referrer satisfaction. This could lead you towards a series of algorithms that target the most common exams and pain points to decrease patient wait times, optimize exam throughput, and improve radiologist productivity and clinical confidence.

This approach to AI strategy development not only drives to your larger organizational objectives, but ultimately unburdens radiologists, enabling them to complete reports more efficiently and confidently. So, when designing your AI strategy, keep this big picture approach in mind and consider:

How your strategy can help achieve broader organizational goals while simultaneously delivering value across the entire medical imaging enterprise, and
How to effectively measure and monitor a chosen solution to ensure it is meeting the objectives that were previously identified.

Identifying goals and objectives Clear and measurable objectives should be defined that are based on the outcomes that your organization hopes to achieve with AI, and the role that radiology and other imaging departments will play in realizing them. This requires a solid understanding of who the consumers of AI will be, what drives them, and the key challenges they face. To keep your objectives assessable and ensure you can effectively evaluate outcomes:

Ensure your objectives are outcome-based: Outcomes Based Specifications (OBSs) focus on the results and value to be achieved according to target clinical, operational, and financial criteria rather than rigid prescribed feature sets.
Consider intangible metrics: Outcomes such as improvements in radiologist confidence are important to track as they can have a big impact on the bottom line. While difficult to measure, they can often be quantified using downstream metrics such as productivity, satisfaction scores, and turnover.
Set clear timelines and milestones: Interim goals can be used to track your progress, highlight successes, and course correct as necessary, which not only keeps the AI project on track, it also builds momentum and garners buy-in from radiologists and other stakeholders.

Invest in monitoring tools To assess whether AI is meeting the goals and objectives that you've defined in your program, it is essential to be able to benchmark and measure current versus future performance. When evaluating AI solutions, look for tools and mechanisms that can assist in benchmarking and measuring AI's contributions to your OBSs from its initial implementation and following its continual adoption. An effective AI solution should include tools for ongoing performance monitoring and display at-a-glance which AI models are most effectively used. To inform the clinical and operational value algorithms have achieved, a robust AI strategy should also include:

Key performance indicators (KPIs) and the consideration towards the resources required to manage and track them.
Input from multidisciplinary stakeholders who are best positioned to assess whether desired outcomes are being met.
A mechanism to analyze AI's effectiveness in terms of quality and performance.

Are you interested in learning more about how to build an effective AI strategy? Click here to download our AI Buyer’s Guide, which discusses the challenges, considerations, and tactics for developing an AI strategy, building a business case, and evaluating AI models and vendors.