AI-enhanced Innovation and Creativity

Why Innovation?

The pace of change can be overwhelming for innovation teams. But, we can’t stand still or we will be outpaced in the marketplace. We’re driven to innovate not just to stay apace with the competition, but because creativity is an essential part of human nature. It drives our search for “new strategies to diagnose, treat, and prevent diseases.”

Generative AI is emerging as a powerful accelerant to the innovation process. It can improve your process in three key ways: capacity, speed, and quality.

A growing body of research, shows that it delivers impact across the innovation process: "Generative AI may be a game changer..., as the delegation of tasks to an artificial agent can result in faster iterations and reduced costs." The result is a new "transformative phase in which individuals and organizations increasingly turn to ai systems…"

Indeed, a May 2024 survey by McKinsey found that 65% of companies are using Generative AI in “at least one business function.” A full 23% are using GenAI in product or service development, and 10% listed innovation as their most common use-case.

Innovation Process

Creativity is the process of generating new, novel ideas. Innovation is the process of successfully implementing those novel ideas within your organization. AI and creativity can be fused to enhance innovation! Common methodologies, including Design Thinking, TRIZ, or Double Diamond, all follow a common pattern:

Problem identification - where the team recognizes gaps and clearly articulates problem statements, goals and objectives for solving them.
Research and ideation - the team gathers and analyzes data relating to the problem space, and generates potential ideas.
Concept development and testing - the ideas are refined and tested in an iterative fashion.
Implementation - planning and launch of the selected idea, performance monitoring and scaling.

Generative AI can be applied at all stages of the innovation process. You can use it to explore the problem space to identify trends and reveal insights. Ideation and concept evaluation are fertile grounds for GenAI. Indeed, it may even already "outrival human idea generation regarding overall idea quality.”

And GenAI can greatly speed up prototyping - enabling your team to rapidly generate high-fidelity concepts, mockups, and code. It has even been used to design a new microchip, where “... the language model seems to be a ‘force multiplier’, allowing for rapid design space exploration and iteration” in the development of a novel processor.

The Double Diamond is a common innovation framework, with successive expansion and narrowing of focus. Using GenAI essentially expands the Double Diamond across a broader problem and solution space, facilitating the generation, consideration, and development of additional ideas or concepts.

“By expanding the problem and solution spaces… language models create an opportunity to access and generate larger amounts of knowledge, which in turn results in more possible connections of problems and solutions.”

The original double diamond framework (above), as conceptualized by Marion and Fixson (2019), and the artificial intelligence-augmented double diamond framework (below). Source: Bouschery, Blazevic and Piller.

Problem Identification

The first stage of the innovation process is selecting and articulating the problem to address. Indeed, many innovation activities fail because they attempt to solve the wrong problem. Leveraging AI for sentiment analysis of customer data at scale can give the team a clearer understanding of your customer’ needs at scale.

Ideation

During Ideation, you can prompt the AI system to review a broader scope of knowledge collected during the exploration phase, and use it to generate novel concepts that align to “a particular context, user segment, or user needs.”

To assess which ideas are worth developing, innovation teams often score or rank their ideas to facilitate filtering out the less promising ones. One methodology is to score them on Novelty, Benefit to the Customer, and Feasibility. You can then calculate a Quality score by multiplying those scores, so Q=(N*B*F).

And the stronger you can drive alignment and a shared mental model of the vision from the outset, the closer to it your outcomes will be.

Concept Development

Research shows that when we invite Generative AI into our process and treat “algorithms rather as members of an innovation team,” as opposed to a replacement for it, we improve the quality of our concepts, and radically increase the pace of innovation. It helps us avoid focusing on the wrong problem. Indeed, alignment around problem selection is one of the biggest challenges in innovation. "Augmenting human innovation teams with AI hence calls for investigating many aspects of jointly working with AI rather than only human colleagues."

Jan Joosten, and his team at the BeLab-Behavioral Studies and User Experience Laboratory in Germany, ran a blind test to compare the quality of ideas generated by AI with those generated by professionals.

They recruited a European packaging company, and assigned them two tasks: design new packaging solutions that added value to their customers, and design new packaging solutions that aligned with their ESG goals. The researchers also prompted ChatGPT 3.5 with the same tasks. The human designers generated 43 concepts and the GPT generated 52. The researchers were then graded by the company’s director of innovation in a randomized order. In this study, humans scored lower in both novelty (0.0039) and customer benefit (0.0028) with a negligible difference in feasibility (0.2922). Overall quality score: 0.0039.

Implementation

But that’s not the end of the story. When humans and AI work together, we get better results than humans or AI alone. By using AI, we can “add an entirely new capability” to our teams. This new capability “puts nontechnical users in a position to close the gap between conceptual work (i.e., ideas and concepts) and early look-and feel-like prototypes that can be tested with users.”

And AI can be applied along the entire innovation pipeline. When defining the problem, you can use it to quickly extract knowledge from a vast amount of data.

"NLP models and complementary algorithms can uncover commonalities, filter essential information from non-essential information, predict outcomes and typologies based on semantic text representations, and even generate new knowledge with little human effort."

You can perform sentiment analysis with it to generate unique insights. Once you’ve identified your problem, you can then augment your ideation with AI. Again, another space where humans and AI together can form “superminds” such as in Prediction Markets. And then it can help you create prototypes more quickly and efficiently. AI in innovation is a game-changer!

AI Collaboration Strategies

When collaborating with Generative AI, there are several strategies you can use to get the best results.

1. Use the CARE framework for your prompt: Provide the background context, Explain the action you want the LLM to take, Your expected result, and a clear example:

Content: We need to optimize our administrative processes to reduce patient wait times and paperwork.
Action: Propose solutions for automating and digitizing administrative tasks.
Result: Efficient workflows that minimize manual input and streamline patient check-in and record-keeping.
Example: Introducing a self-service kiosk system where patients can check in and update their information without needing to fill out paper forms.

2. Few-shot - this technique includes several examples of the results you want in your prompt - to give the LLM a clear picture of your desired outcomes.

3. Chain of thought - Asking the LLM to explain the process is another technique that can lead to better results, rather than leaping straight to the output.

4. LLMs excel at interpreting natural language prompts, but that doesn't mean you can’t use structured data as well. You can use Markdown Tables or JSON both in your input and your output from the AI.

5. When you find a prompt or workflow that works, share it with your team. Create a prompt library, or create a custom GPT to increase efficiencies and knowledge sharing.

Reduce Risk with Humans in the Loop

Benefits aside, there are still legitimate barriers to wholesale adoption. Due to the “black box” of their operation, it can be hard to really trust the output of a LLM. If we don’t have transparency around what data the model was trained on, we can miss bias in the output, or encounter hallucinations. But that is starting to change.

Researchers at Anthropic released a paper describing the first “mapping” of the patterns of neuron activations inside the model directly to human-interpretable concepts. They found related terms were situated in close proximity - terms like Golden Gate Bridge, were close to Alcatraz Island, Ghirardelli Square, the Golden State Warriors, California Governor Gavin Newsom, and others.

Until there is full transparency into how these models operate, it falls to us to manage and mitigate their risks - especially in a healthcare context. "While machines can draw attention to trending technologies, customer pain points, or particularly novel technologies, the final decision on how to proceed remains the responsibility of human innovation experts."

Take the time to craft an AI Risk policy for your organization - it can guide you on the best ways to incorporate AI into your products and operations.

These are still stochastic systems with a degree of randomness of the answer, and that is why we keep humans in the loop - augmenting human cognition rather than replacing it outright.

A Call to Action

You may have heard that AI is the “worst it will ever be.” While that remains to be seen, there is still no better time to start working with AI than now. Look at your workflows, and where you can augment your team with these generative tools.

Embrace change and design your process to avoid lock-in as LLMs continue to leap-frog each other in capability. Be flexible so that you can adopt more powerful or more affordable models as they become available.

Leverage the temperature setting to manage the creativity of the LLM’s output space. Dial it up in low-risk ideation sessions, but lower it when working on more high-stakes or patient-facing concepts.

Start small with a low-stakes challenge and augment your team using these techniques. You’ll be surprised at how fast you can move. Buckle up!

Applying Generative AI with the Double Diamond

The examples below offer sample use cases and prompts for applying Generative AI to the Double Diamond process.

Discover

Problem Articulation

Activities

User Research: Conduct interviews, surveys, and observations to understand user needs and behaviors.
Market Research: Analyze market trends, competitors, and industry insights.
Stakeholder Interviews: Engage with stakeholders to gather insights and requirements.
Ethnographic Studies: Observe users in their natural environments to gain deep insights.
Contextual Inquiries: Understand the context in which users interact with the product or service.

Use Case 1: Analyzing Patient Experience Data

Summary: Generative AI can analyze large volumes of patient feedback from surveys and interviews to identify key pain points and areas for improvement in patient experience.
Sample Prompt:

Context: We conducted a survey of 75 families who recently delivered in our maternity ward and are looking for opportunity areas to improve our patient experience.
Action: Analyze the survey data to identify opportunity areas for improving the patient experience.
Result: A summary of patient feedback and suggested opportunities to improve the patient experience.
Example: Overall, patients had a positive experience during their time in the maternity ward. Opportunity Areas to improve the customer experience include:

Pre-Delivery Information: How well the hospital prepared families for the delivery process through prenatal classes, informational materials, and consultations.
Language Services: Availability and quality of language translation and interpretation services for non-English speaking families.
Continuity of Care: Satisfaction with follow-up care and support provided after discharge, including home visits, follow-up appointments, and availability of lactation consultants.

Use Case 2: Generating Research Themes

Summary: AI can identify common themes and patterns in qualitative data from patient interviews and observations, aiding in the discovery of key insights.
Sample Prompt:

Context: We have raw data from patient interviews regarding their experiences in our oncology department.
Action: Identify common themes and patterns in the interview data.
Result: A list of key themes and patterns observed in the patient data.
Example: Common themes include:

Communication: Patients expressed the need for clearer communication regarding treatment plans.
Emotional Support: Many patients highlighted the importance of emotional and psychological support during their treatment.
Waiting Times: Several patients were concerned about the long waiting times for appointments and treatments.

Use Case 3: Market Research Analysis

Summary: AI can analyze market research reports and competitor data to extract key trends and insights relevant to patient care services.
Sample Prompt:

Context: We have gathered market research reports and competitor analysis documents related to telehealth services.
Action: Extract key trends and insights from these reports.
Result: A summary of market trends and competitor insights in telehealth services.
Example: Key trends include:

Growth in Demand: There is a significant increase in demand for telehealth services, particularly in rural areas.
Technological Advancements: Competitors are leveraging AI and machine learning to enhance telehealth platforms.
Patient Preferences: Patients prefer telehealth services that offer seamless integration with their electronic health records (EHRs).

Define

Problem Selection

Activities

Affinity Diagramming: Organize and group research findings to identify patterns and themes.
Personas: Create personas to represent key user segments.
Customer Journey Mapping: Map out the user journey to identify pain points and opportunities.
Problem Statements: Clearly articulate the problem to be solved.
Design Briefs: Develop a clear and concise design brief to guide the development process.

Use Case 1: Creating Patient Personas

Summary: Generative AI can synthesize research data to create detailed patient personas that represent different segments of the patient population.
Sample Prompt:

Context: We need to create patient personas based on our research data from the cardiology department.
Action: Generate detailed patient personas that represent our key patient segments.
Result: Detailed patient personas with demographics, health goals, pain points, and behaviors.
Example:

Persona 1: John, 60, retired, seeks comprehensive information about managing his heart condition.
Persona 2: Maria, 45, working mother, needs flexible appointment times and support for stress management.

Use Case 2: Defining Problem Statements

Summary: AI can help articulate clear and concise problem statements based on patient feedback and research data.
Sample Prompt:

Context: We need to define the problems identified in our patient experience research for the emergency department.
Action: Articulate clear and concise problem statements.
Result: Well-defined problem statements that guide the design process.
Example:

Problem Statement 1: Patients are experiencing long wait times before being seen by a physician.
Problem Statement 2: There is a lack of clear communication regarding treatment progress and next steps.

Use Case 3: Creating Patient Journey Maps

Summary: AI can develop detailed patient journey maps highlighting key touchpoints and interactions.
Sample Prompt:

Context: We have identified key patient touchpoints and interactions in our outpatient surgery center.
Action: Develop detailed patient journey maps.
Result: Comprehensive patient journey maps highlighting patient interactions and pain points.
Example:

Pre-Surgery: Patients often feel anxious and require more information about the procedure.
During Surgery: Patients appreciate real-time updates for their families.
Post-Surgery: Patients need better follow-up care instructions and easier access to support services.

Develop

Concept Generation

Activities

Brainstorming: Generate a wide range of ideas and solutions.
Sketching and Wireframing: Create initial sketches and wireframes of potential solutions.
Prototyping: Develop low-fidelity and high-fidelity prototypes to visualize solutions.
User Testing: Test prototypes with users to gather feedback and iterate.
Co-creation Workshops: Collaborate with stakeholders and users to refine solutions.

Use Case 1: Generating Care Model Ideas

Summary: AI can generate a wide range of innovative care model ideas to enhance patient experience based on identified problem areas.
Sample Prompt:

Context: We need a wide range of care model ideas to improve patient experience in our primary care clinics.
Action: Generate creative care model ideas based on our problem statements.
Result: A variety of innovative care model concepts.
Example:

Idea 1: Implementing virtual care coordinators to assist patients with managing their appointments and follow-up care.
Idea 2: Creating a patient mentoring program where experienced patients support new patients.

Use Case 2: Creating Service Line Prototypes

Summary: AI can help create low-fidelity prototypes for new service lines, allowing for visualization and initial testing of concepts.
Sample Prompt:

Context: We need to develop low-fidelity prototypes for a new telehealth service line.
Action: Generate initial sketches and wireframes for potential service line solutions.
Result: Low-fidelity prototypes that visualize different service line solutions.
Example:

Prototype 1: A user-friendly telehealth platform with integrated EHR access.
Prototype 2: A virtual waiting room experience that reduces perceived wait times.

Use Case 3: Refining Care Models through Patient Feedback

Summary: AI can iterate on care model prototypes based on patient feedback to improve their effectiveness.
Sample Prompt:

Context: We have patient feedback on our new chronic disease management care model.
Action: Iterate on the care models based on this feedback.
Result: Improved care model prototypes that better meet patient needs.
Example:

Feedback: Patients found the scheduling process cumbersome.
Iteration: Simplify the scheduling process by integrating it with the patient portal.

Deliver

Concept Selection & Implementation

Activities

Final Prototyping: Develop high-fidelity prototypes and final designs.
User Acceptance Testing: Conduct final testing with users to ensure the solution meets their needs.
Implementation Planning: Develop a plan for implementing the solution, including timelines and resources.
Launch: Roll out the solution to users.
Evaluation and Feedback: Gather feedback post-launch to assess the solution’s effectiveness and identify areas for improvement.

Use Case 1: Finalizing High-Fidelity Care Model Designs

Summary: AI can help develop detailed and polished care model designs ready for implementation.
Sample Prompt:

Context: We need to create final high-fidelity designs for our new patient-centered care model.
Action: Develop detailed and polished care model designs ready for implementation.
Result: High-fidelity designs that are ready for development.
Example:

High-Fidelity Design 1: A care model that includes personalized care plans and real-time patient monitoring.
High-Fidelity Design 2: A holistic care approach that integrates mental health support with physical health management.

Use Case 2: Patient Acceptance Testing (PAT)

Summary: AI can create test scenarios and gather patient feedback during acceptance testing to ensure the care model’s effectiveness.
Sample Prompt:

Context: We need to conduct patient acceptance testing on our final telehealth care model designs.
Action: Generate test scenarios and gather patient feedback.
Result: Insights from PAT that confirm the care model’s effectiveness or identify areas for improvement.
Example:

Test Scenario: Assess the usability of the telehealth platform during a virtual consultation.
Feedback: Patients found the video quality satisfactory but suggested improvements in navigation.

Use Case 3: Post-Launch Feedback Analysis

Summary: AI can analyze post-launch patient feedback to identify improvements and enhance the care model.
Sample Prompt:

Context: We have launched the new telehealth service and collected patient feedback.
Action: Analyze the post-launch feedback to identify improvements.
Result: A summary of patient feedback and suggested improvements.
Example:

Feedback: Patients appreciate the convenience but requested more language options.
Suggested Improvement: Add multilingual support to the telehealth platform.