AI + You: 5 steps for impactful experimentation

Executive overview

As AI evolves exponentially, the question isn't whether to experiment with it, but how to integrate it effectively. This episode walks through five concrete steps to experiment with AI that drive real business value. Rather than viewing AI as an overnight transformation, leaders should approach it with patient experimentation, starting small, building trust progressively, and staying attuned to both technical and human dimensions of adoption.

Core insight: AI amplifies human decision-making when you design experiments around specific pain points and measure outcomes clearly.

Step one: Learn what's out there

• You're still near the starting line—it's a marathon, not a sprint; save your energy and resources.
• AI will certainly touch every business (language, communication, meetings, sales, marketing, customer service) but timing varies.
• Don't decide today whether you do AI or not. Revisit on a regular basis as new applications emerge.
• Before integrating, ask: Does it answer a customer need? Does it help move your business forward? If uncertain, wait another month or quarter and reassess.

Step two: Focus on the pain points

• Traditional AI analyzes data and tells you what it sees; generative AI creates something new using that data (text, images, music, conversation).
• Identify a specific area of your business with a clear problem that AI could solve.
• Hand hygiene monitoring in hospitals (using computer vision) reduced infection rates and improved detection accuracy over human observation.
• Acute kidney injury (AKI) detection: an algorithm that read electronic health records reduced treatment costs by 20% and cut detection time from four hours to 15 minutes.
• Google DeepMind's AI also matched human expertise in eye disease and breast cancer detection—success came from solving real pain points.

Step three: Don't preach the power of AI. Illustrate it

• Your team won't uniformly embrace AI; technology enthusiasts, designers, and business leaders have different concerns.
• Win skeptics through concrete, measurable experiments with clearly defined outcomes—not large town hall announcements.
• Microsoft illustrated Copilot's value by using it in live meetings, then sharing meeting summaries. This created organic curiosity and adoption far more than forced evangelism.
• Ask your team: What unique uses are you finding? How much time is saved? Is satisfaction higher? Are close rates improving?
• Metrics should evolve through phases: familiarity → novel uses → business efficiency measured in language you already use.

Step four: Onboard and collaborate with AI like any new employee

• Hire AI gradually: start with simple tasks, then progressively increase complexity as trust builds.
• When results disappoint, investigate like you would with underperforming staff—ask questions, offer feedback, check your instructions.
• A single piece of targeted human feedback to a language model can improve its solve rate by 20–30% on a given task.
• The fastest path to progress is a direct feedback loop between user and model; the model learns from your corrections and preferences.
• Models improve for you as an individual user without sharing your data with others—your feedback generates personal memory and smarter responses over time.

Step five: Experiment with your eyes open and be prepared for discomfort

• Check the terms and conditions of any AI service you use, especially before uploading customer, client, or medical data.
• Many consumer services retain the right to use your data to train their models; protect sensitive information by choosing services that guarantee privacy and security.
• AI will reveal gaps and limitations in your existing processes—you must be prepared to embrace change, not ignore what the experiment shows you.
• Compute required to train state-of-the-art models has 10x'd every year for the past decade; exponential progress will continue.
• Accept that mistakes are a key byproduct of experimentation. Errors often lead to your greatest learnings.