Keynote: Making IoT Experimentation Easy

### Overview In this keynote address, Sarah Cooper, General Manager of IoT Solutions at Amazon Web Services (AWS), discusses the challenges and methodologies associated with making IoT product experimentation and post-deployment updates seamless. As the speed of technological innovation continues to decouple and outpace product lifecycles, connected devices often accumulate 'technology debt' once deployed. Cooper argues that an IoT product truly begins its lifecycle the moment it is installed, requiring organizations to integrate an experimentation layer into their technical architecture, tooling, and culture to deliver continuous value to customers.

### Key Topics - **Accelerating Pace of Technology and Technical Debt:** Technologies evolve faster than physical hardware lifespans. Unlike static systems of the past, modern connected products must adapt to avoid becoming obsolete shortly after deployment. - **Outcome Economy and Customer Value:** Customers increasingly evaluate purchases based on sustained value and outcomes (e.g., paying for soil nitrogen fixation rather than fertilizer). Maintaining long-term customer satisfaction requires updating capabilities dynamically in the field. - **The Three Pillars of Experimentation:** Effective experimentation relies on architecture, tools, and culture. A successful experimentation framework ensures low-risk rollouts, minimal blast radius, and data collection to evaluate outcomes. - **Applying the Scientific Method to IoT Design:** Iterating on products requires developing a clear hypothesis, running controlled tests, analyzing contextual and streaming data, and evaluating rollout feasibility for target customer cohorts.

### Technical Details - **Architectural Separation and Blast Radius:** Designing modular systems allows teams to test changes on small cohorts rather than risking full-scale fleet failures. The speaker emphasizes containing the blast radius both across software services and across customer segments. - **Data Temperatures:** Data utilized in IoT experimentation is categorized by 'temperature': - *Hot Data:* Real-time, streaming information required immediately (e.g., processed via Kinesis). - *Warm to Cool Data:* Contextual, behavioral, and diagnostic telemetry used to understand operational anomalies or correlate unanticipated patterns. - *Cold Data:* Historical population data used for baseline establishment and long-term analysis. - **Serverless and Edge Architectures:** - *AWS Lambda:* Serves as an isolated environment for running experiments, data cleansing, enrichment, and rapid algorithm testing in the cloud. - *AWS Greengrass:* Combines Lambda execution and the AWS IoT communication stack onto edge devices (typically running Linux). This enables developers to simulate functional logic in the cloud and push discrete code updates directly to physical devices in the field for localized in situ testing. - **Solar Panel Cleaning Case Study:** An experiment was shared evaluating cleaning regimes on solar panels using algorithms to differentiate shadow from dirt, testing water sheets versus mechanical squeegees to illustrate quick iteration and rollback using serverless functions.

### Takeaways - IoT devices should not be treated as fixed, static products; value is preserved by adapting to new technologies, user skill sets, and changing operational needs post-installation. - Serverless architectures and edge runtimes provide the modularity needed to run discrete experiments, push updates to defined subsets of devices, and roll back quickly if tests fail. - Organizations must pair robust technical tooling and controlled baselines with a culture that encourages engineering experimentation and minimizes fear of failure.