Discover and source to represent the business in data. Gathering data must be iterative and continuous to create, train, and optimize models. New data and data types (text, voice, image, audio, video) should augment and improve machine-learning (ML) models as data becomes more representative of the environment where AI is deployed. Marketplaces and exchanges can enable trusted data sourcing through data scientist self-service.

Capture and ingest data for quality and relevance. Fresh data is a prerequisite for performant AI. Data scientists need representative data moving into their sandboxes and training environments. In a connected intelligence model, data capture happens through streams to handle ephemeral data (e.g., time series). Data fabrics flex to match and shape data flows to keep data and insight in time with the digital experience and outcomes.

Curate and model data for better context. Forrester finds that 62% of global data and analytics decision-makers source external data. This requires constant classification, labeling, and certification of data to understand and govern data for self-service use. AutoML on structured data, computer vision, and behavioral ML on data use can combine to scale and speed up data curation and modeling to meet data scientist and connected intelligence solution demands.

Transform and prepare data for increased relevancy. Shaping data for ML is fraught with business logic, security, privacy, and regulatory considerations. Data scientists, data engineers, and data stewards collaborate and share transformations and preparation steps to streamline the influx of data. DataOps and data governance tools use ML to set data standards, schemas, and controls while providing the transparency and traceability of the data flow for impact and root-cause analysis.

Test and train to engender trust. Make AI testing holistic across data services, data models, business logic, governance, and service levels for data, metadata, and ML models and solutions. DataOps, ModelOps, and DevOps can play a critical role in end-to-end and granular testing and tracing of data flows affecting business logic and routing, as well as the ML model itself.

Deliver and deploy for scale. Consistent use of continuous integration and delivery (CI/CD) processes and practices keep DataOps, ModelOps, and DevOps connected and collaborative. Connected intelligence is built component by component and delivered as a product by data engineers, machine-learning engineers, and software engineers. Feature store platforms, data fabric, cloud, and edge computing systems create the backbone to deploy each component quickly, easily, at scale, and properly governed.

Execute and act dynamically to drive outcomes. Data and ML models must stay responsive and continuously adapt to business conditions and decisions. MLOps, data governance, and continuous lineage analysis ensure that connected intelligence is meeting service-level agreements. CI/CD allows dynamic versioning of data and model components to optimize the system, and monitoring and alerts provide an early warning system of potential outcome degradation.

Observe and evaluate for refinement and ongoing governance. Anomaly detection by DataOps helps preemptively identify and quarantine data on capture to mitigate upstream AI degradation or risk. One global financial firm has unified anomaly detection between data and ML to effectively enhance risks and threats for known and unknown bad-actor behaviors. Data observability and MLOps tools bring context to data anomalies and performance impact to mitigate AI governance and responsible AI risk conditions.