Researchers Propose Framework to Measure Enterprise AI Adoption Through Real-World Usage Data

A team of researchers has introduced a systematic approach to measuring how enterprises adopt AI technologies by monitoring actual usage patterns from deployed systems in production environments.

The Framework#

Rather than relying on surveys or estimates, the method—called adoption telemetry—extracts signals directly from how AI systems are being used in practice. It integrates three key components: pre-launch evaluation checkpoints, real-time telemetry collected from running systems, and established change-management stages that organizations typically progress through during technology rollouts.

Five-Stage Model and Open Standards#

The researchers have operationalized their approach into NANTE, a concrete five-stage model with specific telemetry thresholds. Importantly, they have published these thresholds openly, inviting researchers and practitioners to test and refine them. This transparency contrasts with proprietary models that remain black boxes—here, the team explicitly positions their thresholds as proposed constructs rather than finalized calibrations.

Validation Tools#

Accompanying the framework is an open-source reference implementation capable of distinguishing healthy adoption patterns from five characteristic failure modes. The team tested their system on synthetic populations with known ground truth, demonstrating how the tool can flag when deployments are struggling.

Future Directions#

The researchers acknowledge that their thresholds require real-world validation against actual organizational outcomes. They outline a research agenda for collecting and analyzing data from genuine enterprise deployments to refine the model over time. This positions adoption telemetry as a living methodology rather than a static tool.