Most research on AI's business impact falls into one of two buckets: vendor case studies (which are marketing) or employee surveys (which measure belief, not reality). A new working paper on SSRN, "Mapping AI into Production: A Field Experiment on Firm Performance," belongs to a third, rarer category - an actual controlled experiment.
What Makes a Field Experiment Different
In a field experiment, researchers don't ask participants what they think happened - they observe what actually happened. Companies or teams are divided into groups, some with access to AI tools and some without, and researchers measure real performance metrics over time. This controls for the attribution bias that plagues self-reported studies, where people who adopted something new tend to credit it for improvements that might have happened anyway.
The paper examines how AI deployment affects firm-level performance - not just individual task completion speed, but actual business outcomes. That distinction matters. A tool that makes a marketer 20% faster at writing first drafts may or may not translate into better marketing results, depending on whether draft writing was the actual bottleneck.
Deployment Conditions, Not Just Averages
The framing around "mapping AI into production" suggests the paper identifies which deployment conditions produce measurable gains - not just a single average-effect number. This is the more useful finding for anyone making real adoption decisions. The same AI tool deployed into a content team versus an operations team versus a sales team produces different results, and research that maps those conditions is more actionable than aggregate statistics.
The paper is an academic working paper, meaning peer review may still be pending. But field experiment methodology is the most credible approach available for measuring AI's real-world business impact. The full paper is available on SSRN.