Analysis · Technology
Why Software Spending Complicates the Measurement of AI’s Value
By AWEI · AI-compiled · Published · Analysis prepared · 1 source · doi.org
How much of an AI user’s productivity advantage belongs to AI, and how much to its other technology investments? A 2023 study exposes a measurement problem: accounting for software spending can help separate those contributions, but may also subtract part of the AI investment itself.
Dirk Czarnitzki, Gastón P. Fernández and Christian Rammer examined 5,851 German firms, including 409 AI users, in Artificial intelligence and firm-level productivity. Their survey data covered 2018 and counted active use of both internally developed and externally supplied AI, across methods including language understanding, image recognition and machine learning.
The initial model compared sales while accounting for employment, tangible capital and purchased inputs. It associated AI use with roughly 14% higher sales. Adding company age, industry, innovation activity and past software expenditure reduced the reported association to about 6%, with modest statistical support. These estimates concern output relative to production resources, not project profitability.
The smaller estimate is not automatically more accurate. Software expenditure helps account for wider digital investment that could otherwise be credited to AI. Yet the authors explicitly warn that those expenses may include AI investments, potentially understating AI’s contribution. Their data could not fully separate the two. The reduction therefore illustrates sensitivity to the comparison, rather than establishing the correct size of an AI payoff.
The researchers also used statistical instruments, reweighted comparison firms and examined changes between 2017 and 2018. Results generally supported a positive relationship, but magnitudes varied substantially. This was observational research in one country. Causal interpretations depend on assumptions; AI use was recorded in only one survey wave, and the short panel treated it as a proxy for recent investment.
For businesses evaluating AI, the useful implication is to distinguish AI spending from broader software investment wherever possible. Track output alongside the resources used to produce it. This study supports investigating productivity gains, but supplies neither a guaranteed return nor an adoption-to-payoff timetable.
Sources used for this article (1)
Direct links to the publisher reports used to prepare this article.
- Source 1
- Artificial intelligence and firm-level productivity Dirk Czarnitzki, Gastón P. Fernández, Christian Rammer
