Structured Policy Analysis
The Software Management Multiplier
Why the same operational software makes one firm soar and another stall. AI research grounded in evidence, structured by causal mechanisms. Independent verification required.
Key Findings
Research suggests that when two similar firms adopt the same operational software, the results can diverge sharply, and much of that divergence may trace to management and organizational practices rather than the software itself. Structured management practices have been associated with more than 20 percent of the variation in productivity across plants, a share comparable to R&D or IT. A randomized trial in India found that adopting better management practices raised productivity about 17 percent with no new technology, which suggests management can be a causal lever. The returns to IT appear largest when software is paired with process redesign and complementary people practices, and smaller or absent when it is bolted onto unchanged operations.
Effects vary widely by firm size, age, sector, and the quality of implementation. Findings from manufacturing plants or specific software categories do not necessarily generalize to all firms or all tools. Much of the cross-firm evidence is observational, so causal language is hedged throughout.
Identical IT, different returns
US multinationals in Europe obtained higher productivity from the same IT than non-US firms, and the gap was associated mainly with tougher people-management practices. The software was similar. The management was not.
Management is causal, but it decays
A randomized trial raised Indian textile productivity about 17 percent through management changes alone. Nine years later roughly half the adopted practices had been dropped, so the lever is real but not permanent or free.
Software is a complement, not a standalone
Studies find IT, process redesign, and incentive practices act as complements. A bundle of all three has been associated with larger productivity premiums than any one adopted alone, which can leave tool-only adopters near zero.
Headline software ROI is partly selection
Well-managed firms tend to adopt more and better IT, so raw correlations between software and performance can overstate the causal return. Part of the measured management premium also reflects who the firm hires.
Benefits can lag, sometimes by years
General purpose technologies require intangible complementary investment that is poorly measured. This can produce a J-curve: measured productivity dips first as firms build capability, then rises later. Returns over five to seven years have been much larger than short-run estimates.
Even fast tools depend on context
One controlled study found AI coding assistance cut a benchmark task time about 56 percent. A separate trial with experienced developers on their own large codebases found a 19 percent slowdown. The tool was capable in both. The context differed.
Research Findings
Sources
What this means in practice
Work related to operational software often involves manually moving data between systems, re-keying numbers into spreadsheets, chasing status updates, and rebuilding the same reports each cycle. These manual steps are the process around the tool, and they are typically handled with systems that automate the repetitive parts so the software has something to multiply.
- Ingest data from existing systems, files, and forms
- Automate the repetitive processing, modeling, and tracking steps that surround the software
- Generate consistent outputs and reports for analysis and review
Related Research
Manual Workflows at Scale
Evidence, failure modes, and system outcomes for manual coordination, data entry, and approval processes
Data Fragmentation & Operational Inefficiency
Evidence on how data silos, disconnected systems, and fragmented data sources create operational costs and productivity losses
Does AI Coding Assistance Actually Speed Developers Up?
When AI coding assistance speeds developers up versus slows them down, and why developers are often wrong about which is happening