From AI to ROI: Evidence Shows Real Productivity Gains in Retail

TL;DR: Recent academic research demonstrates GenAI drove 16.3% revenue increases and 21.7% conversion rate improvements in retail operations. Hybrid AI-human systems achieved 11.5% sales increases versus human-only teams. Findings challenge “AI bubble” narrative, with smaller sellers and less experienced buyers seeing disproportionate benefits.

Despite proliferating claims that artificial intelligence represents an asset bubble, new academic evidence reveals significant revenue and productivity gains from GenAI implementation in real-economy operations. Research from Zhejiang and Columbia universities provides quantifiable data that AI benefits extend beyond infrastructure providers to companies implementing the technology.

Research Findings

The study “Generative AI and Firm Productivity: Field Experiments in Online Retail” tested multiple GenAI applications within large retail platforms. The most successful intervention—adding an AI assistant before purchase—generated a 16.3% revenue increase and 21.7% improvement in conversion rates.

Hybrid systems combining AI with human escalation for complex issues achieved 11.5% sales increases when compared directly against human-only teams. This aligns with findings from HSBC’s investment research experiments, suggesting AI works best augmenting human analysts rather than replacing them.

Distribution of Benefits

Smaller and newer sellers experienced disproportionate gains, alongside less experienced buyers. If this pattern repeats across sectors, it suggests AI may narrow capability gaps rather than creating “winner takes all” dynamics, forcing investors to look beyond obvious beneficiaries.

Not every experiment succeeded. AI-generated product titles for Google adverts produced slightly negative results, with significant declines in impressions and clicks. The mechanism for failure remains unclear—either the models required fine-tuning on advert-specific datasets, or Google’s algorithms identified and downranked AI-generated content.

Practical Implications

The research identifies two key takeaways. First, GenAI value appears in ordinary applications: customer service, query refinement, translation, and customer-product matching. Small best practice changes could generate large economic effects as these use cases prove repeatable across sectors.

Second, AI theme winners will extend beyond infrastructure providers. Companies implementing AI for lower costs or increased revenue may deliver stronger results. US companies implementing AI into operations have already outperformed peers that haven’t, with markets pricing in expectations of continued project successes.

Bubble Narrative Challenged

The “AI bubble” narrative persists due to parallels with dotcom history, but critical differences exist. AI spending primarily comes from large incumbents with vast resources rather than speculative start-ups. Whilst S&P 500 capital expenditure as GDP fraction exceeds dotcom levels, equivalent capex represents only 40% of operating cash flow—far below the 70%+ seen during dotcom mania.

Looking Forward

Returns from GenAI have been visible in AI infrastructure companies’ earnings for some time. Evidence now demonstrates benefits accruing to the broader economy through operational implementation.

The findings suggest expectations of AI-generated benefits are more likely to be fulfilled than the bubble narrative suggests, though high-stakes decisions will likely remain human purview for the foreseeable future. Success depends on thoughtful implementation augmenting human capabilities rather than wholesale replacement attempts.

Source Attribution:

Share this article