Why the real GenAI ROI hides in the back office
Since ChatGPT burst onto the scene in 2020, over $30 billion of enterprise dollars have poured into GenAI initiatives. Most of this spending has gone to shiny AI tools, chatbots, or costly internal builds promising quick wins.
The results have been uneven at best; while the top 5 percent have captured millions in ROI, a 2025 report by MIT shows that the large majority has led to only small productivity gains. So what’s the disconnect?
Right tech, wrong application
The problem isn’t GenAI itself, but where and how it’s applied.
Despite back-office automation driving the biggest wins, front-office functions steal 70% of AI budgets. Pressure to meet headline business KPIs leads leaders to pour resources into sales, marketing, and customer service initiatives.
What most fail to see in the busyness of work is that margin, in fact, is king. Take AstraZeneca, a global pharmaceutical company with a ~$15B annual spend. If they replace half of their ~500-person procurement team with AI, it would result in ~$25 savings annually*.
Make that same team more effective in managing that spend by working alongside AI teammates, and you can multiply the value of their labour by >5x. Just 1% efficiency gain alone can deliver ~$150M in annual savings by leading to faster production times, cheaper goods, and a competitive edge.
This is mirrored in the findings of the MIT report: the organisations that saw the highest ROI have barely seen a change in headcount. Instead, their team is delivering much more.
Data blocks back-end progress
So if the maths check out, why aren’t more companies prioritising back-end operations? The short answer lies in data.
Data is the oil that fuels the AI machine, but with most global manufacturers still wrestling with fragmented systems and data silos, most can’t fully take advantage of even simple automation tools.
The idea of a multi-year, multi-million-dollar data cleanup initiative? Daunting. The motto has long been to learn how to walk before you run.
GenAI fundamentally changes this by bypassing lengthy data transformation projects. For example, Magentic’s AI teammates, called Mages, find and consolidate immature master data across systems, spot mismatches, and prioritise cleanup efforts where they matter the most.
AI agents don’t need perfect data. They gradually improve data quality by focusing on high-value fixes and delivering tangible business impact in real-time. Rather than waiting years, leaking value in the process, companies can unlock savings bite by bite.
Internal builds stall innovation
Before stepping into the CEO shoes as the co-founder of Magentic, Robin worked as a consultant at McKinsey & Company. There, he advised supply chain leaders on how to drive change that doesn’t just look good on paper.
The findings in the MIT report echo what he saw at McKinsey day after day: it’s rarely the technology that kills AI projects;it’s the people and process.
Apart from the obvious; lack of internal buy-in and difficulties driving large organisational change, internal builds often falter for two key reasons. First, the stakes are high and expectations unrealistic. When careers are on the line, pressure inflates timelines and scope, creating a ‘cannot fail’ mindset. Speed gets prioritised over experimentation, leading to incremental change over big transformations.
Second: poor user experience. Internal tools often opt for technical integration over usability. Forced adoption breaks vital feedback loops, causing organic usage to stall. External providers, by contrast, see nearly double employee usage rates because they build technology that users love. Competition pushes innovation.
Given these dynamics, it’s hardly surprising that deployments launched with vendor partners are twice as likely to succeed.
