AI workflow gaps lead firms to scale back projects

CambrianEdge.ai has launched in the UK with research into workplace AI use. The study found that 18% of organisations had rolled back or abandoned AI initiatives because of quality problems.

The findings were presented at the House of Lords and were based on a survey of 775 professionals across 104 organisations, including large companies, marketing agencies, startups, law firms and educational institutions.

The research argues that many businesses have focused on giving employees access to AI tools without putting in place the workflows needed for consistent use across teams. More than half of respondents, 55%, said isolated solo use or the lack of a structured human-machine workflow was their main operational bottleneck.

That gap appeared in several ways. According to the research, 62% of organisations had no defined process for handing off AI-generated work for human review, while 27% lacked what it described as collaboration infrastructure, such as shared access, prompt libraries, training or quality standards.

The study argues that these missing processes directly affect results. Among organisations with no infrastructure layers in place, 32% reported significant impact from AI, compared with 100% of organisations that had all five layers identified in the research: shared tool access, formal training, prompt libraries, quality standards and mandatory review processes.

Defined handoff processes were also linked to better outcomes. The survey found that 71% of organisations with structured workflows reported significant outcomes, compared with 38% of those with unstructured workflows.

The report also pointed to a disconnect between executive confidence and day-to-day practice. Citing separate data from a BCG study of 300 global chief marketing officers, it noted that 96% believed AI was driving end-to-end transformation, while nearly half still limited its use to isolated tasks handled by individual employees.

Harjiv Singh, Founder and Chief Executive Officer of CambrianEdge.ai, said the problem was organisational rather than technical. "Most organisations spent the last two years asking which AI model to subscribe to, forgetting to ask how their teams were supposed to work with it," Singh said. "Adding AI to a system built for siloed work is like putting electric lights in a building designed for candles; the architecture needs to change, not just the bulbs. True economic value only materialises when companies abandon a fragmented stack of individual tools and build a shared, continuous workflow."

Quality concerns

The survey suggests weak controls can reverse adoption rather than simply slow it. Eighteen per cent of respondents said their organisations had already pulled back from AI projects or dropped them altogether because of quality failures and broader adoption problems.

The finding goes to the heart of a debate facing many management teams as AI moves from individual experimentation to formal business use. While many employees now use AI tools every day, the research suggests companies often lack clear standards for review, approval and accountability.

CambrianEdge.ai compared those concerns with users of what it described as an AI-native collaborative platform. According to the company, this group recorded a 98% active engagement rate after onboarding, and 56% said their main remaining obstacle was execution velocity.

This shift came when teams consolidated research, creation, distribution and analysis into one environment instead of relying on separate tools. The study argues that moving from disconnected individual habits to a continuous team workflow changes how organisations assess the value of AI.

Policy debate

The research was presented alongside a panel discussion that included representatives from academia, business and public life. The wider study was carried out with partners including the Cambridge Central Asia Forum, Stanford SEED, US-India Strategic Partnership Forum, Gutenberg, Digimentors and Brand Communion.

Lord Raj Loomba, Member of the House of Lords and Founder and Chairman of The Loomba Foundation, linked the findings to a broader policy and governance question. "AI has arrived at a threshold where the technology is ready, but the organisational architecture has not kept pace," Loomba said. "If we do not address this through deliberate design, we risk reducing a transformative technology to a mere collection of individual tools. The conversations we must have now, in Parliament and in boardrooms alike, must ensure that as we shape this future, intelligence remains human."

The report said businesses should define how work moves from AI generation to human review and final deployment before trying to scale adoption. It also argued that national AI readiness measures should move beyond simple rates of tool adoption and pay more attention to workforce fluency, training and quality control.

For companies under pressure to show returns on AI spending, the findings add to evidence that adoption alone does not guarantee productivity gains. In this survey, the difference between progress and retreat appeared to depend less on the model being used than on whether organisations had built processes that let people work with it together.