New research involving 776 Procter & Gamble experts suggests individuals using AI can perform as well as traditional two-person teams.
During one-day workshops, participants developed product ideas for various P&G business units. The teams consisted of one commercial and one technical expert, with half of both teams and individuals receiving access to GPT-4 and GPT-4o. The data showed teams without AI outperformed individuals by 24%. However, individuals using AI improved by 37% - matching the performance level of non-AI teams.
Teams with AI support achieved the highest overall performance (39% improvement), though the difference compared to individuals using AI wasn't statistically significant. However, teams using AI were about three times more likely to produce solutions ranking in the top 10% of quality scores.
Groups using AI worked 12-16% faster while delivering longer, more detailed solutions. More than 75% of AI-generated content was kept by a "substantial proportion" of the groups, according to the study.
AI balances expertise differences
Another finding is that the technology seems to level the playing field between different types of experts. Without AI, technical experts stuck to technical solutions while sales experts focused on market aspects. With AI assistance, both groups started producing more well-rounded proposals.
This effect was particularly noticeable among less experienced product development employees. Without AI, they performed relatively poorly, even in teams. With AI support, they suddenly achieved performance levels comparable to teams with experienced members.
Contrary to common assumptions about new technology causing stress, participants using AI reported more positive emotions like enthusiasm and energy, with fewer signs of anxiety and frustration.
The researchers suggest companies should reconsider viewing AI as just a productivity tool. Instead, they propose treating the technology as an additional team member.
Understanding the limitations
The study comes with several important caveats. The research design may have limited AI's potential by using it primarily as a chatbot. While chats excel at quick ideation, solution quality depends partly on chance and heavily on the user's expertise.
The volume of text generated requires professional evaluation, and marketers, for example, may lack the skills to properly evaluate technical suggestions - they might as well blindly trust the AI. What they are really doing is producing words, not necessarily useful knowledge.
The researchers' view of AI as a team member, rather than just a tool, appears to stem mainly from how human-like chatbots can seem in conversation. According to study leader Ethan Mollick, "Although it is not human, it replicates the core benefits of teamwork - improved performance, sharing of expertise and positive emotional experiences. This team player perspective should lead organizations to think differently about AI."
But in product development, where structured processes already exist, a more effective approach might be using AI to standardize these processes themselves, rather than humanizing AI and treating it as a conversation partner. Companies could then make these AI-enhanced processes - or even just their outputs - available to teams.
The one-day workshop format is another limitation, as it doesn't capture the complexity of real-world enterprise environments, where work typically involves multiple iterations over extended periods of time. Questions remain about how much AI-generated content would survive extended development cycles.
Success with AI implementation varies significantly based on corporate culture, existing workflows, and technical infrastructure. Perhaps most importantly, questions remain about how AI integration affects skill development and knowledge transfer over time - crucial factors for maintaining competitive advantage.
