Four studies showing how AI actually strengthens teams

The AI discussion too often focuses on individual productivity. ChatGPT for the marketer, Copilot for the analyst, Gemini for the developer. Everyone their own assistant. It sounds logical, but the research shows otherwise: standalone AI access hurts teams as often as it helps. What works is context and the right task distribution.

Four studies, published between 2023 and 2025 in leading journals, show where the real gains are. And where AI fails without that shared context.

1. AI lifts the whole team to the level of the best

Brynjolfsson, Li and Raymond (Generative AI at Work, Quarterly Journal of Economics, 2025) tracked 5,172 customer-support agents as they got access to an AI assistant. Average productivity gain: +15%. For the least experienced workers: +34%.

Why? The AI spreads the approach of the top performers. What normally takes years of shadowing happens in weeks. Provided the AI knows how the team works.

2. AI breaks down walls between departments

A field experiment at Procter & Gamble with 776 professionals (The Cybernetic Teammate, Dell'Acqua et al., NBER 2025) found something remarkable: an individual with AI performed as well as an entire team without. And the typical silos between R&D and Commercial disappeared.

AI acts as a translator between disciplines. Technical jargon becomes commercially understandable, commercial requirements become technically concrete. For a lean SME team, this gives every member access to expertise you otherwise wouldn't have in-house.

3. AI alone is no guarantee. Task allocation is everything.

The largest meta-analysis to date (Vaccaro, Almaatouq and Malone, When combinations of humans and AI are useful, Nature Human Behaviour, 2024) looked at 106 studies and 370 effect sizes. The uncomfortable finding: human-AI combinations on average underperform the best of the two alone.

But there was a clear pattern. Gains in creating: concepts, prototypes, first drafts. Losses in deciding: judgement, risk weighing, strategic choices.

4. Without context, your team navigates blind

BCG and Harvard Business School (Dell'Acqua et al., Navigating the Jagged Technological Frontier, 2023, now published in Organization Science, 2026) gave 758 consultants access to GPT-4. On tasks inside AI's "jagged frontier": +12% tasks completed, 25% faster, 40% better quality.

But on tasks outside that frontier: 19 percentage points worse than without AI. And the treacherous part: consultants trusted the AI most precisely there.

Generic AI hides its own boundaries. A tool configured for your domain makes them visible.

The homogenisation question

A common objection: "If everyone uses the same AI, won't everyone start thinking the same way?"

That objection holds. Doshi and Hauser (Generative AI enhances individual creativity but reduces the collective diversity of novel content, Science Advances, 2024) showed that generative AI increases individual creativity but reduces the collective diversity of outcomes.

The solution isn't "no AI", but a shared tool that builds in diverse perspectives. Standalone chatbots amplify the problem (everyone gets the same generic output). A team AI that knows your organisation's context can actually introduce variation and alternative scenarios.

The question isn't whether, but how

The studies all point the same direction: AI has a team effect that's orders of magnitude larger than an individual effect. But that team effect only emerges if the AI:

- knows the team's way of working

- delegates the right tasks (not everything)

- takes the organisation's context into account

- deliberately builds in diversity of perspectives

That's why a shared digital colleague configured for your department, rather than a general assistant for every employee, isn't a cost optimisation. It's a strategic choice.

The question for you isn't whether your team uses AI. It's whether your team delegates the right tasks and whether the tool knows how your organisation works. That's the difference between +15% and +34%.

Curious how a digital colleague would look for your department? See our Early Access Program.