What Harvard and Kellogg are telling executives about AI

When the academy and the consultancies agree, it's worth paying attention. Harvard Business School's 2025 working paper “The Cybernetic Teammate” (Dell'Acqua, Sadun, Mollick, Lakhani and colleagues) ran a pre-registered field experiment with 776 professionals at Procter & Gamble — real product-development work, randomly assigned, properly measured. The headline finding: individuals working with AI matched the performance of two-person teams working without it. AI didn't just speed people up; it replicated part of what a teammate provides — breadth of expertise the individual didn't have.

Northwestern's Kellogg School, meanwhile, has been blunt about the other side of the ledger. Their executive-education faculty describe the corporate AI landscape in one sentence: companies have multibillion-dollar AI budgets and they're stuck in pilots and proofs of concept, unable to scale to meaningful impact. Kellogg researchers — including Matthew Groh, Julio Ottino, and Brian Uzzi — have published an adoption framework aimed at exactly that failure mode, and demand for the school's AI strategy programs (the largest in its executive-education history) tells you how many leaders recognize themselves in the diagnosis.

Read together, it's an operating manual

The Harvard result tells you what AI is worth when it's deployed against real work: one person with the engine does the work of two. The Kellogg critique tells you why most companies never collect that value: they buy technology instead of changing a workflow, and pilots without owners die in committee. Stanford's AI Index 2026 splits the difference with measured 14–26% productivity gains — but only in the functions where the work is structured and countable.

For a mid-sized business, the synthesis is more actionable than for an enterprise. You don't have a committee to die in. Pick the workflow where one-person-does-the-work-of-two matters most — the phone, the follow-up, the paperwork — baseline it, deploy against it, and measure. That's not our invention; it's what the research now says out loud.

Why we cite the professors

Vendor case studies — ours included — deserve your skepticism. Field experiments with 776 participants and pre-registered designs deserve more weight. When we tell a client that an AI front office or document engine can carry real load, we'd rather point at Harvard's data and Kellogg's framework than at our own marketing. Then we prove it against your baseline, which is the only evidence that should ultimately matter to you.