Article: OpenAI’s analysis of 800,000 ChatGPT messages sent by U.S. professionals shows that nearly one-in-six work-related prompts now ask the model to handle tasks that traditionally belong to a different occupation. Strip out routine writing and summarising, and the figure jumps to 43.5 %. The shift is already reshaping how employees spend their day, and it matters most to the smallest firms, where every extra skill can shave weeks off a wait for a specialist.

Why the numbers matter

The study found that cross-departmental requests cluster in certain functions. Marketing and engineering lead the pack, with marketing-related queries appearing in 8.9 % of messages from other roles—the highest rate in the data set. A salesperson might ask ChatGPT to crunch customer-lifetime-value numbers that a data analyst would normally handle, while a marketer can troubleshoot a broken link or tweak on-page SEO before ever involving an engineer.

For small business owners, the effect is sharper. Owners now draft legal review notes, run quick financial forecasts, and perform ad-hoc analysis that used to require a lawyer or CFO. In teams of two to five people, 18.9 % of AI-generated tasks fall outside the employee’s core role, versus 16.3 % in enterprises with more than a hundred staff. The math is simple: larger firms can pass the request to a dedicated colleague; tiny outfits must solve the problem themselves.

The upside: speed and sharper hand-offs

When an employee generates a first-pass analysis or a rough legal brief, the hand-off to a specialist becomes faster and more focused. The specialist receives a clearer problem statement, better-prepared data, and fewer “please explain what you need” emails. That cuts wait times, reduces internal friction, and raises request quality.

What’s at risk

The study does not claim AI will make experts obsolete. It frames the technology as a front-line responder that brings the problem closer to its source.

How companies can adapt

  • Map decision points – Spot where a request typically stalls and test whether AI can produce a usable first draft.
  • Define authority limits – State which outputs need sign-off from a qualified professional.
  • Train for the new skill set – Offer micro-learning on prompting, interpreting AI suggestions, and spotting likely errors.
  • Measure impact – Track time saved on hand-offs against any rise in error rates or rework.

Small vs. large orgs: divergent paths

The next frontier

OpenAI’s data capture a practice that is evolving fast. As models become more reliable and domain-specific, the share of cross-departmental tasks will likely climb. Organizations that treat AI as a mere productivity tool rather than a catalyst for workflow redesign risk falling behind. Those that redesign processes now—deciding which decisions stay human and which can be AI-augmented—will reap faster cycles and more agile teams.

Takeaway: AI is already moving the first response to the problem’s origin, especially in small businesses where every extra capability trims delay. The real advantage comes not from renaming jobs, but from rethinking who does what, and where a machine can safely step in.