How businesses are putting ChatGPT to work, and what to measure next

New OpenAI Signals data looks at how people are using ChatGPT across countries and how their behaviour changes over time. The headline for UK teams is simple: usage is moving from asking questions towards completing tasks, so the next step is to design business workflows that are measurable, repeatable, and safe.

What changed in how people use ChatGPT

The update frames a shift from conversational prompting to more work oriented outcomes. While the underlying source is global, the value for businesses is the direction of travel: people are not only using ChatGPT to answer, they are using it to get things done, and those patterns vary by location.

Why country level visibility matters for UK adoption

The signals include country level insights on adoption and usage trends. Even if your organisation is not modelling each country, the practical takeaway is that usage and behaviour are not uniform. That means you should expect differences across teams and roles inside your UK business, and you should measure those differences rather than assuming one prompt approach works everywhere.

Where to start, pick workflows not experiments

If the trend is from asking to doing, your implementation should follow the same logic. Start with workflows where teams already work with drafts, summaries, or first pass outputs, then connect ChatGPT output to a defined review and approval step. Define what done means for the business, for example a usable customer response draft, an internal briefing, or a structured plan ready for sign off.

How to measure productivity and ROI without guesswork

Because the signals focus on adoption and usage trends, you can mirror that measurement internally. Track which teams use ChatGPT, how often, and what kinds of tasks they attempt. Then add a simple outcome metric tied to your workflow, such as time saved to reach a review ready state, fewer revisions, or reduced time spent on first drafts. The point is to treat ChatGPT usage as operational data, not a one off trial.

Risk management, make use safe by design

A move from asking to doing usually increases impact, so risk controls need to scale too. Put guardrails around what inputs are allowed, require human review for anything that goes to customers, and keep an audit trail of approvals. Use your defined workflow stages to ensure AI output is always verified before it becomes operational or customer facing.

Questions to answer this week

• Which teams are likely to move from asking to doing first based on their day to day tasks
• What is the review ready definition for your chosen workflow
• How will you track adoption internally, by team and by task type
• What input and output rules are required to keep risk under control