Real cases

The headline number is the easy bit.

These cases use published evidence. I look at what changed, who gained, what the headline left out, and what you can reasonably learn from it.

Case reading rule

A result from another organisation is a clue. It is not your business case, however much someone wants it to be.

Customer support · Fortune 500 software company

Generative AI increased issues resolved per hour.

14%average productivity increase

What was observed

A field study followed 5,179 customer-support agents using a generative AI assistant. Productivity rose by 14% on average. The gains were much larger for less experienced and lower-skilled workers.

What is more interesting

The system appeared to spread practices used by stronger workers. The case concerns learning and access to useful knowledge as much as speed.

What this does not prove

This was one company and one type of work. Experienced workers gained less. A 14% average should not become a general promise for every service team.

Adoption lesson

Segment your evidence. Ask who benefits, who does not, and whether AI helps people build capability or merely complete more tickets.

UK government · Microsoft 365 Copilot trial

People reported time savings, with wide variation.

26 minaverage self-reported saving per day

What was observed

A cross-government trial collected survey and usage evidence from thousands of civil servants. Participants reported saving time, especially in drafting, summarizing, and preparation.

What is more interesting

Some participants described more space for higher-value work. The gain depends on whether teams protect and redirect that space.

What this does not prove

The central time figure was self-reported and benefits varied. Perceived savings do not prove that services, decisions, or workloads improved.

Adoption lesson

Ask people where time was saved, then observe what happened next. Pair experience data with task evidence and service results.

Send the useful details

A useful case includes the bits that went wrong.

Future cases will include where the team started, what it wanted, who was affected, what changed, what went wrong, and what people would do differently next time.

Approved, anonymized evidence only
No invented customer stories
Numbers with context and method
Lessons that others can test