# KPIs and Reporting to Leadership — AI Strategy, Ethics and Governance

Source: https://www.geekswithgeeks.com/en/ai-strategy-governance/ops-kpis

> Track a small balanced set of value, quality, risk and cost metrics and report them honestly.

## A balanced scorecard

Report a handful of metrics that cover four angles. **Value**: share of tickets handled by AI (containment), time saved, adoption. **Quality**: accuracy or grounded-correct rate, escalation rate, customer satisfaction. **Risk**: incidents by severity, overdue reviews, fairness gaps, policy exceptions. **Cost**: spend against budget and cost per resolved task. Show trends and targets, and include bad news. A dashboard that only shows good numbers loses leadership trust.

## Three KPIs, run

I ran this with example numbers: 6,000 of 10,000 tickets were handled by AI (containment 0.6); 600 of those were escalated (escalation rate 0.1); spend of 4,500 over 5,400 resolved tickets gives 0.83 per resolved ticket.

```python
tickets, ai_handled, escalated, cost = 10_000, 6_000, 600, 4_500
print(round(ai_handled / tickets, 2),                 # containment
      round(escalated / ai_handled, 2),               # escalation rate
      round(cost / (ai_handled - escalated), 2))      # cost per resolved ticket
```

Output:

```
0.6 0.1 0.83
```

## Pair every value metric with a guardrail metric

If you push containment up, also watch satisfaction and escalation quality. Optimising one number alone can quietly damage what actually matters.

**Quiz:** Why pair a value metric like containment with a quality metric like satisfaction?

- [ ] Containment is not measurable
- [ ] Satisfaction is always lower
- [x] Raising one alone can hide damage to the thing that matters
- [ ] Pairs of metrics are required by law

*Answer:* Raising one alone can hide damage to the thing that matters. A balanced scorecard stops teams from gaming one number at the expense of customers.