Public sector AI success should be measured by citizen outcomes, not adoption metrics

When citizens renew a passport or book a medical appointment, they don't care whether artificial intelligence processed their request. They care whether the service was fast, reliable, and easy to use.

Yet as governments accelerate AI deployment, many risk measuring success by the technology itself rather than the outcomes it delivers, according to Fawad Qureshi, Field CTO at Snowflake.

"People want to talk about the number of copilots, the number of prompts, summaries, documents generated or AI adoption rates," Qureshi said, as first reported by Automation Watch. "So what? Is the citizen's life improved? That's the outcome that needs to be there."

Why it matters

Public sector organizations face mounting pressure to demonstrate AI progress, but focusing on technology metrics rather than service outcomes can lead to misallocated resources and missed opportunities. When departments optimize for adoption statistics instead of citizen experience, they may report impressive numbers while failing to deliver meaningful improvements in the services people actually use.

The trap of activity metrics

Technology has always generated its own measures of success. Software developers were once judged by lines of code written. Today, AI introduces new numbers to celebrate: chatbot conversations, prompt volumes, token consumption.

But higher numbers don't necessarily mean better outcomes. Qureshi pointed to reports of one organization spending hundreds of millions of dollars on AI tokens after encouraging widespread use without sufficient controls—an example of how activity can become disconnected from value.

The problem reflects economist Charles Goodhart's observation that when a measure becomes a target, it ceases to be a good measure. Once organizations reward activity rather than results, people optimize the metric instead of the outcome.

What citizens actually experience

Government departments need ways to track digital transformation: cloud migrations, online services, AI deployments. Each provides evidence of change.

The problem, Qureshi said, is that citizens never experience those metrics directly.

"When I want to renew my passport, I want it immediately. I want the ambulance to arrive on time. I want my taxes refunded without problems. I want benefits processed," he said. "That's what matters."

He illustrated the point with an Urdu proverb: "Eat the mangoes, don't count the trees."

Measuring what actually matters

The NHS offers examples of meaningful measurement. AI increasingly helps clinicians identify conditions like cancer and stroke from medical images. But the number of scans processed tells only part of the story.

"If AI is processing one million scans in a year, fantastic," Qureshi said. "But the outcome is whether patients receive faster diagnosis, earlier treatment and better care. Were there fewer false positives and fewer false negatives?"

The same principle applies across government. Rather than measuring chatbot conversations, organizations could track first-contact resolution. Rather than celebrating automated case handling, they could measure how quickly planning applications are approved. Rather than reporting adoption rates, they could focus on citizen satisfaction, consistency, fairness, and waiting times.

These measures are harder to collect than technology statistics, but they reflect the experience of the people public services exist to serve.

AI as means, not end

Governments worldwide face pressure to demonstrate AI progress. New pilots and digital services provide visible evidence of innovation.

But Qureshi believes AI should remain in the background.

"Government should not compete on how much AI it is using," he said. "It should compete on how well citizens are served."

The distinction matters because AI is not the objective of digital transformation—better public services are.

"AI is a means to an end, not the end," he said.

These details were first reported by Automation Watch.