From Manual to Predictive

In March 2025, the Prime Minister said: "No person's substantive time should be spent on a task where digital or AI can do it better, quicker and to the same high quality and standard."

He's right. And right now, across Whitehall and beyond, that's exactly what's happening.

The Government's own State of Digital Government Review laid out the scale:

• Only 47% of central government services offer digital pathways
• £26 billion spent on digital and data in 2023, but less than 20% on permanent staff
• Legacy systems make up nearly a third of the IT estate, many rated "high risk"
• 50% of digital and data roles unfilled in 2024
• The DVLA still opens 45,000 physical envelopes every day

The National Audit Office estimates fraud and error cost taxpayers £55-81 billion in 2023-24. Data analytics could tackle this, but "savings so far have been modest compared to the amount potentially achievable."

This isn't a technology problem. The technology exists. It's an execution problem.

Data lives in multiple systems. Some modern, some legacy, some technically unsupported but somehow still critical. Getting it out means different logins, different formats, different quirks. One system gives you CSV. Another gives you XML. A third gives you a password-protected Excel file by email.

Then the real work begins. Date formats don't match. Names are inconsistent: Ministry of Defence, MoD, MOD, M.O.D. Codes that should match don't. Fields that should be populated aren't.

Cleaning takes hours. Sometimes days. By Friday, the data is ready. You've got maybe a day to actually analyse it before the next cycle begins.

The AI Opportunities Action Plan describes "agentic systems" as the next frontier: "systems that can be given an objective, then reason, plan and act to achieve it."

That sounds abstract. Here's what it means practically:

• Ingestion agents pull from multiple sources, handle different formats, adapt when things change.
• Cleaning agents standardise dates, normalise names, deduplicate records.
• Enrichment agents match records across datasets, pull reference data, fill gaps.
• Summary agents produce analyst-ready outputs with notes on what's changed, what's unusual, what needs human attention.

The human doesn't disappear. The agent does mechanical work and flags judgement calls. The analyst reviews, validates, investigates flagged items, and does the actual thinking.

When you remove six hours of preparation from someone's week, you don't just get six hours back. You get capacity for work that wasn't happening at all.

McKinsey estimates data and analytics could create $1.2 trillion annually in value across public and social sectors globally. But that value comes from what people do with freed capacity, not from the automation itself.

Prediction is pattern recognition pointed at the future. You observe that certain conditions precede certain outcomes. You notice those conditions occurring. You infer the outcome is more likely.

Experienced analysts do this intuitively. Predictive systems do it at scale, continuously, without getting tired or forgetting patterns.

This is why automation comes first. Without it, prediction stays in the sandbox.