Experiment

Next Best Attention

Of everything I could look at today, what deserves my attention most?

OperationsNow -> Next
KNOWREMEMBERUSEEXCHANGELEARNANTICIPATE

01

Problem to address

High-volume operational teams cannot inspect everything. Connecting records and making a case easier to understand helps once someone already has it in front of them. It does not answer the first question: of thousands of records, which few deserve that scarce attention today?

02

What the evidence tells us

There are 12 million unique data points. Teams need to flag unusual cases, see what to verify, and know where to direct effort first.

03

The proposition to explore

The idea

Give operational teams a transparent ranked shortlist of the cases or records that most deserve attention now, with an explanation of why.

04

Hypothesis

If operational staff receive a transparent, ranked shortlist of records that most deserve attention, scarce expertise can move from broad scanning to targeted investigation without requiring black-box prediction.

05

How it could work

Combine a small number of inspectable indicators, such as unusual change from baseline, conflicting records, known risk conditions and previous outcomes. Rank cases and show why each appears. Capture the result of investigation to improve future triage.

06

Example: bring it to life

Instead of 2,700 records requiring review, a regulator opens the day to 12 priority cases: 3 unusual changes from history, 4 cross-record inconsistencies, 2 repeat patterns and 3 time-sensitive risks, each with an explanation.

07

AI opportunity

AI is optional. Start with inspectable rules, thresholds, historical comparison and simple anomaly methods. AI can later combine signals and explain prioritisation.

08

Potential value

  • Better allocation of scarce attention
  • Faster investigation of high-value cases
  • Less repetitive scanning
  • A practical LEARN → ANTICIPATE feedback loop

09

Reduce uncertainty without overcommitting

Smallest meaningful experiment

Run the ranking retrospectively against one period of known cases using only transparent rules. Ask regulators whether the shortlist would have changed where they spent time.

10

Open questions and risks

What we still need to learn

  • False positives can create workload
  • Poor thresholds can systematically deprioritise important cases
  • Need a clear way to challenge or override ranking