Experiment
Autonomous Evidence Scout
AI identifies what evidence is missing for an important question and finds the cheapest credible way to close the gap.
01
Problem to address
Defra can have large volumes of data and still lack the specific evidence needed for a decision. Data collection is often system-driven rather than question-driven.
02
What the evidence tells us
Some outcomes cannot be measured, so policy is sometimes made within risk. Important ecological knowledge sits outside Defra. Analysts report missing end-to-end data flows and stale information.
03
The proposition to explore
The idea
Start with the decision question, identify the evidence that is missing, and find the lowest-cost credible way to close the gaps that matter.
04
Hypothesis
If Defra starts from the decision question and explicitly identifies missing evidence, it can focus collection on gaps that materially change confidence instead of accumulating more data indiscriminately.
05
How it could work
Given a question, the scout identifies required evidence, checks what already exists, assesses confidence and proposes the lowest-cost route to close important gaps, including using existing visits, partner data or targeted research.
06
Example: bring it to life
For “Is peat restoration working here?”, the scout finds good intervention and satellite data but weak ground observation. It proposes three observations for the next scheduled site visit and identifies an existing local monitoring source.
07
AI opportunity
Evidence gap detection, source discovery, uncertainty assessment and orchestration of collection opportunities. AI should propose, not silently collect or repurpose data without governance.
08
Potential value
- More purposeful data collection
- Faster closure of high-value evidence gaps
- Better use of existing field interactions and partner data
- Less collection of information that is never used
09
Reduce uncertainty without overcommitting
Smallest meaningful experiment
Take one live policy question and run the evidence-scout process manually with AI assistance. Compare its proposed evidence plan with the current approach and validate with domain experts.
10
Open questions and risks
What we still need to learn
- May overstate whether external data is fit for purpose
- Legal/ethical reuse constraints
- Needs a usable data catalogue and quality metadata