Experiment
Personalised Defra
Move from asking people to find the right service or scheme to proactively showing what is relevant to their circumstances.
01
Problem to address
Users navigate generic guidance and must infer how national rules apply to their specific circumstances. They often cannot see what they are eligible for or what consequences follow until late in the journey.
02
What the evidence tells us
Farmers want to know what they are eligible for before applying, and whether an intervention will work on their land. That can often be calculated from data Defra already holds.
03
The proposition to explore
The idea
Use permissioned context about a person, organisation or place to surface the services, interventions or actions that are most relevant to them, with reasons and uncertainty.
04
Hypothesis
If Defra can contextualise current rules and evidence around the person, organisation or land, users can make better-informed choices earlier and avoid entering unsuitable journeys.
05
How it could work
Start from a permissioned contextual view of the user and relevant assets. Combine current rules, known constraints and evidence to show relevant options, explain why, and expose uncertainty or conflicts.
06
Example: bring it to life
A farmer sees: “Four interventions may apply. A and B fit these parcels; C conflicts with an existing agreement; evidence for D on this soil type is weak.” They can explore the reasons before committing.
07
AI opportunity
AI can translate complex rules into contextual explanations, compare options and surface likely relevance. Deterministic eligibility rules should remain rule-based where possible.
08
Potential value
- Less guidance-search burden
- Fewer unsuitable applications
- Potentially better intervention choices
- Earlier, more transparent decisions
09
Reduce uncertainty without overcommitting
Smallest meaningful experiment
Prototype personalised recommendations for one scheme using a small number of known rules and static farm profiles. Test comprehension, trust and decision quality against current guidance.
10
Open questions and risks
What we still need to learn
- Recommendations may be mistaken for guarantees
- Need rigorous rule/version management
- Personalisation must not create unfair or opaque access