2. Design phase: Data collection
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et dolore magna.
Why should I do this?
Generate, collect, purchase or scrape the data required to build the AI solution. This includes discovering available datasets, identifying data quality problems, and deriving initial insights into the data and perspectives on a data plan.
- Workflow activities Miro Board
Target users
- CTOs and technical leaders
- Data scientists and ML engineers
- Product managers
- Research organizations, think tanks
Checklists in play
- Checklist 1: Risk Identification & Mitigation (Q3, Q4, Q5)
- Checklist 2: Participatory Co-Creation, Feedback, and Model Refinement (Design Q3, Design Q4)
- Checklist 3: Legal & Regulatory Compliance (Q7)
Questions to consider
Checklist 1 · Q3
What approach(es) do you use to assess and address data fairness on an ongoing basis?
Checklist 1 · Q4
What shows that the collected datasets meet data quality requirements
Checklist 1 · Q5
What technical measures have been put in place to ensure the safety and security of the AI system to mitigate potential risks throughout its lifecycle?
Checklist 2 · Design Q3
Did their participation inform decisions about what data was collected, what labels were used, and what outcomes were optimized?
Checklist 2 · Design Q4
What factors contributed to the inclusion and exclusion of specific data subject/groups
Checklist 3 · Q7
Is there a data protection officer or a data steward ensuring compliance with relevant data protection laws, specifically regarding:
- Consent management and ensuring consent is acquired through transparent, easily accessible frameworks, while allowing revocation of the same by the data subject (i.e., person who grants consent);
- Registration of the organization as a data processor, if required;
- Establishing clear and “lawful” basis (including clearly stipulated purpose for personal data collection and processing);
- Compliance with any restrictions or obligations regarding cross-border transfer of personal data;
- Creating an internal data management policy regarding data security, access, processing, downstream transfers, storage, and deletion of personal data;
- Liaising with the designated data protection authority that governs the data processing activities of the organization in question; and
- Any additional legal obligations that must be fulfilled to process personal data for AI innovation.
Investment types
Every investment project is unique
The application of the six steps will vary accordingly. To provide examples that align with your project, common characteristics of AgDev investments were researched and three ‘investment types’ were developed.
AgriConnect: a digital solutions investment
Existing FAIR datasets were consolidated and assessed for suitability for retrieval and model training.
Metadata and source provenance were reviewed.
Participatory co-creation, feedback, and model refinement
- Are datasets representative across different farming systems and communities?
- Are marginalized groups represented in historical datasets?
Lessons learned and challenges that arose
The team identified that historical datasets disproportionately reflected commercial farming environments and lacked sufficient local language context. It was also observed that certain datasets had unclear licensing terms regarding AI training.
Decisions taken and documentation maintained
Additional local agricultural sources were prioritized to improve representation and an assigned data steward assessed the data's integrity and compliance to data protection laws. Bias assessment findings were formally recorded in a dataset governance report.
AgroThrive: a policy and advocacy investment
Reports from donors, NGOs, ministries, development banks and agricultural programs were consolidated into a searchable repository.
Participatory co-creation, feedback, and model refinement
- Are local or community-led reports represented alongside donor-funded publications?
- Are dissenting, counter positions or minority perspectives included?
Lessons learned and challenges that arose
The team discovered that donor-funded reports heavily dominated the corpus because they were more accessible and standardized. Smaller local studies and informal knowledge sources were underrepresented, creating a risk that the AI would reinforce existing power imbalances in policy discourse.
Decisions taken and documentation maintained
[MISSING – no content in source row assigned to this card]
NGBT: a field research investment
The team consolidated agronomic trial data, extension guides, farmer interviews, nutritional studies and climate resilience reports into a retrieval-ready knowledge base.
Participatory co-creation, feedback, and model refinement
- Are local farming practices and indigenous agricultural knowledge sources represented?
- Are skeptical or critical farmer perspectives included?
- Are farmers aware their testimonials or interviews may be reused by AI systems?
Lessons learned and challenges that arose
The team realized that most available documentation focused on successful trial outcomes and productivity improvements, while less attention had been paid to documenting unsuccessful adoption experiences or implementation barriers.
Decisions taken and documentation maintained
A deliberate decision was made to include limitations and suitability datasets alongside success stories.