LocationsAI Value Chain
The Data-to-AI Value Chain is a framework describing the sequence of technical, institutional, financial, and governance activities through which data is transformed into AI that creates public value. It encompasses five interdependent stages. ELM acts across all five as the governance and coordination layer that enables the chain to function equitably. Click any stage to explore.
Left to itself, the chain produces the failures documented in our LLA programmes: extraction without consent, curation without community voice, training without access, deployment without sustainability, and impact without return. ELM sets norms, aligns incentives, secures government commitments, and builds the shared infrastructure that no single actor can build alone.
What Breaks Here
- FPIC is not always adhered to — consent, compensation and provenance tracking remain inconsistent
- Communities have no visibility into downstream use of their data or who benefits from it
- Women, elderly, rural and dialect speakers systematically excluded from recording and design
- Contribution not sustained beyond initial project — incentives remain unclear for most communities
How ELM Acts
- Through government commitments that set data sovereignty as a national standard
- By convening actors around shared norms and enforceable FPIC protocols
- By mobilising funding for community-led collection that individual programmes cannot sustain alone
ELM Intervention
- Sets data sovereignty norms for all publicly funded collection
- Helps scale new methodologies, tools and platforms
- Orchestrates sustainable funding for community data collection
What breaks here
- Annotation decisions made by external actors — no community review
- Once data enters a corpus, the trail back to the community disappears
- No shared standards — each programme rebuilds curation from scratch
- Translationese trap: datasets translated from English, not native-authored
How ELM acts
- Through shared standards that travel across borders and programmes
- By anchoring community review rights in procurement and policy
- Building peer networks that spread best practice without reinventing it each time
ELM intervention
- Establishes shared curation standards across the ecosystem
- Safeguards community review rights
- Mainstreams community-centred best practices
What breaks here
- Limited funding for compute — what exists is mostly credits, not local infrastructure investment
- Compute is not localised: hardware capabilities and the soft skills to operate them are rarely built in-country
- Every LLA programme rebuilds the same pipelines from scratch — no shared infrastructure
- Community data trains commercial models with no value return to contributors
- Technical depth dangerously thin — one researcher leaving means knowledge lost
How ELM acts
- Through coordinated advocacy and funding mobilisation for open, locally owned infrastructure
- By brokering access to shared compute for under-resourced teams while pushing for local capacity
- By establishing accountability norms around how models are documented, licensed and reused
ELM intervention
- Supports fine-tuning resources for local languages
- Shares infrastructure and compute for model evaluation and research
- Advocates for localised compute investment — hardware capability and the skills to use it
What breaks here
- Critical funding gap — grants end before products reach citizens
- Governments not spending on AI in local language public services
- Gap between prototype and production large and consistently underestimated
- Communities excluded as co-designers — leading to low adoption
How ELM acts
- Through government deployment commitments that create an institutional home and operational budget
- By setting equitable access standards that prevent over-charging communities
- By embedding community feedback loops into governance frameworks
ELM intervention
- Supports public service integration commitments
- Promotes equitable access pricing
- Develops governance frameworks with community feedback loops
- Defines purpose limitation requirements
What breaks here
- No value-return mechanisms to communities who contributed data
- No community satisfaction measurement in any current LLA country
- Impact tracking has almost zero dedicated funding
- The loop never closes — communities see no return
How ELM acts
- Through accountability mechanisms requiring governments to report on value return publicly
- By building shared channels for benefit-sharing that no single programme can maintain
- By closing the feedback loop so impact evidence shapes the next round of collection and investment
ELM intervention
- Provides shared channels for value-return reporting
- Develops community-centred impact assessment frameworks
- Ensures feedback loops back to the collection stage
These stages are not a one-way flow. Deployment experience feeds back into curation. Community responses shape what gets built next. Impact assessments influence funding. ELM builds these feedback loops explicitly.