Most AI grants are made one organisation at a time, on the merits of a specific pitch. Most AI portfolios are not designed; they accumulate.
For a funder with five, ten, or fifty portfolio organisations, this is a problem. The same pool of capital ends up underwriting AI experiments at radically different stages of readiness — funding a Stage 1 organisation as if it were Stage 3, or supporting a Stage 4 organisation as if it still needed pilot funding. The result is wasted capital, frustrated grantees, and a portfolio AI strategy that exists only in slide decks.
This page describes the four-dimension, four-stage framework we use with Tilted Ground clients, a stage-specific playbook for what to fund and what not to fund, and the two self-assessments — one for portfolio readiness, one for your own internal team — that this work depends on.
The four dimensions
An organisation's AI readiness is the intersection of four dimensions. The weakest dimension bottlenecks the rest.
Data readiness
Is the organisation's data structured, accessible, and consented? Most nonprofits sit on years of programme data trapped in PDFs, WhatsApp threads, and field notebooks.
Technical capability
Is there anyone in-house, or contractually close, who can scope, build, maintain, and evaluate AI systems? AI literacy at the leadership level is often the binding constraint.
Use case clarity
Has the organisation identified specific, bounded problems where AI is the right tool — not general aspirations to "use AI"? Use case clarity is the strongest predictor of whether a pilot produces learning or noise.
Implementation environment
Is the deployment context — schools, anganwadi centres, district offices, low-bandwidth smartphones — known and designed for? This is where most well-engineered pilots quietly fail.
The four stages
Data infrastructure and technical capability need strengthening before AI deployment will deliver value. This is not a failure. It is an honest starting point that most social sector organisations share — Google.org and Fast Forward's Philanthropic Reset finds that 45% of nonprofits report they are not spending enough on technology, and that more than half of foundations spend less than 5% of their own budgets on technology, cybersecurity, or engineers.
- Data hygiene work
- Leadership AI literacy programmes
- One internal AI champion's capability development
- A landscape audit of what data the organisation has, where it lives, and who controls it
- AI deployment into programme delivery
- Formal partnerships with AI companies
- AI outcome commitments to other funders
One or two use cases are scoped clearly. There is some technical capacity, often through an external partner.
- A single tightly scoped pilot, with its wraparound: an evaluation framework defined before launch, a technical advisor on retainer, structured documentation of what works and what does not
- Good first pilots: AI-assisted research synthesis, meeting transcription with action-item extraction, first-draft report generation, an internal FAQ chatbot over staff-facing knowledge
- Retrieval-augmented generation over an existing knowledge base is almost always the right starting architecture
- Custom fine-tuning is not
Examples in motion: The Google.org Generative AI Accelerator with Fast Forward: 21 nonprofits, up to $2M each, with Google engineer training and AI coach support. Tarjimly cut translation response times in half for nearly 200,000 refugees. Quill delivered a 30% improvement in student writing. Bayes Impact gave caseworkers 25% more time in direct engagement with beneficiaries.
A pilot has worked. The organisation now needs architecture to scale it without scaling the failure modes. This is the most expensive stage and the highest-leverage one — and the one most funders skip, because the deliverables are infrastructural rather than visible.
- Data flows, evaluation pipelines, partnership agreements, second-line technical hires
- AI partnership conversations: Google.org, Anthropic, Sarvam AI, Microsoft
- Government adoption pathway, designed in — not on
- Wadhwani AI's Kisan-eMitra expanding from one PM Kisan use case to three additional schemes
- Qure.ai's TB screening moving from Chhattisgarh validation into Tamil Nadu state-level rollout
- GiveDirectly's anticipatory cash aid to 7,500 people in Nigeria five to seven days before peak flooding
The organisation has shipped multiple AI systems and become a credible thought partner in its domain. Its primary responsibility now is knowledge transfer — not just building more.
- The organisation as an ecosystem node: convening budgets, case study publication, candid documentation of what failed
- Digital Public Good architecture for solutions with replication potential
- AI policy engagement — Stage 4 organisations have evidence that should inform national strategy
- Central Square Foundation's FLN evidence base → NIPUN Bharat national mission
- EkStep as institutional home of the Sunbird Digital Public Goods stack
- Karya, targeting $1B in wages for underserved communities by 2030
The funder's own readiness
The Google.org AI Readiness Playbook for Funders (April 2026) makes the point directly: AI education for your own internal teams must be a top priority, not a side project. A 2025 Project Evident study found that only 36% of funders felt confident assessing the technical feasibility of AI components in grantee proposals.
Two diagnostic patterns to look for inside your own organisation:
High AI usage among staff, but low confidence in choosing the right tool for a given task. The training response should focus on confidence, not access — workflow-specific deep dives, not generic "Intro to AI" sessions.
Strong internal use, weak ecosystem-facing translation. The team uses AI to draft emails but not to evaluate grantee AI proposals or support portfolio AI capacity. The fix is to explicitly link AI fluency to ecosystem outcomes — proposal evaluation, technical due diligence, portfolio-level AI partnerships.
What this changes about how you fund
Three things change once a portfolio is mapped this way.
The unit of funding changes. A Stage 1 organisation needs operational support. A Stage 3 organisation needs infrastructure. The grant instrument should match — sometimes capacity-building, sometimes pilot, sometimes a multi-year platform investment. Fast Forward finds that only 27% of funders currently offer flexible or unrestricted multi-year support; AI readiness work is one of the strongest cases for raising that share.
The timeline changes. Stage 1 to Stage 2 transitions take twelve to eighteen months and look like nothing is happening. Funders accustomed to quarterly milestones will misread this as failure. It is not. It is necessary.
The partnership architecture changes. Stage 3 and 4 organisations should be plugged into AI company partnerships at the portfolio level, not negotiated grantee by grantee. A funder's leverage with a model provider — Google.org, Anthropic, Sarvam — is much higher than any single grantee's. Use it.
The two self-assessments
Portfolio readiness. The instrument we use with clients is sixteen questions: four for each of the four dimensions, scored zero to three, with a combined score that maps to a stage. It takes about five minutes per organisation. It is built for social sector organisations in India and the Global South — the questions are about programme data on WhatsApp and paper, consent frameworks for vulnerable populations, government partners with sovereignty requirements that have not been mapped, and what happens when an AI system produces a harmful output.
Internal team readiness. The Google.org playbook ships a parallel five-minute survey for your own staff, scored on Frequency, Confidence, and Mission Readiness — designed to surface the Confidence Gap and Ecosystem Void described above. We recommend running both. The portfolio instrument tells you where to invest your grants; the internal instrument tells you where to invest in yourselves.
A note we attach to every result: this is a starting point, not a verdict. Your stage can change quickly with the right investments — and different programmes within the same organisation can be at different stages simultaneously.
Further reading
Google.org, AI Readiness Playbook for Funders (April 2026). Google.org and Fast Forward, The Philanthropic Reset: How Philanthropy Can Lead in the Age of AI (2025). Fast Forward with Google.org, 2025 AI for Humanity Report. The Agency Fund, GenAI Evaluation Playbook.
Tilted Ground companion frameworks: Five Questions for Funders, Three Kinds of Scale, and Funding AI in the Global South: Six Case Studies.