"Scale" is used to mean a dozen different things in the social sector. The word is so overloaded that funders, founders, and practitioners often find themselves agreeing in language and disagreeing in substance.
This framework maps the three distinct types of scale we see in the field, what each one actually requires to achieve, and where technology — particularly AI — has begun to change the geometry of what is possible. It draws on Kevin Starr's Scale Really Matters (Stanford Social Innovation Review, 2026), Anish Malpani's horizontal-versus-vertical refinement, and our own experience advising Indian social sector organisations through the AI inflection.
The three types
Steady Impact
Deep, local work. More impact requires more money, always. Local NGOs running night schools, community health camps, grassroots legal aid clinics, tribal nutrition centres.
Who it's for. Funders who want proximity to specific people and places, and accept that the cost of impact does not fall over time. This work does not move the needle on population-scale problems — that is not its purpose, and demanding that it do so misreads what the work is for.
The role of technology. Modest. Tools that reduce administrative load — digital case management, simple data collection — help, but they do not change the fundamental cost structure.
Linear Growth
Organisational scaling — more services for more people. An NGO expanding from three districts to fifteen. A training programme growing its batch sizes. A diagnostic service replicating its clinic format across cities. Akshaya Patra adding centralised kitchens.
Who it's for. Funders comfortable with predictable, proportional growth and who value the operational maturity that comes with steady expansion. Doubling reach requires roughly double the resources, with diminishing returns as the marginal site is often harder to set up than the first.
The role of technology. Operational. AI-augmented programme management, retrieval over institutional knowledge, decision support for field staff — these compress the marginal cost of each new district. They make linear growth more efficient. They do not, on their own, make it exponential.
Exponential Impact
The curve of impact detaches from the curve of cost. The idea travels through systems, governments, and norms — not through the organisation's own delivery.
Who it's for. Funders willing to fund slow, uncertain, often invisible work with a long time horizon — and to resist the urge to push the work into premature expansion.
What it requires. A scalable solution, a doer at scale (usually government, sometimes a market institution), and patient capital. All three. Remove any one and the model collapses back into linear growth.
Indian examples. CSF's FLN evidence base absorbed into NIPUN Bharat — now running across 1.2 million schools. Rocket Learning's WhatsApp architecture approaching 35% of national early childhood care enrolment. CEGIS research adopted by Chief Secretary offices across multiple states.
The two axes of exponential scale
Anish Malpani's addition to Kevin Starr's framework: exponential scale comes in two forms, requiring different strategies and different funder relationships.
Horizontal Scale
Same intervention, broader reach. More of a proven thing reaching more people. Compounds through replication.
Mid-Day Meal scheme reaching 120 million children daily. Vaccine reminders through mMitra. Salt iodisation and fortification programmes. Once validated, the gain comes from running it across more sites and populations.
Vertical Scale
A new economic logic that others adopt because you've made it undeniable. Compounds through legitimacy — governments run the model because the evidence is no longer ignorable.
BRAC's graduation approach adopted across 40 countries. India's FLN mission. Pratham's Teaching at the Right Level shaping curricula beyond the original organisation. ASER's assessments forcing foundational literacy onto the national agenda.
Both are exponential. Both produce outcomes at population scale. But the work to get there looks completely different — and confusing one for the other is one of the most consistent sources of strategic error in the sector.
Where AI changes the geometry
For most of the social sector's history, the limiting factor on exponential scale was not idea quality. It was the cost of distribution — getting a proven intervention into the hands of frontline workers, parents, citizens, and government officials in their language, on their phones, in their workflow. AI is starting to compress that cost.
Voice and language as a distribution layer
Population-scale outreach in twenty-two Indian languages used to require twenty-two parallel programmes. Sovereign voice AI platforms — Sarvam, AI4Bharat models, BHASHINI — collapse this into one. The "Listen at Scale" deployments by Sarvam and EkStep: 1.4 million senior citizens contacted for a single scheme, with a 42% increase in daily enrolments; 414,000 persons with disabilities profiled in a single wave. The voice layer is to this decade what mobile penetration was to the last.
Digital Public Goods as horizontal-scale infrastructure
The Sunbird stack underneath DIKSHA, the ABHA architecture under ABDM, the ONEST registry — these are the rails on which horizontal scale now runs. An intervention designed to be DPG-compatible from day one inherits national reach. An intervention designed in isolation has to build it. DPG-compatibility is no longer a technical preference; it is a scale precondition.
AI as a vertical-scale accelerant
Retrieval-augmented generation over government knowledge bases, automated evidence synthesis, signal aggregation across portfolios — these shorten the time between an intervention working in the field and the evidence becoming visible to the policy actors who can adopt it. Vertical scale used to be measured in decades partly because evidence took decades to assemble. AI compresses that bottleneck without compressing the political work that follows.
The point is not that AI scales everything. It does not. AI does not turn Steady Impact work into Exponential Impact work. What it does is reduce the cost of the parts of exponential scale that used to be expensive — distribution, language, evidence synthesis — and make a wider set of organisations capable of crossing the threshold.
What is most often missed
Timeframe. Systems change is multi-decade, not five to ten years. Funders who push vertical-bet organisations into premature expansion break the thing that makes the model replicable. The fastest way to fail at vertical scale is to be measured on linear-growth metrics.
Cost structure. Organisations built for government adoption should eventually look like the governments they are trying to enable. UK-anchored leadership running Indian interventions, expatriate cost structures embedded in Indian programmes, donor-language reporting that the eventual government partner cannot read — these are structural drags on both cost and local agency. The scale an organisation can achieve is bounded by the institutional shape it carries.
Acknowledgements & Further reading
Framework draws on Kevin Starr (Mulago Foundation), Scale Really Matters, Stanford Social Innovation Review, 2026; Anish Malpani (Without), LinkedIn response, April 2026. Adapted for the Indian social sector context by Shruti Keerti, Tilted Ground.
Tilted Ground companion frameworks: AI Readiness for Funders, Five Questions for Funders, and Funding AI in the Global South: Six Case Studies.