The European Commission's Project Cycle Management Guidelines (2004) set out the problem tree as part of the analysis stage of the logical framework approach, ideally built in a participatory workshop. The group picks a starter problem, puts its direct causes below and its direct effects above, and keeps asking what causes each one. The tree is then turned into an objectives tree by rewriting each negative situation as a positive achievement.

Workshops produce plausible trees, and a plausible tree can still be wrong. Each causal arrow should be checked against what data exists: NFHS for health and nutrition, ASER and the National Achievement Survey for learning, PLFS for work, the Census for demography, and administrative data from the scheme itself. Where no data exists, a short diagnostic study (a few focus groups, a rapid household survey, observation in five schools) is cheaper than a programme built on a wrong cause.

India's Janani Suraksha Yojana, launched in 2005, paid cash for giving birth in a health facility. Lim and colleagues (The Lancet, 2010) found it raised antenatal care and facility births, and in their matching analysis associated JSY payment with 3.7 fewer perinatal deaths per 1,000 pregnancies. They also called for attention to the quality of obstetric care in facilities, a supply side that the cash did not reach.

A national average hides the places where a problem is worst. NFHS-5 (2019-21) puts stunting among children under five at 35.5 per cent for India, against 46.5 per cent in Meghalaya, 42.9 in Bihar, 39.7 in Uttar Pradesh and 39.6 in Jharkhand, and 23.4 in Kerala and 25.8 in Goa. Design for the distribution of the problem: where it is concentrated, among which groups, and whether the same cause operates everywhere.

Rewriting each cause as a positive condition gives an objectives tree. The design team then chooses which branches to act on. No programme addresses every cause. The choice depends on what evidence says can be changed, what other actors already cover, the organisation's comparative advantage and the budget. Writing down which causes you are deliberately leaving to others makes the programme's boundary visible to funders and partners.

In India the last mile of most social programmes runs through a small number of frontline workers: the ASHA, the anganwadi worker, the auxiliary nurse midwife, the teacher, the gram rozgar sahayak. Each new programme tends to add a register, an app or a meeting to their week. Nepal's Female Community Health Volunteer programme, run since 1988 under the Ministry of Health, and Pakistan's Lady Health Worker Programme, launched in 1994, both show how much a health system comes to depend on this cadre.

Participation can mean very different things, from a consultation meeting after decisions are made to shared control over the budget. Be specific about which decisions participants will influence: the problem definition, the choice of intervention, the selection criteria, the timing of activities, the grievance process. Kerala's Kudumbashree, set up in 1997 as the State Poverty Eradication Mission during the devolution to panchayats and the People's Plan Campaign, built its structure on neighbourhood groups of 10 to 20 women, federated into area development societies at ward level and community development societies at local government level.

A study shows that an intervention worked somewhere. Design asks whether it will work in this district, through this organisation, for these people. Muralidharan and Prakash (AEJ: Applied Economics, 2017) found that Bihar's bicycle programme for girls continuing to secondary school raised girls' age-appropriate enrolment by 32 per cent and cut the gender gap by 40 per cent. The gains came mostly in villages farther from a secondary school, which the authors read as a fall in the time and safety cost of getting to school.

Suppose a team wants to design a livelihoods programme for the poorest households in a district. A short review would find BRAC's Targeting the Ultra Poor programme in Bangladesh, the six-country trial published in Science in 2015 by Banerjee, Duflo and colleagues (Ethiopia, Ghana, Honduras, India, Pakistan and Peru), and the long-run Bangladesh study by Bandiera and co-authors in the Quarterly Journal of Economics (2017).

Designers tend to search for success stories. Evaluations that found no effect are at least as useful because they remove options. The Hyderabad microcredit evaluation found no significant rise in consumption and no significant change in health, education or women's empowerment 15 to 18 months after branches opened, and very few differences two years later. Credit may still help in other ways. A design that promises poverty reduction through microcredit alone needs a stronger argument than the evidence currently gives.

Teams that consider one option tend to fund it. List at least three distinct ways of addressing the chosen causes, including doing less, working through an existing scheme, and a cash alternative. Then compare them on the same criteria. The comparison often shows that the most familiar option is neither the cheapest nor the most likely to work.

A useful discipline is to ask whether the programme would do more good than giving its budget to participants as cash. Muralidharan and Prakash (2017) report that Bihar's bicycle programme was much more cost-effective at raising girls' secondary enrolment than comparable conditional cash transfer programmes in South Asia. The cash benchmark does not always win. It forces the designer to state what an in-kind or service programme adds.

The delivery model should carry the mechanism. If the mechanism is trust and repeated contact, as in many health behaviour programmes, a community worker who lives in the village fits. If the mechanism is rapid, uniform transfer of money, a digital payment system fits. If the mechanism is changing what teachers do in classrooms, the channel has to reach teachers through the education department's own mentoring and supervision structure.

The graduation approach was adapted across countries: the asset offered varied from livestock to small trade inventory, while the combination of asset, consumption support, training and coaching was kept. Document every adaptation and the reason for it, so an evaluator can tell what was tested.

A delivery choice that looks neutral can have different effects for women and men. Training scheduled at midday excludes women doing household work. Transfers paid into a household head's account usually go to a man. Asset transfers of livestock add to women's unpaid work unless fodder and care are planned for. Bangladesh's Female Secondary School Stipend programme, introduced in 1994, paid the stipend into girls' own bank accounts and conditioned it on attendance, marks and remaining unmarried.

Pakistan's Benazir Income Support Programme was launched in July 2008. Its National Socio-Economic Registry was built through a door-to-door survey using a Poverty Score Card based on a proxy means test, covering 27 million households in 2010-11, and eligibility was set by a threshold on the score. More than 85 per cent of those households were resurveyed in the 2021 update, after which three in four beneficiaries came from the bottom two expenditure quintiles.

Many Indian schemes identify beneficiaries from lists compiled years earlier. PM-JAY's initial list of about 10.74 crore families came from the Socio-Economic Caste Census of 2011, which was already seven years old at launch in 2018. Households that became poor after the survey, new households formed by marriage or migration, and people missed by the survey are excluded until the list is updated.

A person can be eligible and still not receive a programme because of the steps between eligibility and receipt: documents, a bank account, a working mobile number, biometric authentication, travel to a block office, knowing the scheme exists. These steps fall hardest on people who are already marginal: migrants, people with disabilities, older people, women without documents in their own name, homeless people.

Between every pair of boxes in a theory of change sits an assumption. Trained teachers will use the method. Parents will send children in harvest season. The block office will release funds on time. Women will control the asset. Designs fail more often at these arrows than in the boxes. List each assumption, rate how much evidence supports it, and rate how much damage it would do if it were false.

A budget translates the design into resources. Reviewers read it to check whether the plan is serious: whether there are enough staff to deliver the dosage, whether monitoring and evaluation are funded, whether participants' costs are covered, whether safeguarding has a line. A budget built by inflating last year's figures does not test the design. Building it bottom up from activities does.

Cost per output (per training held, per kit distributed) is easy to compute and says little about value. Cost per outcome (per child reading, per girl enrolled, per household out of extreme poverty) lets the design be compared with alternatives. At design stage the effect is an estimate taken from the evidence review, so the cost per outcome is a range. Comparing options on that range is still far better than comparing them on cost alone.

A design meant for government adoption has to be costed at government norms as well as at the NGO's own prices. Honoraria for frontline workers, training allowances, travel rates and material costs are fixed in scheme guidelines and state orders. If the pilot pays mentors twice the government rate, or uses printed materials the department's budget head cannot fund, the pilot's results describe a programme the state cannot buy. Build two columns in the budget from the start.

Some funders now pay for outcomes. In the Educate Girls development impact bond in Rajasthan, the UBS Optimus Foundation provided US$270,000 of upfront capital, Educate Girls delivered the programme in 166 government schools, an independent evaluator measured results, and the Children's Investment Fund Foundation repaid the investor according to them. About 80 per cent of payments rested on learning gains, measured by a clustered randomised trial, and about 20 per cent on enrolling out-of-school girls.

Indian law shapes what a programme budget can contain. Under the Foreign Contribution (Regulation) Act 2010, as amended in 2020 with effect from 29 September 2020, section 7 prohibits transferring foreign contribution to any other person, and section 8(1)(b) caps administrative expenses met from foreign contribution at 20 per cent in a financial year unless the Central Government approves more. Corporate funding under section 135 of the Companies Act 2013 must fall within activities in Schedule VII.

A risk is something that may happen and would affect the programme or the people it touches. At design stage, list risks across several categories, rate each for likelihood and impact, and decide a response: avoid it by changing the design, reduce it, transfer it (insurance, contract terms), or accept and monitor it. The output is a risk register owned by a named person, reviewed on a fixed schedule.

Programmes can harm the people they aim to help. Cash transfers can raise the risk of violence in some households. Microfinance groups can pressure members over repayment. Targeting can stigmatise. Volunteers given access to children can abuse them. Collecting data on caste, religion, HIV status or sexuality can expose people. A design-stage harm assessment asks, for each activity, who could be harmed, how, and what would prevent it.

A programme in India that works with children must be designed around legal duties that apply to its staff and leaders. Under section 19 of the Protection of Children from Sexual Offences Act 2012, any person who knows or apprehends that an offence under the Act has been or is likely to be committed must report it to the Special Juvenile Police Unit or the local police. Section 21 makes failure to report punishable, with a longer maximum term for a person in charge of an institution or company who fails to report about a subordinate.

Protection from sexual exploitation and abuse (PSEA) addresses the risk that staff, volunteers or contractors use their position over participants, who often depend on them for a benefit. The risk is highest where staff decide who receives something of value: selection for a scheme, a loan, relief goods, a job. Design can reduce it by removing single points of discretion, publishing criteria, and making sure participants know that the benefit is free and how to report abuse.

Programmes collect personal data: names, phone numbers, Aadhaar numbers, health status, caste, photographs. The Digital Personal Data Protection Act 2023 and the DPDP Rules 2025 impose their duties from 13 May 2027, so a programme running past that date has to design data collection now with a lawful basis, notice, purpose limitation and security. From the same date section 17(2)(b) exempts processing for research, archiving or statistical purposes when its conditions are met, which helps evaluation but does not cover routine programme operations.

The implementation plan converts the design into a sequence of work. It sets out the phases (preparation, pilot, expansion, consolidation), the activities in each, who is responsible, the resources needed and the dependencies between them. In South Asia, it also has to fit the calendar: agricultural seasons, monsoon, school terms, festivals, examination months, the government financial year from April to March, and the model code of conduct before elections.

Fidelity, the degree to which the programme is delivered as designed, falls as programmes grow, mostly because supervision gets thinner. Plan the ratio of supervisors to frontline staff, how often supervisors visit, what they check, and what support they give. Plan for turnover: in many NGO field teams a sizeable share of staff leave each year, and every departure takes training and relationships with it.

A pilot tests whether the design can be delivered and whether its riskiest assumptions hold, before money is committed at scale. It should be run under conditions close to those at scale: with the staff, partners and budget per participant that would be used later. A pilot run by the best staff, with extra funding and close attention from the founders, will look better than the programme can be.

Monitoring data in many programmes flows upward for reports and is rarely used by the people who could act on it. Design the monitoring system around decisions: which decisions will managers and field staff make each month, and what information do they need for each? A short list of indicators that are checked and acted on is worth more than a long list that fills a dashboard nobody opens.

Many grants fix activities, targets and budget lines for three years, which makes adaptation a compliance problem. At design stage, negotiate the terms that allow learning: outcome-level targets with flexibility on activities, a budget reallocation threshold (for example up to 10 or 15 per cent between lines without prior approval, Illustrative), scheduled review points, and a learning budget for small tests.

Every programme ends, or changes hands. The design should say what happens then: whether the activity continues through government, a community institution or a market, or whether the programme is time-limited by design, as graduation programmes are. Exit planning shapes design choices from the start. A programme that intends handover to the education department should use the department's staff, schedules and cost norms from the pilot onward.

In India, Pakistan, Bangladesh, Nepal and Sri Lanka the state runs the schools, health systems, social protection and rural employment programmes that reach most poor households. An NGO programme that reaches a few thousand people is valuable mainly if it changes what those systems do. Designing with government in mind from the start changes many design choices, compared with hoping for adoption after a successful pilot.

Rural employment in India now runs under the Viksit Bharat Guarantee for Rozgar and Ajeevika Mission (Gramin) Act 2025, which replaced MGNREGA from 1 July 2026. According to the Government of India's backgrounder (PIB, December 2025), it guarantees 125 days of wage employment per financial year to rural households whose adults volunteer for unskilled manual work, is planned through Viksit Gram Panchayat Plans, lets states pause works for up to 60 days at peak sowing and harvest, and gives each state a normative allocation, with spending beyond it borne by the state.

JEEViKA began in 2007 as the World Bank supported Bihar Rural Livelihoods Project in 42 blocks of six high-poverty districts, mobilising women into self-help groups, village organisations and cluster-level federations. The World Bank's completion report (2017) records that the National Rural Livelihoods Mission, designed in 2010, was to a large degree based on the Bihar model and on other World Bank supported livelihood projects. Bihar's own design had drawn on Andhra Pradesh's experience of building local institutions.

The WHO and ExpandNet guide Nine steps for developing a scaling-up strategy (2010) distinguishes horizontal scaling up (expansion and replication to new places and people) from vertical scaling up (institutionalisation through policy, budgets and legal change). It asks designers to plan actions that increase scalability from the start, using its CORRECT checklist: credible, observable, relevant, relative advantage, easy to install, compatible with the user organisation, and testable.

Effects found in pilots often shrink when programmes grow. Bold, Kimenyi, Mwabu, Ng'ang'a and Sandefur (Journal of Public Economics, 2018) studied contract teachers in Kenya: teachers hired on a fixed-term contract by an NGO raised test scores, while teachers on identical contracts hired by the government produced no effect. The authors trace the gap to implementation constraints and political economy forces set in motion as the programme went to scale.

Muralidharan and Niehaus (Journal of Economic Perspectives, 2017) argue for experimentation at scale: larger sampling frames, more treated units and larger units of randomisation, run through the systems that would deliver the policy. Evaluating inside the government system answers the question a policymaker asks: what will happen if the state itself does this, across whole districts, with its own staff and budgets.

An NGO in a district of eastern Uttar Pradesh is asked by a CSR funder to design a three-year programme to improve reading in government primary schools. This example is Illustrative and used to walk through the design steps. The team first gathers data: an assessment in 40 schools finds most Class 3 to 5 children below Class 2 reading level, teacher vacancies in about a fifth of schools, and sharp attendance drops in the wheat harvest.

The evidence review points to level-based instruction (TaRL), with two delivery models: camps run by NGO instructors and a government model with teachers supported by mentors. The team compares three options, with costs taken from budget estimates and effects from published evaluations adjusted down for expected fidelity loss.

The plan has three phases. In year one the model runs in one block with monthly review of three indicators: share of scheduled level-based sessions held, mentor visits per school, and the share of children moving up a reading level each term. A decision point at the end of year one uses agreed rules. Years two and three expand to the other blocks if the rules are met, with a district-level evaluation designed with the education department.

The Udaipur nurse attendance programme is a design failure in a precise sense. The intervention changed nurses' incentives through monitoring and pay deductions, and in the first months attendance rose by 15 to 29 percentage points according to J-PAL's summary. The design relied on the local health administration to enforce deductions, and that administration had its own reasons not to. Monitoring machines were damaged and absences were excused.

For years microcredit was presented as a route out of poverty for poor women. The Hyderabad evaluation found more business investment and higher profits among businesses that already existed, and no significant rise in average consumption. Programme designs that promised poverty reduction through credit alone set targets the mechanism could not reach, and monitoring systems counted loans disbursed and repayment rates, which said nothing about well-being.

A frequent failure in South Asia is the parallel programme: an NGO or donor project that hires its own workers, runs its own registers and reporting, and pays incentives the government cannot match. It can show strong results while funded. When funding ends, nothing in the government system has changed, frontline workers have been drawn away from routine tasks, and the community has learned that services arrive and disappear with projects.

A pre-mortem is a short exercise run before the design is finalised. The team is told to imagine that three years have passed and the programme has failed badly. Each person writes down, independently, the most likely reasons. The reasons are then pooled, grouped and checked against the design: is each one addressed in the risk register, the assumptions, the monitoring plan or the budget? The exercise surfaces doubts that people hesitate to raise in a planning meeting.