How to Write a ToR That Gets You Useful Research

If we had to name the single biggest reason research engagements go wrong, it would not be methodological. It would not be capacity. It would not even be budget. It would be that the scope of work was unclear from day one, and that the gap between what the client thought they were asking for and what the agency thought they had agreed to do only became visible at week six, when the first draft did not satisfy either side.

We see this from both sides. As clients commissioning research, we have written ToRs we now wince at. As an agency receiving them, we read perhaps a hundred ToRs a year — from NGOs, foundations, governments, intermediaries — and we can usually predict, in the first ten minutes of reading, whether the engagement will produce useful research or produce a relationship-ending disappointment. The signal is almost never the budget. It is the clarity.

This post is an attempt to make the elements of clarity explicit. It is written for the development sector practitioner who has been told "draft a ToR for the X study" and is not sure where to start, and for the foundation programme officer who needs to commission a piece of work and does not want to waste either party's time. It also includes the part everyone needs and almost no one talks about: how to cost a study honestly in India, and what a given budget can — and cannot — actually buy.

The Anatomy of a Good ToR

A good Terms of Reference (call it ToR, SoW, RFP, or "the brief" — the language varies, the substance does not) does seven things at minimum. Most weak ToRs do three or four; almost none do all seven well. The diagram below is the working checklist we use ourselves.

1. The question — written as one sentence

If you cannot write the central research question as one sentence, the study is not yet ready to commission. This is the single most common diagnostic of trouble. "We want to understand the entire ecosystem of women's livelihoods in Bundelkhand" is not a question. "What share of women in our 30-village sample have transitioned out of agriculture into non-farm work since 2018, and what predicts that transition?" is a question. The first will produce a 200-page report you will not read. The second will produce a 30-page report you will use to design your next programme.

2. The purpose — the decision it will inform

State what you will do with the findings, in plain language. "We will use this to decide whether to expand the NRLM-linked livestock programme to three new districts in 2027" is the kind of sentence that anchors everything. Compare it with "to learn more about" or "to inform our strategy" — both of which mean nothing operationally. A study without a decision attached to it is a study that will land on the strategy team's desk, get circulated, get not-read, and become invisible in six weeks. Studies tied to a decision get used.

3. The audience — who actually reads the output

The CEO who needs a one-page memo. The programme team that needs a 40-page report with specific recommendations. The funder who needs a 12-slide deck. The academic community for whom a working paper makes sense. The community partners who need a 4-page Hindi-language back-brief. It is almost always more than one, and the different audiences usually need different artefacts. Saying so explicitly in the ToR is what gets you a deliverables set that works for all of them, rather than one document that satisfies none.

4. The scope — including what is not in the study

The hardest discipline. The temptation is always to expand: "and we should probably also look at…", "while you're in the field could you also…". Resist this in writing the ToR. Be explicit about what is out of scope. "This study does not include cost-benefit analysis", "we are not measuring health outcomes in this round", "the comparison with the Bihar programme is for a later phase." Boundaries protect both parties. Without them, scope creep is guaranteed and the budget will be unrealistic from the start.

5. Method — preferred, or method-agnostic

Two valid approaches. Either: "We need a mixed-methods design with at least 600 household surveys and 30 in-depth interviews across three districts" — clear, specific, costable. Or: "We are method-agnostic; please propose the design that best answers the question within the budget." Both work. What does not work is the middle ground: "Use rigorous methods including but not limited to qualitative and quantitative approaches as appropriate." That sentence means nothing. It will produce a proposal that hedges, an engagement that drifts, and a final report that everyone is mildly unhappy with.

6. Deliverables — be physical and specific

"A comprehensive report" is not a deliverable. "A 30–40 page report in English (PDF + Word source), plus a 12-slide PowerPoint deck, plus a 4-page hindi back-brief for community partners, plus the cleaned dataset and the analysis code" is a deliverable specification. State the language. State the format. State the length range. State who owns the data and code afterwards. State how many rounds of revision are expected. State who signs off.

7. Timeline — with the real reason behind it

"By August 15" is a date. "By August 15, because that is when our board meets to decide the FY2027 strategy and findings need to land in board papers two weeks prior" is a date with a reason. The reason matters because it tells the agency what flexibility exists. If a draft on August 15 is acceptable as long as the final is October 1, that is very different from August 15 being the absolute final. Real reasons let real conversations happen about phasing.

8. Budget — transparent, in a range

The biggest single thing that wastes time in research commissioning is hiding the budget. The reason it gets hidden is the worry that "if we tell them the budget, they'll quote us up to it." This is a real risk, but a smaller one than the cost of not stating a range. Without a budget signal, the agency designs in the dark. They either over-design (and you reject the proposal as too expensive) or under-design (and you accept a proposal that promises what cannot be delivered). Stating "₹8–12 lakh" upfront produces proposals you can actually evaluate against each other. We discuss what each tier buys you in detail below.

9. Evaluation criteria — written down

"We will evaluate proposals on quality of approach (40%), team experience (30%), proposed cost (20%), and ImpactMojo / community linkages (10%)" is the sort of statement that produces proposals you can compare. Without it, agencies will privately conclude that the decision is being made on relationships, on lowest price, or on whatever is in the commissioner's head — and proposals will calibrate to those (worse) signals.

Five Anti-Patterns We See Every Week

Most weak ToRs are not weak because the commissioner is lazy. They are weak because they fall into one of a small number of recognisable patterns. Naming the patterns helps.

What Kind of Research Do You Actually Need?

The second-most-common reason research engagements go wrong is not unclear scope — it is asking for the wrong type of research for the question. A ToR that requests an outcome evaluation when what the client actually needs is process tracing produces a methodologically clean study that does not inform the decision. A quantitative survey commissioned to answer a question about mechanism produces large tables that nobody can act on. A single cross-sectional study commissioned to demonstrate "change over time" produces conclusions that no honest evaluator can defend.

Before you write the methods section of a ToR, walk explicitly through three dimensions. Each has predictable mismatches.

Dimension 1 — What question are you really asking?

Outcome evaluation answers "did it work" and "how much effect did it have." It needs a credible comparison — a counterfactual that tells you what would have happened without the intervention. It produces effect sizes, statistical significance, and (with enough rigour) causal claims. Use it when the decision in front of you is "should we scale this, continue this, or stop this," and when you have a comparison group available — or can construct one through randomisation, matching, or natural experiment.

Process tracing answers "why did it happen" and "what was the mechanism." It needs detailed, time-ordered evidence about how an intervention unfolded. It produces causal narratives, mechanism descriptions, and identification of what worked, for whom, when, and why. Use it when the decision is "how should we redesign this," when the intervention is novel, or when the outcome evaluation is producing puzzling results you need to explain.

Theory-based evaluation combines both, structured around a theory of change. It tests the causal links inside the theory — both whether the outcomes materialised and whether the intermediate mechanisms operated as predicted. The most demanding design and the most informative one. Used by serious large-funder evaluations (USAID, FCDO, Gates Foundation flagship programmes); under-used in mid-tier Indian NGO commissioning, where it would often be the better fit.

The most common mismatch on this dimension: a client wants to know "why our programme works" (a process / mechanism question) but writes a ToR for an outcome evaluation with statistical estimates. The resulting study produces an effect size and tells them almost nothing about why. Or the reverse: a client needs to demonstrate impact to a funder (outcome question) but commissions a thick qualitative case study. The funder rejects the case study as anecdotal.

Dimension 2 — What kind of evidence will be persuasive?

Quantitative evidence is persuasive when the audience wants representativeness, generalisation, or specific magnitudes. It requires a sufficient sample size, valid measurement instruments, and the ability to translate fuzzy concepts (well-being, agency, empowerment) into measurable variables — which is harder than it sounds. Use when the audience is a funder, board, government counterpart, or any reader who needs to be persuaded that the finding generalises beyond the cases studied.

Qualitative evidence is persuasive when the audience wants depth, mechanism, lived experience, or context. It requires careful case selection (typological, extreme-case, deviant, paradigmatic — there is a craft to this), skilled interviewing, and disciplined analysis. Use when the audience is a programme team, the community partners, or any reader who needs to understand how rather than how much.

Mixed-methods combines both — and is the most common choice in modern development research because most real questions need both kinds of evidence. The under-discussed problem is that mixed-methods is regularly under-budgeted: clients pay for the quant component and the qual component but not for the genuine integration work (where the two strands inform each other, contradict each other, and produce findings neither alone could). Without budget for integration, mixed-methods produces two parallel studies, not one integrated study.

Common mismatch: commissioning a quantitative-only study to answer a question about lived experience, mechanism, or meaning. The numbers will not answer it. Reverse mismatch: commissioning a qualitative-only study to demonstrate effect at scale. The cases will not persuade a sceptical funder.

Dimension 3 — Snapshot or change over time?

Cross-sectional studies capture one point in time. They describe "what is now." They are cheaper, faster, and operationally simpler. They are not capable of supporting claims about change unless paired with credible historical baseline data (which usually does not exist).

Pre-post designs measure outcomes before and after an intervention. They are the minimum credible "change" design. The honest interpretation requires either a comparison group (to control for what would have happened anyway) or strong theoretical justification that the change is attributable to the intervention rather than to other factors changing simultaneously.

Longitudinal designs track the same units across multiple rounds. They are the strongest design for causality and developmental questions. They are expensive, operationally demanding (attrition is the killer), and require sustained client commitment that often does not survive funding cycles. India's strongest longitudinal datasets — the India Human Development Survey, Young Lives, the JPAL panel data — exist because someone invested in five-to-twenty-year horizons.

Retrospective designs rely on participant recall to reconstruct a baseline. They are cheaper than true longitudinal designs and weaker on every dimension that matters. Use only when no baseline data was collected and you have no alternative — and always discount the findings accordingly.

The most common mismatch here is the most damaging: a client commissions a cross-sectional study, and then writes about "change" in the report. There was no baseline. The "change" is a comparison between the present and a half-remembered past, or between the treated and untreated groups (which were not equivalent at baseline). The findings are not defensible. They get used anyway, and they do not survive any rigorous scrutiny.

What ₹2L, ₹10L, ₹40L Actually Buy You (India, 2026)

The most underdiscussed part of research commissioning in India is honest costing. Most commissioners under-budget by 30–50% because they have no internal baseline for what specific components cost. Here is the working baseline we use. These are 2026 mid-range market rates for credible research agencies and independent senior researchers in India; informal-sector rates are lower but variable, and top-tier international consulting firms are 3–5× higher.

Where the money actually goes

Inside any ₹10L study, the rough breakdown is:

Cost line	% of total	What it covers
Research team (senior + mid)	35–45%	Design, analysis, writing — 1 senior @ ₹15–25K/day, 1–2 mid @ ₹8–12K/day
Field team & data collection	20–30%	Enumerators (₹800–1,500/day), supervisors, training, piloting, transit, per diems
Transcription & data cleaning	5–8%	Often outsourced; ₹40–80 per minute of audio for Indian languages
Project management overhead	10–15%	Coordination, client communication, ethics review, contracting — usually invisible
Travel & logistics	5–10%	Flights, hotels, vehicles, local logistics
Tools, software, design	2–5%	Survey platform, analysis software, report design, translations
Contingency	5–10%	Field delays, monsoon, ethics re-submission, additional revision rounds

Standard 2026 day rates for credible Indian research agencies and independent researchers:

• Junior researcher (MA, 1–3 years experience): ₹4,000–6,000 per day
• Mid-career researcher (MA + 4–8 years, or PhD + 1–3 years): ₹8,000–14,000 per day
• Senior researcher (PhD + 8+ years, or recognised expert): ₹15,000–30,000 per day
• Principal investigator (sector-recognised, named on grant): ₹25,000–60,000 per day
• Field enumerator (qualified, trained): ₹800–1,500 per day plus per diem
• Field supervisor: ₹1,800–3,000 per day plus per diem
• Transcription (audio → text, Indian language): ₹40–80 per minute of audio
• Translation (Hindi/regional → English, professional): ₹1.50–3.00 per source word

Multiply these rates against the time the work genuinely takes — including the parts that are easy to under-budget: design (a senior researcher's 5–10 days), analysis (a senior plus mid, 15–25 days), writing and revision (10–20 days). A "small" mixed-methods study honestly costed is rarely under ₹6–8 lakh; many quotes lower than that are either being subsidised by the agency, are cutting corners that will hurt the engagement, or are being quoted by less-experienced teams.

Common Budgeting Mistakes

The Pipeline From ToR to Useful Research

Everything above is in service of a process that, when it works, looks roughly like this:

The kickoff conversation that saves the engagement

One specific practice we recommend to every client: budget for a 90-minute kickoff meeting in the first week of the contract, with the full research team plus the client team, where the agency walks back through their understanding of every element of the ToR. Question. Purpose. Audience. Scope boundaries. Method. Deliverables. Timeline. Budget assumptions. Evaluation criteria.

This sounds redundant. It is not. Eight out of ten times, the kickoff surfaces at least one substantive misalignment — usually about scope ("oh, we assumed the regional offices would not be in the sample"), or about audience ("we did not realise the funder also needed to sign off"), or about the deliverable format. Catching this in week one costs nothing. Catching it in week sixteen costs the engagement.

A Working Checklist

If you remember nothing else from this post, remember this: before sending a ToR out, walk through these questions. Most ToRs that pass this checklist produce engagements that work; most that fail it produce engagements that struggle.

One Final Suggestion

Send your draft ToR to one trusted person on the agency side of the field before sending it to formal procurement. Not to compete, not to bid — just to read. Ask them: "If you received this, would you understand what we want? Would you bid?" That single twenty-minute conversation is the most efficient time you will spend on the entire engagement.

The agencies on the other side of these documents are mostly trying to do good work. We see well-resourced commissioners produce bad ToRs all the time. We also see under-resourced commissioners produce excellent ToRs that result in research worth twice what they paid for it. The difference is almost never the budget. It is the clarity. Build that habit, and the work that follows almost takes care of itself.