Fast take: GiveWell's June 25, 2026 webinar (recorded June 9, 2026) shared four detailed lookbacks that tested prior models and changed funding decisions. This post translates those findings into a practical guide for researchers running qualitative evaluation of grants and shows how to operationalize lookbacks with AI-enabled tools. Read the original recap at GiveWell and see how an end-to-end research platform like Evidano speeds transcript coding, cross-segment comparisons, and secure reporting. In short, lookbacks reveal where models and monitoring diverge; qualitative data often supplies the why. Below you'll find a compact snapshot, what changed, implications for M&E and program teams, and a 7-step workflow you can run in Evidano this month.
Key Takeaways
Evidano is an AI-powered qualitative data analysis platform that helps teams run fast, auditable lookbacks by unifying transcripts, surveys, and monitoring data.
Targeted qualitative work paired with independent quantitative checks often changes impact estimates and funding decisions, as GiveWell's lookbacks in 2025–2026 demonstrated.
- GiveWell's June 25, 2026 webinar presented four lookbacks where qualitative interviews and independent surveys revised reach, uptake, or cost-effectiveness assumptions.
- New Incentives was about 2× more cost-effective than first estimate in the 2020→Lookback 2025 case, while Evidence Action dispensers saw only ≈1/3 expected users per independent surveys.
- A reproducible, auditable 7-step workflow (two weeks) can take teams from raw transcripts to a decision brief using AI-assisted coding, translation, and triangulation.
Findings Snapshot, summary sentence
The table below summarizes four GiveWell lookbacks where qualitative field work and independent surveys altered impact inferences or funding choices.
Findings Snapshot
| Date | Grant / Program | Give Amount (approx.) | Key outcome | Implication |
|---|---|---|---|---|
| 2020 → Lookback 2025 | New Incentives (cash for childhood vaccinations) | GiveWell-directed > $140M total; initial grant $17M | About 2× more cost‑effective than first estimate; ~8, 000 lives saved (lookback) | Cost-effectiveness models underestimated reach, monitor implementer track record & qualitative signals |
| 2022 → Lookback 2026 | Evidence Action; Dispensers for Safe Water | ≈ $65M | Only ~1/3 as many users as expected (monitoring vs independent surveys) | Commission independent surveys, re-evaluate renewal when routine monitoring disagrees |
| 2023 → Lookback 2026 | Against Malaria Foundation, net distributions (DRC) | Grant in 2023 (AMF large-scale funding) | Net campaigns reduced child death risk ≈ 25%, about as cost-effective as estimated | Combine multiple quantitative sources + qualitative interviews to validate causal timing |
| 2016–2022 → Lookback 2025 | Results for Development, government advisory (pneumonia treatment, Tanzania) | $13M across three grants | Amoxicillin-DT stock rose from ~48% to 90%, government financing transitioned | Harder to infer counterfactual, interrogate theory of change for technical assistance grants |
What happened (plain English)
GiveWell expanded lookbacks in 2025–2026 to evaluate whether past grants achieved the impact their models predicted.
Panels at the June 9, 2026 webinar described four lookbacks where additional monitoring, especially qualitative field work and independent surveys, changed the inferred impact or the funding decision.
- New Incentives: unexpectedly high reach and lower per-enrollee cost led to a major upward revision in cost-effectiveness.
- Evidence Action dispensers: independent surveys showed much lower use than routine monitoring, informing a decision not to renew in that context.
- AMF nets: mixed-methods validation (natural experiments + interviews) supported original estimates.
- Technical assistance (RfD): measurable system changes observed but attribution is uncertain, highlighting limits of counterfactual inference.
Implications for researchers: qualitative evaluation of grants
For M&E teams
M&E teams should treat qualitative work as hypothesis-testing: use interviews and implementer conversations to explain model deviations such as reach, cost-per-beneficiary, or uptake.
M&E teams should triangulate routine monitoring with independent surveys and purposive interviews to catch reporting biases.
For program directors and funders
Program directors and funders should plan lookbacks at grant close or midline and decide in advance which indicators will need qualitative explanation if numbers diverge.
Program directors and funders should collect early indicators of government ownership and plausible causal pathways for technical assistance grants, for example relationships and procurement changes.
For qualitative researchers
Qualitative researchers should prioritize depth over breadth for lookbacks, focusing on targeted interviews with implementers, local officials, and beneficiaries to reveal uptake barriers or enabling conditions.
Qualitative researchers should document assumptions that fed original cost-effectiveness models so interviews can directly test those assumptions.
FAQ: Qualitative evaluation of grants
What is a lookback and why does it matter?
A lookback is an evaluation that revisits past grants to test whether the original impact models and monitoring matched real-world outcomes.
Lookbacks matter because they surface differences in reach, uptake, and cost-effectiveness and explain why models diverged from outcomes, using qualitative interviews and independent surveys.
How did GiveWell's lookbacks change funding decisions?
GiveWell's lookbacks changed funding decisions by revising impact inferences when qualitative work or independent surveys conflicted with routine monitoring.
In GiveWell's cases, New Incentives was upgraded for cost-effectiveness, Evidence Action dispensers were downgraded in that context based on independent surveys, and AMF estimates were largely validated after mixed-methods checks.
When should teams run a lookback versus relying on routine monitoring?
Teams should run a lookback when routine monitoring diverges from model predictions or when a grant's assumptions require qualitative explanation.
A lookback is especially useful at grant close or midline, and for technical assistance grants where attribution and counterfactuals are difficult to infer from quantitative data alone.
How can a platform like Evidano support lookbacks?
Evidano supports lookbacks by unifying transcripts, monitoring spreadsheets, and survey exports, applying AI-assisted coding, and producing cross-segment analyses and exportable visuals.
Evidano reduces time-to-decision with features such as auto-transcription, translation, PII redaction, hierarchical codebooks, frequency tables, co-occurrence networks, and secure data handling.
Do more, faster with Evidano (mapped to GiveWell lookbacks)
Problem: scattered transcripts, surveys, and monitoring
Scattered inputs slow analysis, so ingest interview transcripts, monitoring spreadsheets, and survey exports into Evidano for unified analysis.
Problem: slow, inconsistent coding across interviews
Slow coding can be accelerated by importing or building a codebook and using AI-assisted coding to apply hierarchical codes and subcodes, then reviewing and finalizing to reduce initial coding time from days to hours.
Problem: comparing segments (regions, implementers) is manual
Manual comparisons can be replaced by running cross-segment analyses and frequency tables in Evidano to see which implementers or districts explain divergence from model predictions.
Problem: translating multilingual field notes and protecting PII
Multilingual notes and PII concerns can be addressed with Evidano's transcription with custom dictionaries, translation, and PII redaction, plus end-to-end encryption and the commitment that customer data is not used to train third-party models.
Problem: stakeholders want clear narratives and evidence
Stakeholders' need for clear narratives can be met by generating exportable visuals such as co-occurrence networks, word clouds, thematic hierarchies, and clickable quotes to support funding decisions and donor reports.
Checklist: a 7-step lookback workflow you can run in Evidano (2 weeks)
This checklist summarizes a 7-step workflow to run a focused qualitative evaluation of a completed grant within approximately two weeks.
- 1) Gather inputs: upload transcripts, monitoring spreadsheets, survey CSVs, and the original grant model (assumptions).
- 2) Create a short codebook based on model assumptions (reach, uptake, procurement, governance).
- 3) Auto-transcribe and translate audio with a custom dictionary, redact PII.
- 4) Run AI-assisted coding, review disagreements, and lock codes.
- 5) Triangulate: run cross-segment frequency and co-occurrence analyses to find where outcomes diverged.
- 6) Pull representative quotes and generate an evidence matrix (quantitative + qualitative indicators).
- 7) Produce a decision brief and visuals for stakeholders, archive the searchable corpus for future lookbacks.
Wrapping up & next steps
GiveWell's lookbacks, summarized in the June 25, 2026 recap, show that combining targeted qualitative work with independent quantitative checks changes both impact estimates and funding decisions.
For teams running similar evaluations, the priority is reproducible, auditable workflows that move quickly from raw transcripts to evidence-backed recommendations.
- Ready to try this approach? Run a pilot lookback in Evidano with one grant dataset to see how thematic coding, cross-segment analysis, and secure reporting shorten your time-to-decision.
- Start a trial or book a demo at Evidano or Try Evidano for free.