Run a rapid needs assessment

What You Will End Up With

A short report on priority needs, built from a secondary data review and a limited round of field collection, opening with an executive summary. Start as soon as you can, and within the first two weeks of a sudden-onset disaster at the latest.

A rapid assessment is a broad look at basic needs that points to priorities for assistance. It is not a detailed survey, it is not statistically representative, and it cannot tell you how to design a specific localized intervention. In-depth sectoral assessments come later.

Before You Start

"Good enough" here means a simple approach over a complicated one, not second best. In an emergency it may be the only practical option. It fits best when the emergency is new and sudden, access is relatively stable, the need for information is urgent, and partners will share information and resources.

Settle these before anyone travels:

• The decisions. What has to be decided, who decides, by when, and where the information will come from.
• The plan. Objectives, scope, methodology, analysis and resources, all written down.
• The team. One or two people can run the secondary review, often from outside the affected area. Field collection needs more staff, support and funding.
• The ethics. Follow the CHS commitments on rights and dignity, and on people's primary role in finding solutions.

Keep the analytical framework specific to this disaster and fix it before collecting anything. Teams often collect too much.

Steps

The timeline has two phases. Phase 1 is the first 3 days after onset, a secondary data review that ends in a situation analysis. Phase 2 runs about 2 weeks: roughly 1 week of joint primary data collection, then roughly 1 week of joint inter-sectoral analysis, ending in a report on priority needs.

1. Define decisions and write the analysis plan. Work back from the decisions. Your plan should state what data you need, where you will get it and which comparisons matter, and only then should you design tools. Cover where impact is greatest, who is most vulnerable and which sectors need action now. In most situations, limit disaggregation to two or three categories, such as geography and population group, so sampling stays manageable.

2. Review secondary data. This is a careful desk study drawing on government, NGO, media and other material, and seasoned staff can carry it out fast and from a distance.

   • Work through compiling, organizing, validating, consolidating and analyzing, tagging each item with date, location, sector and reliability.
   • Question every source: the method of collection, the producer, the timing, the purpose, and whether other independent sources agree.
   • Expect outdated, seasonal or contradictory figures, and numbers reported only at national or provincial level.
   • If you use an assistant to tag and summarize non-sensitive documents by those fields, do not assume its output is accurate. Ask it to quote the passage each tag comes from, then check the tags against the original by hand.

3. Choose sites and groups. In the first days and weeks, purposive sampling is the best approach: you pick places and people for a reason. Time, access, security and resources will limit you. Do not visit only the worst-hit areas, which may overestimate impact. Choose a spread, such as coastal and inland or urban and rural. Keep the tool short, and if a department wants extra questions, ask what the data is for and whether it can come from elsewhere or later. Sector standards such as Sphere indicators can help you judge conditions in a given sector.

4. Collect at community or group level.

   • Use direct observation, key informant interviews and community group discussions. In the first two weeks, collecting at community or group level is more realistic than at household level.
   • Disaggregate by sex and age wherever possible, and include questions that let you compare groups.
   • Interview a balanced number of women and men, draw key informants from different groups, and run men's and women's discussions at the same time and in private where you can.
   • Get consent, protect anonymity, and tell people about referral and feedback channels.
   • Manage expectations: repeat visits with nothing visible to show for them breed assessment fatigue and unrealistic hopes. Fewer sites with more time at each gives better data, and you can stop when a common pattern emerges.

5. Analyze, validate and state confidence. Compare field findings with the secondary review. How much confidence you can place in a finding depends on two things: whether experts or decision-makers agree with each other, and how strong the supporting evidence is. Ask each team member to put a personal judgment on paper before the group discusses, so nobody is swept along by the room. Say what the analysis rests on, admit gaps and why they exist, and describe the result as a dynamic but incomplete picture that you will revise. Because the sample is purposive, say plainly that findings cannot be extrapolated to the whole affected population.

6. Report and share fast. Keep it short. Open with an executive summary, use bullets, cite sources and add maps and graphs. Speed can matter more than detail, so share provisional findings and update them. Send them to colleagues, peers in other organizations, coordinators, authorities and affected communities. For security you may need separate internal and public versions and restricted access to data.

Using AI Safely

Treat an assistant as a processing aid, never the authority on what people need.

• Keep identifiable data out. No respondent details, no notes with names, and no locations of people at risk. Your data collection promises consent and anonymity, so keep that promise.
• Respect restrictions. If data is sensitive enough to need restricted access or separate internal and public versions, it does not go into an AI tool.
• Check every output against the source. Accuracy of secondary data cannot be assumed, and the same caution applies to anything an assistant produces. Verify each figure, date and claim in the original document before it reaches your report.
• Label AI-assisted judgments. If an assistant helped you spot patterns, say so, so readers can weigh the evidence properly.

Common Mistakes

Visiting only the worst-hit sites. It may overestimate the impact. Pick a range of site types.

Asking the wrong questions. Skip household-specific questions to individuals, areas bigger than a community, and technical sector questions that non-specialists may misunderstand. Push back on questions added to satisfy a department.

Treating a purposive sample as representative. It is not, and a rapid assessment does not replace a detailed survey. State the limit in the report.

Hiding gaps or waiting for a perfect report. Share provisional findings, name what you could not cover, and update as you learn more.

Repeat visits that change nothing. Multiple visits without visible outcomes create fatigue, so keep the number of visits down.