FGD, KII: what we are talking about
The sector runs on acronyms, and three of them carry most of the meaning:
| Term | What it is | What it brings |
|---|---|---|
| FGD Focus group discussion | A guided discussion between six and twelve people alike enough to speak freely to each other | What is built in the exchange: a disagreement, a shared norm, an objection nobody would have voiced alone |
| KII Key informant interview | A one to one with someone whose position gives an overview | Information a group will not give in front of witnesses, and perspective |
| Individual interview | A one to one with someone affected, not necessarily well placed | The detail of a personal path, an account the group flattens |
When qualitative work is the right tool
It does not replace a survey, it answers different questions. Four situations where it is the right tool, and where a quantitative survey would be of no use:
- A figure is surprising and nobody knows why. Satisfaction drops twenty points between two rounds: the cross tabulation shows the drop, it does not give the cause. A stock out, a change of team, a queue in the sun, a rumour.
- You do not yet know what to ask. Before building a questionnaire, qualitative work surfaces people's actual vocabulary and the categories that matter to them, which are almost never those of the logframe.
- The subject cannot be ticked. Perceived fairness of targeting, sense of safety, power relations within a household: no answer option captures that.
- Accountability. Understanding why a complaints mechanism goes unused takes listening, not counting the complaints received, which are precisely at zero.
Where these tools sit in the full monitoring and evaluation cycle is described on what is MEAL.
The session types
Typing a session is not documentary labelling: the type drives the composition of the group, the choice of facilitator, and the way the summary has to be read. The types used in practice:
- Individual interview, including with a key informant.
- Mixed focus group, when the subject does not suffer from having both groups present.
- Women's focus group and men's focus group, when it does.
- Youth focus group, because a teenager does not speak in front of the village elders.
- Other, with free text: a group of people with disabilities, a management committee, a group of recently displaced people.
Preparing and running the session
Most of it is decided before the discussion starts. What gets prepared:
- The interview guide. Six to ten open questions, from the most general to the most sensitive, with follow ups planned. It is a thread, not a questionnaire to be read out word for word: a discussion that follows the guide to the letter is a poor focus group.
- Consent. Explained out loud, including what will be done with what is said, who will read it, and the fact that participation can be withdrawn at any time. Recording is asked for separately, and can be refused.
- The composition. Participants alike enough to talk to each other, with no hierarchy between them. A mother in law and her daughter in law in the same group, a site manager and the people under him: speech will not be free, whatever the size.
- The place and the time. Somewhere you cannot be overheard from outside, at an hour that does not compete with work or the water run, otherwise the group empties of exactly the people you wanted to hear.
- Two roles, not one. A facilitator who prompts, and a note taker who watches the silences and the glances and does not speak. One person cannot do both.
- The language, handled separately below, because it is the most underestimated parameter of all.
The language of the exchange
An automatic transcript that does not know the spoken language guesses it. And automatic detection goes wrong all the more when the recording is noisy, several people speak at once, and the language is thinly represented in the models. Those are exactly the conditions of a field focus group: all three factors are present every time.
Declaring the language before the session is therefore the single act that weighs most on the quality of the result, at a cost of two seconds. In Opti' it is picked from a catalogue of 230 languages, which covers the international languages but above all those in which focus groups are actually held: Wolof, Fula, Bambara, Dioula, Hausa, Zarma, Mooré, Lingala, Kikongo, Tshiluba, Swahili, Kinyarwanda, Kirundi, Amharic, Tigrinya, Oromo, Somali, Tamazight, Kabyle, Malagasy, Sango, Chichewa, Shona and Zulu, along with the creoles and the dialectal variants of Arabic.
That declaration is used in three places: it is passed to the audio transcription engine, which no longer has to infer it; it is passed to the reading of handwritten sheets, where it stops ambiguous handwriting from being pulled back into plausible French or English; and it appears in the context provided when the summary is written.
From recording to transcript
Three sources feed one session, and they combine: you can record the audio, photograph the note taker's sheet, and paste in an addition typed up that evening.
| Source | What happens |
|---|---|
| Audio | Recording from the device, cut automatically into segments of about five minutes. Each segment is a standalone file, transcribed independently then stitched back together. |
| Photo of a sheet | Photograph of a handwritten sheet or a flipchart. The text is extracted by optical reading, in the declared language. An illegible passage is flagged as such, never filled in with plausible content. |
| Pasted text | Notes already typed up, added straight into the session. |
Why segmenting changes everything in the field
A one hour recording in a single file is a gamble: if the app goes into the background, if the tab closes or the battery dies, everything is lost. Cut into five minute segments, each written to the device before any network attempt, the incident costs the last few minutes at worst.
The rest of the robustness follows the same logic: transcription is queued and runs when a connection is available, a session can be recorded entirely offline then transcribed on return to base, and the app going into the background during a recording does not cut the segment in progress.
From transcript to findings
A transcript is not an analysis. Between the two, two steps that nothing replaces:
- Reread and correct the raw transcript. A misheard name, an off topic passage, an observation from the note taker to add. It is also the moment when the facilitator sees what the machine did not understand.
- Draw out the structure. Recurring themes, points of convergence and divergence between participants, striking verbatims, warning signals. Disagreement is often more informative than agreement: it shows where a line runs through the community.
The summary has to take the session type into account: a women's focus group and a key informant interview do not call for the same reading, and treating them alike erases exactly what separating them was meant to capture.
On the analysis method itself, from verbatim to recommendation, the article advanced qualitative analysis in humanitarian settings goes further than this page: here we define and frame, there the method is worked through.
What qualitative work does not tell you
This is the section missing from most guides, and it is the one that avoids the costliest mistakes in front of a donor.
- No representativeness. Eight people do not represent a population, however carefully chosen. A focus group does not generalise.
- No percentages. Writing that sixty per cent of participants find the distribution fair, when there are eight of them, gives the appearance of measurement to a conversation. If the question is "how many", it is answered by a survey on a calculated sample.
- A selection bias that is almost always there. Those who come are those who are available, mobile, allowed out and told about the meeting. The most vulnerable are often exactly the ones missing.
- A desirability effect. In front of an organisation that distributes aid, people say what keeps the assistance coming. That is rational, and it is corrected through the choice of facilitator and the wording of the questions, not by ignoring it.
The qualitative cycle in Opti'
In Opti Gemba, qualitative sessions live inside the MEAL assignment, alongside the quantitative data, and not in a separate tool. The full chain:
- Create the session with its type, its date, its place, its number of participants, its facilitator, the language of the exchange and its context notes.
- Capture: segmented audio, a photo of a handwritten sheet, pasted text, in any combination, including entirely offline.
- Transcribe, then reread and correct the raw transcript, which is editable.
- Summarise: themes, convergences, divergences, verbatims, warning signals. The summary is editable too.
- Feed it into the report: the summary can be copied, exported to Word or PDF, and picked up in the assignment's narrative report, where it sheds light on the figures from the quantitative side.
A session can be filled in by someone without a full account, through a share link, which makes it possible to involve external facilitators or local partners in the collection. The quantitative side of the same assignment is described on KoboCollect and the KoboToolbox connection.
Frequently asked questions
What is the difference between a focus group and a KII?
A focus group, or FGD for focus group discussion, brings together six to twelve people alike enough to speak freely, and looks for what is built in the exchange: a disagreement, a shared norm, an objection nobody would have voiced alone. A key informant interview, or KII, is a one to one with someone whose position gives an overview: a village chief, a midwife, a site manager. The first gives collective depth, the second gives information a group will not offer in front of witnesses.
How many participants in a focus group?
Six to twelve is the sector's usual working range. Below six, one or two dominant voices carry the whole discussion and the group effect disappears. Above twelve, the quieter participants stop speaking altogether and the transcript becomes hard to attribute. The number matters less than the composition: a group where a mother in law and her daughter in law sit together will not produce free speech, whatever its size.
Why separate women's and men's groups?
Because on many subjects, the presence of the other group changes what gets said. Sexual and reproductive health, violence, control of household money, domestic workload: these are subjects where a mixed group produces a surface consensus. Separation is not an absolute rule, it is a decision taken subject by subject, and it carries the choice of facilitator with it, often a woman for a women's group.
Should a focus group be recorded?
Where consent is given, yes, because manual note taking loses the verbatims and unconsciously selects what confirms the facilitator's expectations. But recording can be refused, and that refusal is respected without discussion: on a protection subject, a recording can be a real risk to participants. In that case the handwritten note sheet remains the source, and it is perfectly usable.
Can percentages be drawn from a focus group?
No, and it is the most frequent mistake. Writing that sixty per cent of participants considered the distribution fair, when there are eight of them, gives the appearance of measurement to a conversation. Qualitative work says why and how, it says neither how many nor how widespread. If the question is how many, it is answered by a survey on a calculated sample, not by a group discussion.
How does Opti' handle focus groups?
Each session lives inside the MEAL assignment concerned, alongside the quantitative data. It carries its type, its date, its place, its number of participants, its facilitator and the language of the exchange. Three sources feed it: audio recorded from the device, the photo of a handwritten sheet read optically, and pasted text. Opti' produces an editable raw transcript, then a written summary that takes the session type into account, editable as well, and picked up afterwards in the narrative report.
Does the module work without an internet connection?
Yes. Every audio segment is written to the device before any network attempt: a dropped connection, a closed tab or a move to the background loses no recording. Transcription is queued and runs when the connection returns. A session can therefore be recorded entirely offline in the field, then transcribed on return to base.
Handle qualitative and quantitative in the same place
Opti' is an all in one MEAL platform covering the whole cycle, from questionnaire design to the final report. It is made of three modules:
OptiBot
The MEAL AI assistant: XLSForm generation, methodological advice and documents (ToR, methodologies).
Find out more →
Opti Gemba
Assignment management: ToR, a Kobo and ODK compatible questionnaire, online and offline collection, focus groups, analysis and report.
Find out more →
Opti Academy
Personalised AI led MEAL training, fitted to your role and your context.
Find out more →