Run Your First Field Survey in AISurvey: A Step-by-Step Guide

You have just landed a research contract, you have a team of interviewers, and two weeks to field it — some points in Tashkent, some in villages across the Fergana Valley, and a handful out in Karakalpakstan. This guide walks you through the whole journey in AISurvey: from an empty account to collected, exported data. We will move through eight steps, pausing at each point where the Uzbek field differs from the textbook.

Step 1. Create a project for the study

After signing up, the first thing to do is create a project — a container for a single study. The questionnaire, interviewers, access settings, and all collected responses live inside it. One client contract equals one project; do not dump several waves or several cities into the same one.

Give the project a clear working name — for example, "Service satisfaction — Fergana, June 2026." When an agency runs three or four projects in parallel, a human-readable name saves hours. Sketch out which regions you will work in right away: that determines how many language versions of the questionnaire you need and how you will split the team.

Before you open the constructor, have three things ready: the final instrument text signed off by the client, the sample design (how many interviews, in which regions and mahallas), and the list of interviewers split into teams. The logic of the sample itself, adapted to Uzbek realities — the missing sampling frame, the 2026 census, the mahalla as a cluster — is covered in our overview of sampling methods; settle those questions before you start adding questions.

Step 2. Build the questionnaire in the visual constructor

Open the constructor and add questions. All the field types you need are available: single and multiple choice, text, number, scale, date, plus GPS, photo, and audio — the last three matter for quality control and for capturing context (a shop's signboard, the state of a household, a respondent's spoken consent).

Group questions into logical sections — a screening block, the main block, sensitive topics at the end. The consent block is best placed right at the start: a recording of verbal consent or a marker of written consent is both an ethical and a legal requirement (the "On Personal Data" law requires a lawful basis and the data subject's consent). If you would rather not start from a blank page, take a ready-made template and rewrite it to fit.

A few rules that pay off during the field stage:

• Keep the questionnaire shorter than you want to. In 40 °C heat in the south and in Karakalpakstan, a long doorstep interview means refusals and careless answers in the second half.
• Word things neutrally. Acquiescence bias runs strong in the Uzbek field: a respondent tends to agree so as not to offend a guest. More in our breakdown of questionnaire design principles.
• Decide on scales and open-ended questions early — they are harder to translate and code across three languages.
• Use required fields and validation where it matters — age as a number, phone by a mask — so you do not end up with garbage in the data.

Step 3. Make the questionnaire multilingual

This is not optional; it is a requirement. Build the instrument in at least Russian and Uzbek, add a Tajik version for Samarkand and Bukhara, and a Karakalpak version for Karakalpakstan. Type Uzbek in Latin script, but remember: interviewers in the field must be able to read Cyrillic too, because showcards and a share of respondents are still on it.

In the constructor, every question and every answer option has a field for each language. Fill them in parallel, and always do a back-translation of the key wordings (Russian to Uzbek especially): a machine translation of an "somewhat agree / somewhat disagree" scale almost always breaks the meaning. In the field, the interviewer switches the questionnaire language to match the respondent with a single tap — starting in Russian in Tashkent, switching to Uzbek in a village near Andijan, and to Tajik in a Samarkand mahalla, the neighborhood community.

Give showcards separate thought. Uzbek is officially in Latin script, but older and rural respondents often read Cyrillic more comfortably — that matters for self-administered blocks and scale cards. We cover the fine points of translating and localizing scales in our piece on open versus closed questions.

Rule: the language of the questionnaire should match the respondent's language, not the interviewer's. One interviewer may run interviews in three languages in a single day — give them all versions in one instrument.

Step 4. Set up skip logic

Conditional branching (skip logic) shows the respondent only the relevant questions. The classic example: a block about car brand opens only for those who answered "yes" to "Does the household own a car?"

This is also where you build in Uzbek selection realities. Because of labour migration — millions of men work in Russia and Kazakhstan, especially from the Fergana Valley — daytime households skew heavily toward women, the elderly, and children. If you specifically need an adult man or a randomly selected household member, write the selection rule (a Kish grid, "last birthday") directly into the questionnaire logic, or the interviewer will survey whoever is convenient. While you are at it, define clearly what counts as a "household": large, multi-generational families are common in Uzbekistan, and without an explicit definition the interviewer will count their own way.

Skip logic also lets you carefully route around sensitive topics where appropriate — for instance, in the politically sensitive context of Karakalpakstan. And keep the quota effect in mind: if you give an interviewer a quota that is easy to fill with "convenient" respondents (youth by the metro, neighbors), that is exactly what they will fill — so hard selection rules inside the questionnaire are more reliable than verbal instructions.

Step 5. Publish and onboard interviewers

Once the questionnaire is ready, publish it and invite interviewers by access code. No app to install: the interviewer opens a link in their phone's browser, enters the code, and the instrument is there. This removes an enormous headache around a mixed fleet of devices and low-end phones.

Hand out codes to teams by geography and language:

1. Tashkent — the most Russified and digital team, short travel legs.
2. Fergana Valley (Andijan, Fergana, Namangan) — dense population, mostly Uzbek-language interviews, account for migration.
3. Karakalpakstan — long distances, the Karakalpak version, a separate briefing and logistics.

Do not forget the gender mix: in conservative and rural households a male stranger is often not received, and sensitive topics (health, family) almost always call for female interviewers.

A short briefing before the start removes half of the field's problems. Walk the team through every screen of the questionnaire, talk through how to switch language, how the GPS is captured, and why an interviewer must never "fill in" a questionnaire from memory in the evening. Remind them about access too: a visit to the mahalla chairman (raisi) and an official letter from the client or the hokimiyat sharply reduce the share of refusals and the risk that someone calls the local police officer. This is the operational side, which our separate piece on field quality control is devoted to.

Step 6. Pilot offline and run the field

Offline mode is the whole reason digital CAPI exists in Uzbekistan. 4G coverage drops in remote mahallas, in the mountains of Surkhandarya, in the Karakalpak desert, and simply in stairwells. The interviewer downloads the questionnaire to the device in advance, runs the interview fully offline, and the data syncs by itself once a signal returns — no response is lost. Why this is more reliable than paper is covered in our guide to offline collection.

Before the full launch, always collect 10–20 pilot questionnaires with real interviewers. A pilot catches what you cannot see at your desk: a confusing Uzbek wording, a broken branch, a GPS reading that will not lock in a basement, branches that open the wrong block. Look at the pilot data as if it were final: distributions on the key questions, interview duration, how fully the open-ended fields are completed. If something looks odd at twenty questionnaires, it will be a disaster at a thousand.

Plan the field calendar too. During Ramadan, daytime energy is low and evenings are taken by iftar — interview windows shrink. In autumn the cotton harvest and the spring planting pull rural residents into the fields, and people are harder to find at home during the day. Throughout the field, the interviewer records a GPS point and, where needed, a photo — your anchor for the quality control that follows.

Step 7. Monitor collection in real time

While the team is in the field, the monitoring dashboard shows how many questionnaires have been collected, how the load is distributed across interviewers, and where and when interviews took place (by GPS and timestamp). For a geographically dispersed team — Tashkent, the valley, Karakalpakstan — this is the only way to avoid waiting for the end of fieldwork to learn that something went wrong.

What to check daily:

• Suspiciously fast interviews — a possible sign of fabrication.
• GPS points that do not match the assigned mahalla.
• A skewed sample — for example, almost all women (migration again) or a single interviewer who stands out from the overall pattern.

Real-time monitoring gives you one more advantage: you can stop or redirect an interviewer while they are still in the field, instead of receiving a thousand spoiled questionnaires at the end. Noticed that one district is systematically failing to close its quota? Move a neighboring team there today, not next week. We cover back-check methodology and fabrication signals in detail in our piece on field quality control. The logic of the sample itself is in our overview of sampling methods for Uzbekistan.

A pre-launch checklist for your first field study

Most first-project failures are baked in the night before launch, not in the field. Before you hand out access codes, run through a short list — it costs half a day and saves a whole wave.

Check at your desk

• Translation and back-translation. Have a native speaker read the Uzbek version aloud, and the Tajik and Karakalpak versions for Samarkand, Bukhara, and Karakalpakstan. Verify the back-translation of the key scales: "somewhat agree" must mean the same thing in all three languages.
• Skip logic tested. Walk every branch: answer "no" where a block should close and "yes" where it should open. Check the within-household respondent-selection rule too (a Kish grid, "last birthday") — migration leaves mostly women and the elderly at home during the day. A hidden broken branch on a thousand questionnaires is a lost block of data.
• Required fields and validation in place. Age as a number, phone by a mask, an upper bound on open numeric fields — so that "250 years old" never reaches the export.
• Showcards matched to language and script. Put scales and option lists on cards in the language and script the respondent actually reads: Latin for most, Cyrillic for older and rural respondents, separate cards for the Tajik and Karakalpak routes.

Check in the field

• A 10–20 form pilot run and the data reviewed. Not an "all good" report, but the real distributions, interview durations, and how fully the open-ended fields are filled.
• Interviewers onboarded by access code and trained on offline sync. Each one should, once, download the questionnaire, run an interview offline, and watch the answers sync when a signal returns — before they head to a distant district.
• GPS and consent enabled. Confirm that the GPS point is being recorded and that the consent block sits first and is required.
• Access cleared, devices charged. Arrange a letter from the client or the hokimiyat and a visit to the raisi in advance. On the hot southern and Karakalpak routes, give every pair a power bank: in 40 °C heat both battery and signal drain fast.

Rule: if you cannot honestly tick a line, the field is not ready to launch. Better to lose a day now than a wave later.

Common first-project mistakes to avoid

These are the rakes we see teams step on when they first move from paper to digital in Uzbekistan. Every one is cheap to prevent and expensive to cure.

What not to do

• Fielding only a Russian version. It may pass in Tashkent, but in the Fergana Valley, Karakalpakstan, Samarkand, and Bukhara a single Russian questionnaire means systematic undercoverage and broken representativeness. Build the Uzbek version, plus Karakalpak and Tajik where needed, from the start.
• Assuming connectivity at the doorstep. 4G coverage drops in remote mahallas, mountains, and the desert. If the interviewer has not downloaded the questionnaire in advance, they will either lose data or start "filling it in" from memory in the evening. Offline mode is the norm, not the exception.
• Skipping the pilot. The temptation of "the team is experienced, let's go" costs thousands of spoiled questionnaires. Twenty pilot interviews almost always catch something.
• Ignoring the raisi and hokimiyat access step. Without a visit to the mahalla chairman and an official letter, the refusal rate climbs and someone may call the local police officer. It is not bureaucracy; it is your pass into the field.
• Changing the questionnaire mid-collection. Any edit to a wording or option after launch makes the "before" and "after" data incomparable. All changes belong in the pilot; once you launch, the questionnaire is frozen.

If your first project clears these five traps, the rest is field discipline, which we cover in our piece on field quality control. When you are ready, build your questionnaire in the constructor.

Step 8. Export the data to Excel and Word

When collection is complete, export the results to Excel — each questionnaire becomes a row, each question a column, a ready dataset for analysis and pivot tables. Open-ended answers come out as-is, ready for coding, and the GPS coordinates and timestamps stay in the data for a final check. For the client report, export to Word — a structured document that only needs your conclusions added. No manual re-typing from paper and none of the errors that come with it — which means an entire class of data-entry typos disappears as well.

That closes the loop: from an empty project to data in hand. Ready to try it — build your first questionnaire in the constructor or browse the blog for methodology breakdowns tailored to the Uzbek field.