%202.png)
TL;DR
Adapting online research for low-tech populations means redesigning your study methods to work for people with limited devices, expensive data, low digital literacy, or all three. About 43% of Africa’s population lacks smartphone access, and mobile data can cost 30% of a person’s monthly income. This guide defines 30+ terms researchers need to know, from IVR and USSD to channel-native design and audio-first methods, and connects each one to practical field decisions backed by evidence from emerging markets.
Introduction
Standard online surveys assume a lot: a smartphone, a stable connection, comfort with typing, and enough data budget to load a webpage. Those assumptions exclude billions of people. Roughly 43% of Africa’s population still has no smartphone access, and in countries like Zimbabwe, a single gigabyte of mobile data costs an average of $43.75. When researchers ignore these realities, they collect data that represents the connected minority and miss everyone else.
This glossary exists as a working reference for anyone adapting online research for low-tech populations. Whether you’re running consumer studies across African markets, conducting academic fieldwork with migrant communities, or evaluating CX for a brand expanding into emerging economies, the terms and methods below will help you make better design decisions. Each entry explains what the concept means, why it matters, and how it connects to real field evidence.
Explore WhatsApp-native research tools built for these exact challenges.
Section A: Populations and Contexts
These terms define who you’re trying to reach and the barriers standing between them and your survey.
Low-Tech Population
A group of people whose access to, comfort with, or ability to afford digital technology limits their participation in conventional online research. The term covers several overlapping subgroups:
- Device-constrained: People who own only a feature phone, share a single device with household members, or have no personal phone at all. According to GSMA data, sub-Saharan Africa has approximately 750 million mobile connections, but only 250 million are smartphones.
- Connectivity-constrained: People living in areas with intermittent, slow, or prohibitively expensive internet. Rural communities and informal settlements are disproportionately affected.
- Literacy-constrained: People who can make calls and navigate basic phone menus but struggle with reading text or completing digital forms. This includes both text literacy and digital literacy gaps.
- Affordability-constrained: People for whom the cost of a smartphone represents nearly 30% of their salary, making the jump from 2G feature phones to 4G-enabled devices unrealistic.
Adapting online research for low-tech populations starts with understanding which of these constraints apply to your target group. Often it’s a combination.
Digital Divide
The gap between people who have meaningful access to digital technology and those who don’t. It’s not a binary. Someone might own a smartphone but only afford to turn on mobile data a few times per week. Someone else might have connectivity but lack the skills to navigate an app-based survey.
In Africa, the divide is especially stark. In 2019, only 10 out of 45 African countries tracked by the Alliance for Affordable Internet met the affordability threshold of 1GB costing 2% or less of average monthly income. Infrastructure alone doesn’t close the gap. Policies addressing affordability and digital skills matter just as much. For additional Africa-specific data, Yazi maintains a collection of regional data resources worth bookmarking.
Digital Literacy
The ability to find, evaluate, create, and communicate information using digital technologies. In research contexts, it means: can this person understand a survey question on a screen, navigate between questions, and submit a response?
Measuring digital literacy in low-tech populations is harder than it sounds. Researchers at ScienceDirect have noted a “dearth of validated survey measures for capturing digital literacy of populations who have limited prior exposure to technology.” This creates a chicken-and-egg problem: the very tools you’d use to assess literacy require literacy to complete. Audio-based screening and phone-assisted onboarding are practical workarounds.
Feature Phone
A mobile phone with basic calling, SMS, and sometimes a simple browser, but without the touchscreen, app ecosystem, or processing power of a smartphone. Feature phones dominate in many rural areas across Africa, South Asia, and parts of Latin America.
Why it matters for researchers: feature phones can handle SMS and USSD surveys, and they can make and receive calls (supporting IVR and CATI). They cannot run WhatsApp, load web-based surveys, or download research apps. Any method requiring a smartphone automatically excludes feature phone users from your sample. Recognizing this distinction is foundational when adapting online research for low-tech populations.
Shared Device Household
A household where multiple people use the same phone, tablet, or computer. This is common across emerging markets, where one smartphone might serve an entire family.
Research implications are significant. A survey sent to a phone number might be completed by someone other than the intended respondent. Sensitive questions about health, finances, or relationships may go unanswered if other household members can see the screen. Session-based surveys that don’t store visible chat history, combined with clear instructions about who should respond, help mitigate these risks. The Fei et al. working paper from the Joint Data Center flags that “shared phones amongst multiple respondents may affect data privacy and confidentiality.”
Data Poverty
The condition of being unable to afford sufficient mobile data to participate in routine digital activities, including online research. This goes beyond “no internet.” Many people technically have access but ration every megabyte.
The numbers tell the story. Zimbabweans pay an average of $43.75 for 1GB of mobile data. Israelis pay $0.02 for the same amount. Every megabyte your survey consumes is a cost your participant bears. A survey that loads images, plays video, or requires participants to upload media can easily burn through 5 to 10MB, a meaningful expense in data-poor contexts. For practical strategies on keeping costs down, see this guide on low-data-cost research methods.
Last-Mile Connectivity
The final leg of network infrastructure that connects end users to the broader internet or mobile network. In urban areas, last-mile connectivity is usually adequate. In rural and peri-urban areas, it’s often the weakest link: towers are sparse, signals are unreliable, and speeds drop to 2G levels.
For researchers, this means a method that works perfectly in Nairobi or Lagos might fail in a rural district 200 kilometers away. Testing your survey over a 2G connection before launch is a basic but frequently skipped step.
Emerging-Market Research
Research conducted in economies characterized by rapid growth, increasing but uneven technology adoption, diverse languages, and underdeveloped formal sampling infrastructure. The playbook that works in the US, UK, or Germany often breaks down in these contexts.
An MIT thesis on the subject put it directly: “Formal user research approaches designed for conventional markets may not be effective in emerging market scenarios.” Google’s Next Billion Users team, led by researcher Nithya Sambasivan, has documented how people in emerging markets create “infrawork,” improvised workarounds to access technology through shared devices, neighborhood charging stations, and data-sharing arrangements. Understanding these workarounds is essential for designing studies that actually reach participants.
Section B: Data Collection Methods
These are the tools and channels available when adapting online research for low-tech populations. Each has trade-offs.
WhatsApp Survey
A survey delivered and completed entirely within WhatsApp, using automated messages that feel like a chat conversation. Participants tap quick-reply buttons, type short answers, or send voice notes without leaving the app they already use daily.
Evidence on effectiveness is strong. Innovations for Poverty Action (IPA) conducted a randomized evaluation comparing WhatsApp, SMS, and IVR with Venezuelan migrants. WhatsApp surveys had the highest response rates, driven by both higher initial engagement and higher completion rates. However, IPA also notes that WhatsApp surveys “work best with populations that have internet access, smartphones, and some level of technological literacy,” which means they aren’t universal.
Modern WhatsApp survey platforms support complex branching logic, audio-based questions for low-literacy cohorts, and proactive quality controls. They represent the current best option for smartphone-owning populations in markets where WhatsApp penetration exceeds 80%.
Want to see what a WhatsApp survey looks like in practice? Compare features and pricing for WhatsApp-based research tools.
SMS Survey
A survey delivered via text message to any mobile phone, including feature phones. Questions arrive as SMS messages, and respondents reply with numbers or short text.
SMS surveys are the go-to choice in countries where smartphone penetration is low but basic mobile phone ownership is widespread. GeoPoll, a platform focused on African data collection, notes that SMS works well for short, simple surveys but is “constrained by character count, not ideal for complex topics.” You’re limited to about 160 characters per message, which rules out nuanced questions or detailed response options. SMS also lacks the visual formatting, buttons, and media capabilities of WhatsApp.
USSD Survey
A survey conducted over Unstructured Supplementary Service Data, the same technology behind balance checks and mobile money menus (like *123#). USSD works on any mobile phone, requires no internet connection, and doesn’t cost the respondent data.
The key advantage is accessibility: if someone can dial a phone, they can complete a USSD survey. The key disadvantage is that sessions are temporary, usually timing out after 180 seconds of inactivity, and the interface is clunky (numbered menus, no formatting). USSD is best for very short surveys of 5 to 8 questions with pre-defined answers.
IVR (Interactive Voice Response)
An automated phone system that reads questions aloud and captures responses via keypad presses or spoken answers. Think “Press 1 for yes, press 2 for no.”
IVR is the primary option for zero-literacy populations. No reading required, no smartphone required, no data required. It works over basic voice networks. The downside: IPA’s randomized evaluation found that IVR had lower completion rates than WhatsApp, likely because automated voice menus feel impersonal and it’s easy to hang up mid-survey. IVR also can’t capture the depth or nuance of open-ended responses.
CATI (Computer-Assisted Telephone Interviewing)
A phone interview conducted by a live human interviewer who reads questions from a screen and records answers in real time. CATI predates the internet and remains the gold standard for data quality in phone-based research.
The trade-off is cost. CATI requires trained interviewers, scheduling, and supervision. Per-complete costs can be five to ten times higher than automated methods. But for populations where phone penetration exists and neither WhatsApp nor SMS reaches the target group, CATI delivers something automated methods can’t: a human who can clarify confusing questions, build rapport, and probe for deeper answers. For more on how costs compare across these methods, there’s a useful in-country fieldwork vs. WhatsApp cost framework.
AI-Moderated Interview
An automated interview conducted in a chat environment (typically WhatsApp) where an AI system asks questions, interprets responses, and generates follow-up probes in real time. It produces interview-like depth at survey scale.
This is a newer category. The AI adapts its probing based on what the participant says, similar to how a skilled human interviewer would dig deeper into interesting responses. The result is richer qualitative data without the cost of hiring dozens of moderators. Learn more about how AI-moderated interviews work and when they’re appropriate.
Diary Study
A longitudinal method where participants record experiences, behaviors, or feelings over days or weeks. Traditionally done with paper journals or dedicated apps, diary studies increasingly run through WhatsApp, where scheduled prompts and reminders arrive in a channel participants already check multiple times daily.
The challenge for low-tech populations is compliance over time. App-based diary tools require downloads, logins, and persistent data connections. WhatsApp-based diary studies reduce this friction, though they still require a smartphone. For feature-phone-only populations, SMS-based diaries or CATI check-ins at regular intervals are alternatives. Yazi’s WhatsApp diary study tool was designed specifically to handle scheduled prompts and automated reminders in-chat.
Mixed-Mode Research
A study design that combines multiple data collection channels to reach a broader population than any single method could. For example: WhatsApp surveys for smartphone users, SMS for feature phone users, and CATI callbacks for non-responders.
Mixed-mode is common in academic and development research across Africa. BFA Global’s case study on micro and small enterprises in Ghana combined WhatsApp chatbot surveys with in-person onboarding, validating that channel choice drives participation. The complexity lies in harmonizing data across modes, since question formatting and response patterns differ between a WhatsApp quick-reply and a CATI interview.
Voice-Note Research
A qualitative research method where participants respond to questions by recording voice messages instead of typing. The recordings are auto-transcribed, translated if needed, and analyzed like any other text data.
Voice notes bypass literacy barriers entirely. A participant who struggles to type a coherent paragraph can speak fluently for two minutes and convey far more nuance. Practitioners in WhatsApp-heavy markets report that voice notes feel natural since many people already use them daily in personal conversations. The workflow is: capture voice note in WhatsApp, auto-transcribe using speech-to-text, translate if necessary, then analyze. For practical setup guidance, see how to collect voice feedback from participants.
Chatbot Survey
A survey delivered through an automated conversational agent that simulates a human-like exchange. Questions are sent one at a time, responses are processed, and the next question is triggered based on the answer.
Chatbot surveys on WhatsApp “meet customers where they are,” as BFA Global’s Ghana implementation demonstrated. The conversational format reduces the intimidation factor of formal surveys. Participants feel like they’re having a chat rather than filling out a form. Branching logic handles routing behind the scenes.
Section C: Design and Adaptation Strategies
These concepts describe how to modify your study design to work within the constraints low-tech participants face.
Channel-Native Design
Building a research study to function entirely inside the communication channel (WhatsApp, SMS, USSD) rather than sending a link that opens an external browser-based survey. In channel-native design, questions appear as messages, responses are tapped or typed within the chat, and media is captured directly in the app.
This matters because every redirect adds friction. A participant who clicks a link that opens a mobile browser needs to wait for the page to load (consuming data), orient themselves to an unfamiliar interface, and trust a new URL. Practitioners on Reddit and research forums consistently report that external links tank completion rates in emerging markets. Channel-native design eliminates these barriers. It’s the core principle behind adapting online research for low-tech populations who are comfortable with WhatsApp but not with web forms.
Low-Bandwidth Optimization
Designing survey flows to minimize data consumption. This means: no auto-loading images, no video embeds, short question text, compressed media if needed, and quick-reply buttons instead of free-text fields wherever possible.
A concrete benchmark: if 1GB costs $5 or more in your target country, and your survey consumes 5MB per respondent, you’ve just asked participants to spend over 2 cents of their own money to help you. That might sound trivial, but it adds up for people who budget data by the megabyte. Keeping surveys text-based and under 500KB is a reasonable target. For a deeper walkthrough, this article on low-data-cost methods for mobile users covers the details.
Quick-Reply Buttons
Pre-formatted response options that appear as tappable buttons in a chat interface. Instead of reading a question and typing “yes” or “no,” participants tap a button labeled with their answer.
Quick-reply buttons reduce cognitive load, eliminate typos, speed up completion, and make data cleaner (no need to code free-text responses). They’re especially valuable for participants with low digital literacy, who may be comfortable tapping a button but struggle with typing on a small keyboard. WhatsApp Business API supports up to three quick-reply buttons per message, which shapes how you design closed-ended questions.
Skip Logic and Branching
Conditional routing that shows or hides questions based on previous answers. If a participant answers “no” to “Do you own a smartphone?”, they skip all smartphone-related questions and move to the feature phone block.
In chat interfaces, skip logic works differently than in traditional surveys. There’s no visible “page” to skip. Instead, the chatbot simply sends the next relevant question. This feels seamless to the participant but requires careful mapping behind the scenes. Poor branching in a chat survey is worse than in a web survey because participants can’t scroll back or see the overall structure. For setup guidance, there’s a practical tutorial on branching in chat surveys.
Multilingual Research
Conducting studies across populations that speak different languages, sometimes within the same country. Nigeria alone has over 500 languages. South Africa has 11 official ones.
The standard approach is machine translation of survey instruments, with human review for nuance. Participants respond in their preferred language, and responses are consolidated back to English (or another analysis language) for reporting. The risk is that machine translation misses local idioms, slang, or culturally specific concepts. For high-stakes studies, back-translation and local language reviewers are worth the investment. For routine surveys, machine translation with spot-checking is usually sufficient. More on this topic in the guide on running studies across local languages.
Audio-First Design
A study design philosophy that prioritizes voice and audio over text at every stage: audio question prompts, voice-note responses, audio consent forms, and spoken instructions. Text becomes the fallback rather than the default.
This is the single most important design shift when adapting online research for low-tech populations with literacy constraints. Multiple sources in the SERP mention audio-based methods, but few explain the full workflow. In practice: the chatbot sends a short audio clip asking the question, the participant replies with a voice note, the platform auto-transcribes the response, and the researcher analyzes the transcript. It works. And it reaches people that text-based methods simply cannot.
Section D: Recruitment and Quality
Getting the right people into your study and keeping your data clean.
Inclusive Recruitment
Recruitment strategies designed to reach beyond standard online panels and find participants who are typically excluded from digital research. This includes community-based recruiting (through churches, market associations, community health workers), QR codes in physical locations, radio announcements, referral incentives, and partnerships with local organizations.
AnswerLab, a UX research consultancy, advises that “recruiting low-tech skilled participants requires a more holistic approach than a typical study. Build extra layers into your screener, offering different levels of criteria to help identify truly lower-tech skilled participants.” Phone-based screening calls, rather than web-based screeners, are often necessary. Expect to budget more time: what takes a week for a standard online panel might take three weeks for a low-tech population.
Yazi’s audience sourcing provides access to participants across 13 African countries, recruited through social media, partnerships, physical QR code placements, and referral programs specifically designed for emerging-market reach.
Opt-In and Proactive Consent
The requirement that participants explicitly agree to receive messages before a researcher can contact them via WhatsApp. This isn’t just good ethics; it’s a technical requirement of the WhatsApp Business API. You cannot send the first message to someone who hasn’t opted in.
The opt-in process typically involves a participant clicking a link, scanning a QR code, or sending a message to the research number first. Template messages (pre-approved by WhatsApp) are then used for initial outreach. Template approvals can take one to three days, so researchers need to plan ahead.
Selection Bias
The distortion that occurs when your sample systematically excludes certain groups. In digital research, selection bias is the elephant in the room. If you run a WhatsApp survey, you automatically exclude everyone without a smartphone or WhatsApp account.
Fei et al.'s working paper is blunt: “The method is not useful for surveying populations with limited mobile phone usage, WhatsApp familiarity, or digital literacy. The selection bias concerns of this limitation demand careful consideration.” The honest response is to acknowledge these limitations in your methodology section, use mixed-mode approaches when budget allows, and avoid making claims about populations you didn’t actually reach.
Fraud and Quality Controls
Techniques for detecting and removing bad data in mobile research panels. Common problems include speeding (rushing through questions without reading them), gibberish responses, straight-lining (selecting the same answer for every question), duplicate accounts, and professional survey-takers who optimize for incentives rather than honest answers.
Controls include red-herring questions that check attention, response-time analysis, open-ended evidence checks (asking participants to photograph something specific), and identity verification through device fingerprinting or selfie matching. These controls matter more in emerging markets, where incentive payments relative to local income are higher, creating stronger motivation for fraudulent participation.
Panel Recalibration
The ongoing process of adjusting a research panel’s composition to maintain demographic representativeness over time. Panels naturally drift as certain groups churn faster than others. Young urban males might over-represent themselves while older rural women drop out. Regular recalibration through targeted recruitment campaigns keeps the panel useful.
This is particularly challenging in emerging markets where census data is outdated or incomplete, making “representativeness” itself a moving target. The best practice is to define your target population explicitly, monitor panel demographics quarterly, and run recruitment drives aimed at under-represented segments.
Section E: Compliance and Ethics
Working with vulnerable or marginalized populations raises the ethical bar.
GDPR (General Data Protection Regulation)
The European Union’s data protection framework, applicable when research involves EU residents or when data is processed by EU-based organizations. Key requirements include lawful basis for processing (usually consent in research contexts), data minimization, the right to erasure, and breach notification.
Even researchers based in Africa may need GDPR compliance if they work with European clients, process data on EU servers, or study diaspora populations in Europe. For a detailed comparison of data protection frameworks, see the GDPR and POPIA comparison guide.
POPIA (Protection of Personal Information Act)
South Africa’s data protection law, modeled on GDPR but tailored to the South African context. POPIA governs how personal information is collected, stored, processed, and shared. It requires a lawful purpose, limits data retention, and gives individuals the right to access and correct their information.
Researchers conducting studies in South Africa or with South African participants need POPIA compliance. The practical differences from GDPR are subtle but meaningful, particularly around data breach notification timelines and cross-border transfer rules.
Data Residency
The physical location where research data is stored. Some regulations require that data about citizens of a particular country remain within that country’s borders, or at minimum within a specified region.
For researchers working across Africa and Europe, this typically means choosing between EU-based and South Africa-based data centers. The choice affects both compliance and latency. For a summary of how Yazi handles this, see the data security executive summary.
Informed Consent in Low-Literacy Contexts
The process of ensuring a research participant understands what they’re agreeing to, adapted for people who cannot read a standard consent form. Methods include:
- Verbal consent: Reading the consent form aloud over the phone and recording the participant’s spoken agreement.
- Pictorial consent forms: Using illustrations to explain the study purpose, what data will be collected, how it will be used, and the right to withdraw.
- Audio consent messages: Sending a voice note in the participant’s language that explains the study, followed by a quick-reply button to confirm agreement.
- Witnessed consent: Having a community leader or trusted intermediary present during the consent process.
Formal ethics boards increasingly recognize these alternatives. If your study involves low-literacy participants and you’re still relying on a written consent form, you’re creating an ethical problem, not solving one.
Choosing the Right Approach: A Decision Framework
No single method works for every low-tech population. Use this matrix to match your target population’s constraints to the best available data collection approach.
| Population Constraint | Recommended Primary Method | Recommended Backup | Key Limitation |
|---|---|---|---|
| Smartphone + WhatsApp + some literacy | WhatsApp survey or AI-moderated interview | CATI follow-up for non-responders | Selection bias (excludes non-WhatsApp users) |
| Smartphone + WhatsApp + low literacy | WhatsApp with audio-first design and voice notes | IVR for simple quantitative questions | Transcription accuracy in local languages |
| Feature phone only + some literacy | SMS or USSD survey | CATI for complex topics | Character limits; shallow data |
| Feature phone only + low literacy | IVR or CATI | In-person data collection | Cost (CATI); low completion (IVR) |
| Shared device household | Session-based WhatsApp survey with no persistent data | CATI to the intended individual | Privacy risk; identity verification difficulty |
| Severe data poverty (any device) | USSD or IVR (zero data cost) | Reverse-billed SMS | Cannot capture rich qualitative data |
The best studies in emerging markets typically combine two or three of these methods. Start with the lowest-friction, highest-reach channel for your primary population, then layer in alternatives for harder-to-reach segments.
Adapting online research for low-tech populations is not about finding one perfect tool. It’s about designing a system of methods that, together, reach the people who matter to your research question.
Book a demo to see how WhatsApp-native surveys, AI interviews, and diary studies work for emerging-market research.
Frequently Asked Questions
What counts as a “low-tech population” in research?
A low-tech population is any group whose participation in standard online research is limited by device access, internet connectivity, affordability, or digital literacy. This includes feature phone users, people in areas with expensive or unreliable internet, shared-device households, and people who can make calls but struggle with forms or apps. The constraints often overlap.
Can I run surveys on WhatsApp if participants have limited data?
Yes, but you need to design for it. WhatsApp text messages consume very little data compared to web-based surveys. Keep surveys text-based, use quick-reply buttons, avoid image and video uploads unless essential, and keep the total exchange under 500KB. In markets where 1GB costs $5 or more, every megabyte matters.
How do WhatsApp surveys compare to SMS and IVR for response rates?
IPA’s randomized evaluation with Venezuelan migrants found that WhatsApp surveys had the highest response rates, outperforming both SMS and IVR on initial engagement and completion. However, WhatsApp requires a smartphone and internet access, so it reaches a narrower population than SMS or IVR, which work on any mobile phone.
What is channel-native design and why does it matter?
Channel-native design means building your entire research study to function inside a communication platform (like WhatsApp) rather than sending a link to an external website. Participants never leave the app. This eliminates page-load times, data consumption from browser rendering, and the confusion of navigating an unfamiliar interface. It’s one of the most effective strategies for adapting online research for low-tech populations.
How do I handle informed consent when participants can’t read?
Use audio consent messages in the participant’s language, pictorial consent forms with illustrations, or verbal consent recorded over a phone call. Quick-reply buttons (e.g., “I agree” / “I don’t agree”) after an audio explanation work well in WhatsApp-based studies. Most institutional ethics boards now accept these alternatives when properly documented.
What about privacy on shared devices?
Design surveys that don’t store sensitive information visibly in the chat. Use session-based flows that minimize persistent data on the device. Avoid asking highly sensitive questions (health, finances, relationships) through WhatsApp if you know devices are shared. CATI, where the interviewer speaks directly with the intended participant, is more appropriate for sensitive topics in shared-device contexts.
How do I reduce selection bias in digital research across Africa?
Acknowledge it explicitly and mitigate it through mixed-mode design. Combine WhatsApp for smartphone users with SMS or IVR for feature phone users, and consider CATI or in-person methods for the hardest-to-reach segments. Report which populations your methods could and could not reach. No digital method alone can represent the full population in markets where 43% lack smartphones.
Is adapting online research for low-tech populations more expensive than standard online surveys?
It depends on the method. Automated WhatsApp and SMS surveys can be cheaper per complete than traditional CATI or in-person fieldwork. Mixed-mode studies that combine several channels cost more to set up but often less than deploying field teams. The real cost increase comes from recruitment (which takes longer and requires more creative approaches) and quality control (which requires more verification steps). Budget 20 to 50% more time for recruitment compared to standard online panels.
