Quick answer: Mixed methods research deliberately combines qualitative data (the why — interviews, open-ended answers, observed behavior) with quantitative data (the what and how many — surveys, rating scales, metrics) in one coordinated study, so each method validates and explains the other. Neither alone is enough: a survey can tell you that 30% of trial users never activate, but not why; five interviews can tell you exactly why, but not whether it generalizes. In 2026, the fastest way to run a mixed methods study is an AI-moderated interview platform like Koji — it asks scale and choice questions for the numbers and probes every answer with adaptive open-ended follow-ups for the story, in the same conversation, then analyzes both automatically.
What is mixed methods research?
Mixed methods research is the deliberate, coordinated use of both qualitative and quantitative methods to answer a single research question. The goal is triangulation — using multiple data types to converge on a more credible, more complete answer than any one method could produce. As the Nielsen Norman Group puts it, mixed methods create "richer, more actionable insights" precisely because the methods cover each other's blind spots.
The two halves play different roles:
- Quantitative answers what, how many, and how often. It is broad, statistical, and generalizable — survey scores, NPS distributions, feature-adoption rates, conversion funnels. It tells you the size of a problem.
- Qualitative answers why and how. It is deep, contextual, and explanatory — interview transcripts, open-ended responses, observed workflows. It tells you the cause behind the number.
If you are still deciding which side of the spectrum a given question belongs on, start with our guide to qualitative vs quantitative research. Mixed methods is what you reach for when the honest answer is "both."
Why mixed methods beats single-method research
The case for combining methods is strongest where single methods fail most visibly.
Surveys measure the gap but hide the cause. Pendo's widely cited feature-adoption research found that roughly 80% of features in the average software product are rarely or never used. A quantitative dashboard surfaces that 80% — but it cannot tell you whether those features are undiscoverable, irrelevant, or simply badly onboarded. Only qualitative probing can. Run the numbers and the conversation together and you get both the size and the source of the problem.
Triangulation is the dominant design for a reason. In a systematic review of mixed methods studies, triangulation was the single most common design, used by about 74% of studies, far ahead of embedded (14%), sequential explanatory (8%), and sequential exploratory (5%) approaches. When a survey shows users are dissatisfied with onboarding and interviews reveal repeated confusion on the same screen, the two findings reinforce each other — a far more compelling, defensible insight than either signal alone.
Modern data collection makes the qualitative half cheap. The historical reason teams defaulted to surveys was cost: interviews did not scale. That has flipped. Conversational, in-app research formats now reach up to 85% completion, versus just 10–15% for traditional static web surveys — and AI-moderated interviews capture open-ended depth at survey-like scale. The trade-off that forced teams to pick one method has largely dissolved.
The four mixed methods designs (and when to use each)
Mixed methods is not one thing. Choose the design that matches your sequence and intent:
- Convergent (triangulation) — Run qual and quant at the same time, then compare. Best when you want to validate a finding from two independent angles. Example: a satisfaction scale plus an open-ended "what is the one thing we should fix?" in the same study.
- Explanatory sequential (quant → qual) — Start with numbers, then interview to explain them. Best when a metric surprised you. Example: activation dropped 8 points last quarter; you survey to size the segments, then interview the affected cohort to find the cause.
- Exploratory sequential (qual → quant) — Start with interviews to discover the themes, then build a survey to measure how widespread they are. Best for new problem spaces. Example: a dozen discovery interviews surface three recurring pain points; you field a survey to rank them across thousands of users.
- Embedded — One method sits inside the other as a supporting strand. Example: a primarily quantitative concept test with a few open-ended "tell us more" probes attached to each rating.
For a deeper menu of the qualitative side, see qualitative data collection methods.
How to run a mixed methods study, step by step
- Write one research question, not two. "How satisfied are users with onboarding, and why?" is a single mixed-methods question. Keep the why and the what bound together.
- Pick your design from the four above based on what you already know.
- Design one instrument that carries both. Combine scale and choice questions (quant) with open-ended questions (qual). The trick is to attach a probe to each number — a 0–10 score followed by "what would move that up two points?" turns a metric into a diagnosis.
- Collect at scale. Field it to enough people that the quant half is meaningful (often 100+ responses) while preserving qualitative depth on every one.
- Analyze both, then triangulate. Aggregate the structured values into distributions and charts; code the open-ended answers into themes; then overlay them — do the themes explain the distribution?
- Report the convergence. The headline insight is where the number and the story agree (or revealingly disagree). For the mechanics, see how to analyze survey data and our research synthesis guide.
The old way vs the AI-native way
| Traditional mixed methods | Koji (AI-native) | |
|---|---|---|
| Instrument | Separate survey tool + separate interview round | One AI-moderated conversation with 6 structured question types |
| Qual depth at scale | Manual interviews, ~5–15 people | Adaptive AI follow-ups on every respondent, hundreds at once |
| Analysis | Export to spreadsheets, hand-code transcripts | Automatic thematic analysis + structured aggregation |
| Time to insight | 2–6 weeks | Hours |
| Bias | Moderator and instrument bias | Consistent, neutral AI moderation — no moderator bias |
Where Koji fits: mixed methods in a single instrument
Most tools force the split — a survey platform for the quant, a separate interview or transcription tool for the qual, and a manual merge at the end. Koji collapses that into one conversation. Its six structured question types are mixed methods by design: open_ended questions capture the qualitative story (with AI follow-up probing), while scale, single_choice, multiple_choice, ranking, and yes_no capture the quantitative, chartable values. A single Koji study can ask an NPS-style 0–10 scale question, then immediately probe "you said 6 — what would make it a 9?" — triangulation built into the flow.
After collection, Koji automatically aggregates the structured answers into distributions and charts and clusters the open-ended responses into themes with verbatim quotes, producing a one-click report that shows the number and the reason side by side. What used to take weeks of fielding a survey, recruiting interviews, and hand-coding transcripts now takes hours — typically 10x faster — with no research expertise required. And because only conversations that pass a quality bar (scoring 3+) consume credits, you pay for signal, not noise.
Run your first mixed methods study with Koji
If you have ever shipped a survey and immediately wished you could ask "but why?", you already need mixed methods. Koji lets you ask both in one AI-moderated study — the scales for the size of the problem, the open-ended follow-ups for the cause — and turns the combined data into a triangulated report automatically. From question to insight in hours, not weeks. Start your first mixed methods study free →
Related reading: Surveys vs Interviews: When to Use Each · How to Analyze User Interview Data · The Death of Static Surveys
