Qualitative coding: what it is and 5 approaches explained

Qualitative coding means assigning labels to segments of text to mark what each segment is about. You then use those labels to find patterns across your dataset. The five main approaches are: descriptive coding (what is happening), in vivo coding (the participant's own words), process coding (sequences of action), structural coding (topic areas), and inductive coding (themes that emerge from the data). Skimle automates inductive and deductive coding across entire datasets, so researchers spend their time interpreting findings rather than reading transcripts line by line.

What is qualitative coding?

Coding is the step between collecting qualitative data and making sense of it. You read through your transcripts, documents, or open-ended responses, and wherever you find something relevant, you apply a label — a code — that describes what that passage is about.

A code might be a short phrase: "trust deficit," "price sensitivity," "fear of disruption," "desire for autonomy." Over dozens of interviews, these labels accumulate into a structure that lets you see patterns: how many participants mentioned a concern, how it connects to other concerns, whether it appears in some segments of your sample but not others.

Coding is not tagging. A tag is a topic marker; a code captures meaning. "Onboarding" is a tag. "Implementation complexity as a barrier to adoption" is a code — it says something about what onboarding means to your participants.

For the relationship between coding and the broader analytical process, see our guide to thematic analysis, which explains how codes eventually build into themes.

What are the 5 main qualitative coding approaches?

Different coding approaches serve different purposes. Choosing the right one depends on your research question, your data, and whether you are exploring or testing.

1. Inductive coding

In inductive coding, you come to the data without a predetermined framework. You read each passage and ask: what is this about? You generate codes in your own words, staying close to the data. The codes are not decided in advance.

This approach suits exploratory research where you want themes to emerge from participants' perspectives rather than from your own expectations. It is the most common approach in academic qualitative research and in any context where you are genuinely uncertain about what you will find.

Skimle's automatic thematic analysis uses inductive analysis by default, grouping segments by shared meaning without a predefined framework.

2. Deductive coding

In deductive coding, you start with a framework — a set of categories, a theoretical model, a predetermined codebook — and apply it to the data. You are asking: where does this passage fit within my existing structure?

This suits research that extends or tests existing theory, or practical analysis where you already know what you are looking for. A consultant reviewing expert interviews to assess a specific set of strategic risks is working deductively.

For deductive coding in Skimle, see predefined categories, which allows you to define your category structure before analysis begins.

3. In vivo coding

In vivo coding uses the participants' own words as codes. If a participant says "we just kept hitting walls," the code becomes "hitting walls." This preserves the texture and voice of the data, which can be valuable in research where the language participants use is itself analytically significant.

In vivo codes often surface unexpectedly striking phrases that capture something important about participants' experience. They are typically used alongside other coding approaches rather than as a standalone method.

4. Descriptive coding

Descriptive coding assigns a topic label to a passage: what is this passage about, at the most basic level? "Working conditions," "family support," "career trajectory." Descriptive codes summarise without interpreting. They are useful as a first pass on complex or unfamiliar data, or as a precursor to a second, more interpretive coding cycle.

See our dedicated post on descriptive coding in qualitative research for a practical walkthrough.

5. Process coding

Process coding focuses on action and sequence: what is happening, and in what order? Process codes use gerunds — "seeking support," "resisting change," "negotiating boundaries." This approach is common in grounded theory methodology, where understanding the social processes at work is central to the analysis.

How do inductive and deductive coding compare?

	Inductive	Deductive
Starting point	Data	Theory or framework
Codes come from	Patterns in the data	Predefined categories
Best for	Exploratory research	Testing or extending existing theory
Risk	Missing structure	Missing unexpected findings
Common in	Academic qualitative research	Consulting, policy, applied research

A third approach, abductive coding, moves between inductive and deductive reasoning — you start inductively, encounter something surprising, and then develop or revise theoretical explanations. For a full comparison of all three, see inductive, deductive, and abductive coding: when to use each.

What is a codebook and when do you need one?

A codebook is a reference document that defines each code in your study: its name, a definition, inclusion and exclusion criteria, and an example passage. Well-maintained codebooks serve two purposes.

First, they impose consistency. In a large dataset, without a codebook, the same passage might receive different codes depending on when you encounter it. The codebook anchors your coding decisions.

Second, they make your methodology transparent. Academic researchers, in particular, need to explain how they coded. A codebook provides that audit trail. For studies with multiple coders, the codebook is what enables inter-rater reliability checks.

You do not need a codebook for quick, exploratory coding. But for any formal research project, any dataset with multiple coders, or any work that will be published or audited, it is worth building one from the start.

What is the difference between a first and second coding cycle?

Many qualitative researchers use two coding cycles.

In the first cycle, you code broadly. The goal is comprehensive coverage — you want to have labelled every passage that might be relevant. At this stage, you will probably generate too many codes: similar codes, overlapping codes, codes that felt important in the moment but turn out to be marginal.

In the second cycle, you consolidate. You look at the full list of codes and group them: which codes are really describing the same thing? Which codes cluster around a shared theme? What are the patterns? Second-cycle coding is where you move from individual labels to interpretive structure.

This two-cycle approach is built into the standard thematic analysis process. It is also how Skimle's AI-assisted analysis works: an initial pass generates segments and codes, which researchers then review, merge, and reorganise through the categories view.

How is AI changing qualitative coding?

Manual coding is time-consuming. A single hour-long interview transcript might run to 8,000 words. Coding 40 interviews at that length — a common sample size in academic qualitative research — means working through 320,000 words. At a thoughtful reading pace, that is weeks of work before any analysis begins.

AI coding tools apply codes across an entire dataset in minutes. The practical consequence is that researchers can analyse larger samples, iterate faster, and spend more of their time on interpretation rather than categorisation.

The important caveat is that AI-generated codes require researcher review. The value of AI in qualitative coding is not that it replaces human judgement — it is that it takes the mechanical labour out of the process so that human judgement can focus on what matters.

For a realistic assessment of what AI can and cannot do in qualitative research, see AI in qualitative research.

Product managers and HR and people teams often have large datasets from user research sessions or employee surveys where the volume makes manual coding impractical — AI coding is particularly useful in those contexts.

Frequently asked questions

How many codes is too many?

There is no fixed limit, but if you end up with more than one code per two or three passages of data, you may be coding at too granular a level. A common outcome of first-cycle coding is 50-150 codes on a moderate-sized dataset. If you have 400 codes on 30 interviews, your codes are probably closer to annotations than analytical categories. The second coding cycle is where you consolidate.

Can you code quantitative data?

Coding in the qualitative research sense applies to text. Quantitative data has its own analytical methods — statistical coding (assigning numeric values) is a different practice entirely. If you have a mixed-methods study with both open-ended text and numerical data, you would code the text qualitatively and analyse the numbers separately before integrating findings.

What software is best for qualitative coding?

The right tool depends on your context. Traditional QDA software like NVivo, MAXQDA, and ATLAS.ti offer comprehensive manual coding environments suited to academic research. AI-native tools like Skimle handle the initial coding automatically and let researchers refine the results. Spreadsheets work for very small datasets. For a full comparison, see qualitative data analysis tools: a complete comparison.

How do I check whether my coding is reliable?

If you are working alone, reliability comes from consistency (use a codebook, code a sample twice and compare) and documentation (explain your coding decisions in a memo or methods section). If you are working with multiple coders, calculate inter-coder agreement using Cohen's kappa or a percentage agreement metric on a shared subsample of the data. A kappa above 0.7 is generally considered acceptable.

Want to apply systematic coding to your qualitative data? Try Skimle for free — inductive and deductive coding across your entire dataset, with every finding traceable to source quotes.

Related reading:

About the authors

Henri Schildt is a Professor of Strategy at Aalto University School of Business and co-founder of Skimle. He has published over a dozen peer-reviewed articles using qualitative methods, including work in Academy of Management Journal, Organisation Science, and Strategic Management Journal. His research focuses on organisational strategy, innovation, and qualitative methodology. Google Scholar profile

Olli Salo is a former Partner at McKinsey & Company where he spent 18 years helping clients understand the markets and themselves, develop winning strategies and improve their operating models. He has done over 1000 client interviews and published over 10 articles on McKinsey.com and beyond. LinkedIn profile