Coding and Thematic Analysis.

Teaching Note

Author

Larry Vincent

Published

February 16, 2026

You just finished your first customer interview. Forty-three minutes of conversation about how someone chooses a streaming service. You asked good questions. You listened. You resisted the urge to lead. Now you’re staring at a twelve-page transcript and thinking: What do I do with this?

The honest answer is that most people do one of two things, and both are wrong. Some skim the transcript, cherry-pick a few quotes that support what they already believed, and call it analysis. Others highlight everything that seems interesting, produce a document full of yellow streaks, and still have no idea what to do next.

Qualitative coding is the discipline that sits between raw conversation and real insight. It’s how you move from “a person said some things” to “here’s what we learned.” And while the word “coding” might sound like it belongs in a computer science class, it’s actually closer to detective work—tagging evidence, sorting clues, and building a case.

What Coding Actually Is

A code is a short label you attach to a segment of text to describe what’s happening in that segment. It could be a word, a phrase, or a short sentence. You are probably familiar with this concept because it is very similar to tagging in the digital world. Whether it is your playlist or a transcript, the point is to encapsulate the meaning of the material to you, personally.

Here’s a passage from a hypothetical interview about choosing a coffee shop:

“I don’t really care about the coffee itself, honestly. I go there because it’s quiet, and nobody bothers me. I bring my laptop, put in my earbuds, and I can get three hours of work done. It’s like my second office.”

You might code this passage several ways: Third place, Productivity, Atmosphere over product, Routine behavior. Each code captures a different dimension of what the respondent is telling you. You realize that this passage isn’t about coffee at all. It’s about finding a functional workspace, about ritual, about a place that plays a specific role in someone’s life. And, of course, a single passage could have many different meanings that you wish to code.

That’s the first lesson of qualitative coding–the most important meaning in an interview rarely sits on the surface, and it may be multifaceted.

Two Directions: Inductive and Deductive

There are two fundamental approaches to coding, and you’ll encounter both in your work.

Deductive coding starts with a predefined set of codes. Before you read a single transcript, you’ve built a codebook based on theory, prior research, or your research questions. You’re looking for specific things. If you’re studying customer loyalty, you might start with codes like Switching costs, Emotional attachment, Habit, and Price sensitivity. As you read, you tag passages that fit those categories.

Inductive coding starts with the data and lets codes emerge from what respondents actually say. You read each transcript without a predetermined framework, creating new codes as you encounter ideas. You don’t know what you’re going to find. That’s actually the point.

In practice, most coding falls somewhere in between. You might begin with a loose framework from your research questions (deductive) but remain open to patterns you didn’t anticipate (inductive). Grant McCracken, whose work on the “long interview” shaped much of modern qualitative practice in marketing, argued that the researcher should come prepared with a sense of what the literature suggests but also be prepared to ignore all of it when the data points somewhere unexpected.

	Deductive	Inductive
Starting point	Theory, literature, research questions	The data itself
Codebook	Built before analysis	Emerges during analysis
Best for	Testing specific hypotheses; structured studies	Exploratory research; discovering new patterns
Risk	Missing what you didn't expect to find	Drowning in codes without structure

The Coding Process: From Raw Text to Themes

Whether you lean deductive or inductive, the coding process follows a predictable arc. Think of it as three stages of increasing abstraction—you start close to the words and end with interpretive claims about what those words mean.

Stage 1: Open Coding. Read through each transcript line by line and tag meaningful segments. Don’t be precious about it. Code generously. If a passage seems interesting, relevant, or surprising, give it a label. You’ll refine later. At this stage, it’s normal to generate dozens of codes from a single interview. Some will be descriptive (“mentions price”), some will be more interpretive (“expresses guilt about switching”).

A few practical tips for this stage: Code actions, not just opinions. When someone says “I checked three different apps before I booked,” that’s a behavior worth coding (Comparison shopping, Decision effort). Also, pay attention to emotional language, metaphors, and contradictions. When a respondent says “I love that brand” but then describes never actually buying it, the contradiction is the interesting part.

Stage 2: Axial Coding. Once you’ve open-coded your transcripts, step back and look at the full inventory of codes. You’ll notice overlap, redundancy, and natural groupings. Axial coding is the process of organizing your codes into categories—clusters of related codes that point toward a larger idea.

For example, imagine you coded five interviews about streaming service choice and ended up with codes like Binge-watching habit, Weekend ritual, Background noise while cooking, Shared watching with partner, and Solo guilty pleasure. These are all different—but they all describe something about how streaming fits into daily life. You might group them under a category you call Consumption Contexts.

Stage 3: Thematic Analysis. Themes are not just categories with a nicer name. A theme is an interpretive claim—it tells you what a pattern in the data means for your research question. A category like “Consumption Contexts” becomes a theme when you can articulate its significance: Streaming services are not chosen for their content libraries alone; they’re chosen for how well they integrate into the rhythms of daily life.

That’s the kind of insight that changes a marketing strategy. And it only emerges through the patient, iterative work of coding.

Beyond Coding: Narrative Inquiry

Coding isn’t the only way to analyze an interview. There’s an older tradition—narrative inquiry—that takes a fundamentally different approach. Instead of breaking a transcript into fragments and sorting those fragments into categories, narrative inquiry treats the interview as a story and analyzes it as a whole.

The distinction matters. When you code, you’re looking for patterns across respondents: what themes appear in interview after interview? Narrative inquiry asks a different question: what is the structure of this person’s experience? How do they construct the story of their relationship with a brand, a product, or a decision? What’s the plot? Who are the characters? Where are the turning points?

Consider a customer who describes switching from one bank to another. A coding approach might tag passages as Service failure, Trust erosion, Competitor awareness, and Switching trigger. Useful. But a narrative approach would examine the story arc.

This customer narrates a journey from loyalty to betrayal to liberation. The switching event is the climax of a story the customer has been telling themselves about who they are and what they deserve.

In marketing, narrative inquiry has proven particularly powerful for understanding how consumers construct brand meaning over time, how they make sense of major purchase decisions, and how identity shapes consumption. Brands, after all, live inside the stories people tell about their own lives.

You won’t be conducting full narrative analysis in this course, but the sensibility is worth carrying with you. When you’re reading a transcript, ask yourself: is this person telling me a story? If so, what kind of story is it? What role does the brand play in the narrative? Sometimes the most revealing insight isn’t in any single passage—it’s in the arc of the conversation itself. And remember, most of us are the heroes of our narratives. This context allows you to build empathy for the respondent.

The Codebook

Whether you start deductive or end up building one inductively, a codebook is your essential reference document. It defines each code so that you—and anyone else on your team—apply them consistently.

A good codebook entry includes three things:

Element	Example
Code name	Third Place
Definition	Respondent describes using a commercial space (café, library, co-working) as a regular work or social environment distinct from home or office
Example quote	"It's like my second office—I go there every morning."

If you’re working with a partner, a shared codebook is what keeps your coding aligned. When two researchers independently code the same transcript and agree on most of the codes, you’ve achieved what’s called inter-coder reliability—one of the key markers of rigor in qualitative research.

A Note on AI

You will increasingly encounter AI tools that promise to code qualitative data for you. Upload your transcripts, press a button, get a thematic framework. Some of these tools—NVivo, Atlas.ti, Dedoose, and newer platforms like Delve—now include AI-powered features that can generate initial codes, suggest groupings, and summarize patterns across interviews.

Here’s what AI does reasonably well: it can speed up the mechanical work of initial tagging, especially with deductive coding where you’ve defined the categories in advance. If you have a codebook and twenty transcripts, AI can apply your codes faster than you can. Recent research suggests AI can be effective for first-pass open coding and descriptive labeling, potentially saving substantial time in the early, most mechanical stages of the process.

But AI often struggles with the interpretive work that makes qualitative research valuable. AI tends to produce surface-level, descriptive codes—it can tell you that five respondents mentioned price, but it’s unlikely to notice that they mentioned price only when they felt defensive about a purchase. Emotional subtext, contradictions, metaphors, shifts in tone, narrative arcs—these are the bread and butter of qualitative insight, and they require a human reader who understands context. As one researcher put it, AI can tag what people said, but it can’t tell you what they meant.

There are also practical risks. If your interviews involve confidential information—which they almost always do in a business context—uploading raw transcripts to a cloud-based AI tool raises serious privacy and ethical concerns. Before using any AI platform for coding, ask: Where does this data go? Who can access it? Is it used to train the model? If you can’t answer those questions clearly, don’t upload.

AI is certainly a promising new tool in the qualitative analysts toolbox, but treat it truly as an assistant, not as a substitute for your analytical mind. It can help you organize. It cannot help you understand.

Putting It Into Practice

For your individual assignment and for your qualitative charrette, here’s a simple workflow:

Transcribe your interview as soon as possible after conducting it, while the conversation is still fresh in your memory.
Read the full transcript once without coding—just to re-immerse yourself in the conversation.
Open code on the second read. Use the margins, a spreadsheet, or qualitative software. Tag generously.
Build or refine your codebook after coding your first one or two transcripts.
Axial code once you’ve finished all transcripts. Print out your codes, spread them on a table (physical or digital), and look for groupings.
Develop themes by asking: What do these categories tell me about my research question? What’s the insight?
Select illustrative quotes for each theme. These are the evidence you’ll present in your deliverable.

One final piece of advice: don’t skip step two. Reading the full transcript before you start coding is the qualitative equivalent of looking at your survey data before calculating a mean. It builds familiarity. It surfaces surprises. And it prevents you from coding mechanically without ever really hearing what your respondent said.

AI Exploration Prompts

Try these prompts to deepen your understanding of qualitative coding:

“I have a transcript from a customer interview about [topic]. Help me develop a preliminary codebook with 8-10 deductive codes based on [theoretical framework]. For each code, provide a definition and an example of what a matching passage might look like.”
“Here are five codes I’ve developed from my interview data: [list codes]. Help me think about how these might group into higher-order categories. What relationships do you see?”
“I’m struggling to move from codes to themes. My research question is [question] and my main categories are [list]. Push me to articulate what these categories mean for my research question—not just what they describe, but what they imply.”

Note: These prompts are designed to help you think through the coding process, not to outsource it. The learning happens when you do the interpretive work yourself.