The best tool for teaching qualitative analysis depends on what you're actually teaching

The best tool for teaching qualitative analysis depends on a question most software comparisons skip: are you teaching students to code, or are you teaching them to do qualitative analysis? Coding tools (Delve, MAXQDA, Quirkos, Taguette) get students labelling text fast, but most of the course stays in the coding mechanics, with theory-building squeezed into the final weeks. Skimle removes that bottleneck so a class can spend its time on judgement, reflexivity and theory from week one.

These are not the same question, and conflating them is the reason a lot of qualitative methods courses leave students able to apply a code to a sentence but unable to say what that exercise was actually for.

What are you trying to teach: coding, or qualitative analysis?

Many comparisons of qualitative software for the classroom ask a sensible-sounding question: how fast can students start coding? They measure time to first coding session, whether instructors can review work without wrestling with file formats, and what a student licence costs. By that measure, web-based tools like Delve beat NVivo easily, and a free tool like Taguette beats everything on cost.

That is the right question if coding the text by hand is the lesson. Teaching students to sit with a transcript, assign a label, defend why this passage got that code and not another, and build a codebook from scratch is a real, valuable, teachable skill. It is also, on its own, not the same thing as teaching qualitative analysis.

The challenges of teaching qualitative coding are well documented in the methods education literature: students struggle to grasp coding as anything more than mechanical labelling unless an instructor is actively pushing them toward interpretation. The coding itself, in other words, is the easy part to assess and the hard part to make meaningful. Optimising a course around "time to first coding session" answers a real operational problem (classroom friction) without addressing the harder pedagogical one (do students ever get past coding to the analysis it was supposed to enable).

Manual coding, even in the friendliest tool, is slow. A class of 20 students working in pairs can realistically hand-code a handful of interview transcripts each across a semester, not the 30 to 50 a thematic saturation argument would actually call for. So the project's sample size (its n) ends up small by necessity, not by methodological choice. More importantly, the class's calendar gets consumed by the coding itself. Weeks one through eight: read transcripts, assign codes, refine the codebook, check that two coders applied it consistently. Week nine, if there is time: compare cases, look for patterns across the dataset, attempt to say something about what it all means. For some courses, that last and most important step gets compressed into a single session or assigned as homework nobody has bandwidth left to do well.

This is not a knock on Delve, MAXQDA, Quirkos, or Taguette. Each does the coding job it is built for competently, at noticeably different price points. The issue is that "fastest to first code" was never the bottleneck that matters most for learning outcomes; it is the bottleneck that is easiest to measure.

What changes if the learning objective is judgement, not mechanics?

If the goal is closer to "students can do qualitative analysis," meaning for example understand the nuances in the data, develop a coding frame, identify or create themes, debate alternative interpretations, reckon with their own position in relation to the data, and build toward an explanation or model, the bottleneck is different, and so is the tool that helps.

Some might argue that the higher order qualitative skills can only be introduced after the student spends significant amounts of time on the basic act of coding by hand. This is akin to monasteries asking novices to do mundane tasks for years before starting to discuss deeper questions, or students of classic Japanese archery spending the first year without even touching the bow. We would argue that while mastery in manual coding is helpful, the other aspects of qualitative research are also worth teaching from the start. Getting to the deeper skills and useful results faster will increase the appreciation towards and perceived value of qualitative research.

Skimle processes every document during upload and produces a full category structure with insights nested inside, following a similar systematic thematic analysis process a trained coder would do, at AI speed. Critically, this is not a black box the class has to take on faith. Every code Skimle assigns is visible, traces back to the exact source passage, and is fully editable. Students are not handed a finished answer; they are handed a transparent coding scheme they can interrogate, challenge, and rework.

That changes what a session can be about:

Alternative framings, tested in real time. Because the mechanical coding pass takes minutes rather than weeks, a class can run the same dataset through two different framings, an inductive open-coding approach and a predefined categories approach matched to an existing theory, and compare what each one surfaces. Trying this by hand even once per semester is a stretch; trying it twice to make a teaching point about how framing shapes findings is normally out of reach.

Reflexivity and positionality, with something concrete to argue about. A common complaint about teaching reflexivity is that it stays abstract: students are told to reflect on their position but have nothing specific to anchor the reflection to. An AI-coded dataset gives the class a live object to interrogate together: why did the model code this excerpt this way? What did it miss that a researcher attentive to a particular lived experience would have caught? Whose framing does the default coding centre, and whose does it flatten? Editing the AI's coding, and debating why an edit is warranted, is itself a reflexivity exercise, arguably a sharper one than asking students to reflect on their own coding choices, which they are naturally inclined to defend rather than interrogate.

Pattern discovery and model-building from week one, not week nine. With the coding pass handled, a class project covering 40 or 60 documents (a number that supports an actual saturation argument) is feasible within a semester. Students spend their time on cross-case comparison, abductive reasoning about what explains the pattern, and producing an actual model or framework, the part of qualitative analysis that the coding was always supposed to be in service of.

Preparation for where the field is actually going. AI-assisted qualitative analysis is no longer a niche choice in research practice; the methodological question students will face in their own future research is not "should I ever touch AI" but "how do I supervise, audit, and critique AI-generated coding before I trust it." That is a skill in its own right, and a class that only ever hand-codes never gets to practise it.

This is not an argument against teaching manual coding

It is worth being direct about this, because the framing above can read as more combative than it is meant to be. Hand-coding remains a legitimate, valuable thing to teach. Some programmes need students fluent in NVivo or MAXQDA because that is what their discipline or their supervisors expect. Some early methods courses are specifically about the discipline of close reading, and the slowness of manual coding is part of the point, not a bug to engineer away. None of that is wrong, and a comparison that pretends otherwise is doing a disservice to instructors who have good reasons for their current syllabus.

The possible move for many programmes is sequencing rather than either/or: an early module that teaches manual coding fundamentals in Quirkos, Delve, or Taguette, followed by a module that uses Skimle to reach theory-building at a scale a semester can actually support. Because Skimle supports REFI-QDA export, students who later need to continue in NVivo or MAXQDA, for a thesis, a dissertation committee, or a job that requires it, can move their coded project across rather than starting over. The two approaches are not in competition; they teach different things, and a well-designed course can use both deliberately.

How to decide for your course

Frequently asked questions

Will students still learn to code if the class uses Skimle?

They will see coding decisions made, made visible, and they will edit and challenge them, which is a different but related skill to producing the first code from scratch. For a course where producing the first code by hand is itself the objective, pair Skimle with a short manual-coding module rather than replacing it entirely.

Is this appropriate for a PhD-level methods course?

It depends on what the course is for. If it trains students in a specific manual tradition their discipline expects (grounded theory's constant comparison, for instance, taught with the classic process in full), a manual tool may better serve the requirement. If the course is preparing students to run their own large-scale qualitative projects, where realistic sample sizes and AI literacy both matter, Skimle is the stronger fit. Our guide for PhD students covers the budget and licensing trade-offs in more detail.

Does using an AI coding tool undermine training in reflexivity?

Not if the AI's coding is visible and editable rather than a black box. A coding decision a student can see, question, and overturn is a better object for a reflexivity discussion than one they cannot inspect at all. The concern would be valid for any tool that hides its reasoning; it is specifically why two-way transparency, tracing every theme back to its source text and back again, matters more in a teaching context than almost anywhere else.

What does Skimle cost for a class project?

Skimle's free tier covers up to 400 pages of analysis, which might be enough for some single-semester class projects. We are flexible to discuss licences for courses depending on your needs.

Can a course use more than one tool across a semester?

Yes, and for many programmes this is the best answer rather than a compromise. A foundations module in a manual coding tool, followed by an applied project in Skimle, teaches both the mechanics and the analysis they are meant to support, without asking either tool to do a job it was not built for.

Want to see how the analysis works before you build a syllabus around it? Try Skimle for free on a sample dataset and see what a class project can look like once the coding pass is no longer the bottleneck. If you teach or supervise in an academic department, it is worth trialling on last year's course data before committing a new syllabus to it.

Planning a methods course? Read more on how to code qualitative data, grounded theory's full coding process, and using AI in academic qualitative research.

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 co-founder at Skimle and 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

Sources

• The Best QDA Software for Teaching Qualitative Research - Delve
• Raddon, Raby & Sharpe (2009), "The Challenges of Teaching Qualitative Coding: Can a Learning Object Help?" - International Journal of Teaching and Learning in Higher Education, 21(3)
• Delve pricing
• Quirkos licences
• Taguette, free and open-source qualitative data analysis