Three Paths to Qualitative Analysis
You have interview transcripts, open-ended survey responses, video interviews, contracts, statements or other documents to analyse. Which tool should you use?
After years of working with qualitative data - both at McKinsey where Olli did over 1000 business interviews, and in academia where Henri has published a dozen studies with qualitative data - we have come to see that the tool question is actually a methodology question. Before comparing features and pricing, you need to decide which approach fits your needs.
There are three distinct paths:
Path 1: Rigorous Manual Analysis - Traditional academic tools require you to systematically code every passage by hand. Transparent, reproducible, but time-intensive. Think NVivo, MAXQDA, ATLAS.ti.
Path 2: Ad-hoc + AI Assistance - This category includes generic LLM tools such as ChatGPT, Microsoft co-pilot, and NotebookLM, as well as coding tools with bolted-on AI features of NVivo, Atlas.ti, and MAXQDA.. These lack the structured rigour of traditional approaches, but lack the coherence and comprehensiveness needed in qualitative research.
Path 3: AI-Native Structured Analysis - Purpose-built for AI from the ground up, but maintaining the transparency and structure of academic methods. This is the approach we took with Skimle.
Each path involves trade-offs. Let me walk through what each offers.
Path 1: Rigorous Manual Analysis
These are the established tools that have defined qualitative research for decades. They’re built around the traditional workflow: read your data, develop codes, manually tag passages, and build themes.
NVivo
NVivo is the most cited qualitative analysis software in academic publications, now owned by Lumivero. It’s designed for medium to large enterprises, academic institutions, and research organisations.
What it does well:
- Comprehensive coding system with hierarchical nodes
- Excellent visualisation tools (word clouds, charts, cluster analysis)
- Strong mixed methods support (qualitative + quantitative)
- Collaboration features via NVivo Collaboration Cloud
- Wide format support including video and audio
AI features (NVivo 15): NVivo has added AI through the Lumivero AI Assistant, including text summarisation, coding suggestions, and sentiment analysis. The AI can suggest child codes within existing categories and auto-code for themes.
Limitations:
- Steep learning curve, especially for advanced features
- Interface feels dated to some users
- All coding is still fundamentally manual - AI assists but doesn’t transform the workflow
Pricing: From €1,100/year for commercial licenses. Student licenses around €90/year. Worth noting: in September 2024, Lumivero acquired ATLAS.ti, bringing both products under one umbrella.
MAXQDA
MAXQDA is developed in Germany and positions itself as the most user-friendly of the traditional tools. It’s the only leading QDA software offering identical features on Windows and Mac.
What it does well:
- Cleaner, more modern interface than NVivo
- Strong mixed methods tools including statistical analysis
- Good collaborative features
- Visual tools for exploring code relationships
AI features (MAXQDA AI Assist): MAXQDA’s AI assistant helps with document summarisation, coding suggestions, and paraphrasing. It can suggest codes for text segments. Available in the Analytics Pro tier.
Limitations:
- The learning curve, while better than NVivo, is still significant
- AI features require the most expensive tier and have limited value in systematic coding
- Core workflow remains manual
Pricing: From €850 for Base edition to €1,600 for Analytics Pro (commercial). Academic pricing starts around €230/year.
ATLAS.ti
ATLAS.ti has invested heavily in AI, particularly through their AI Lab. They’ve been more aggressive than NVivo or MAXQDA in integrating LLM capabilities.
What it does well:
- Intuitive coding interface
- Strong network visualisation for exploring relationships
- Most advanced AI integration among traditional tools
- Web version available for browser-based access
AI features (ATLAS.ti AI Lab): Unfortunately the implementation of AI is clumsy: you get a huge list of disconnected first order codes that you need to manually weed through. After initially working only in the U.S. only, Atlas.ti has now added data residency options (US or Europe) for GDPR compliance.
Limitations:
- Despite AI features, the fundamental approach is still “AI-assisted manual” rather than “AI-native”
- AI does not provide systematic coding and categorization of your data.
Pricing: Similar range to NVivo and MAXQDA. Student licenses around €90/year.
Dedoose
Dedoose is a cloud-based tool developed by UCLA researchers. It’s positioned as an affordable alternative for distributed teams.
What it does well:
- Fully cloud-based - works on Mac, Windows, Linux, Chromebook
- Genuinely affordable (€15/month individual, less for students)
- Real-time collaboration with team coding
- Good security practices
Limitations:
- No significant AI features - all coding is manual
- PDF coding can be clunky
- Learning curve is still present
- For long projects, monthly costs add up
Pricing: €12-18/month depending on user type. Group discounts available.
Best for: Budget-conscious research teams who need collaboration and don’t mind manual coding.
When to Choose Path 1
Choose rigorous manual tools if:
- You are limited to classical methods (e.g., by institution)
- Your institution provides licenses (if not, check out free alternatives like Taguette or use Dedoose)
- You have months (not days) for analysis
- You need detailed audit trails for every coding decision
The honest reality: these tools offer tremendous power and transparency, but they demand significant time investment. A typical thesis project using NVivo might spend 3-6 months on analysis alone, and the first summer job of many researchers I know has been to manually code data page after page after page using these apps...
Because of this heavy bottom-up component, most users are academic. In the business world the time and cost considerations simply make these tools irrelevant.
Path 2: Ad-hoc + AI Assistance
This category includes both the AI-native generic tools that are not designed for qualitative research, such as NotebookLM, as well as the bolted-on AI features of qualitative analysis software, such as NVivo, Atlas.ti, and MAXQDA. Often, the combination of ill-thought AI features and manual coding features means that the user faces a stark time/quality trade-off.
ChatGPT, Gemini, Claude or other LLM models directly
I am adding this to the list as many people are trying to upload files directly (or via some RAG-based database) to LLM's with the instruction to analyse them. This is also the approach used by many workflows which contains an "Analyse with AI" or "Talk with the documents" type of feature as one step as it is simple to implement.
What they do well:
- Universally and instantly accessible via website
- User interface familiar to most users
- Fast analysis
Limitations:
- Lacks transparency: no coding of the files, no tracing of categories to raw text
- Superficial analysis, often biased towards themes from specific texts in the beginning or end of the models context window
- Depending on model and tier, maximum 3 to 20 files can be analysed at any given time. Zipping, merging files etc. can help bypass the limitation.
- Output is not stable: further questions, adding documents or asking probing questions changes the answer each time
Pricing: Starts free, but privacy (GDPR compliant data storage and no training use) require paid tiers.
Best for: When no real need to analyse the documents, and a simple summary with some themes is enough.
AI features of NVivo, Atlas.ti, and MAXQDA
These tools have added AI features to their existing workflows. Unfortunately, the only real value is summarization. Thus far, the coding features are clumsy and lack the structured rigour of traditional approaches; the AI suggests diverse codes for almost every sentence in the document. Coding a large dataset takes surprisingly long time, as there is no automated way to reanalyze all of your documents.
What they do well:
- All tools have "chat with documents" features that allow you to ask questions about the documents and get answers.
- Some tools, such as MAXQDA, have fairly good closed coding feature, allowing the user to define specific codes that AI looks for.
Limitations:
- Open coding features are clumsy.
- Chat with documents are based on basic RAG-model, searching through the documents for relevant information; they lack comprehensiveness.
- No support for document metadata, and poor integration of chat interface with the rest of the tool.
Dovetail
Dovetail is an Australian company that’s positioned itself as the “AI-native customer intelligence platform”. It’s popular in UX research and product teams at companies like Meta, AWS, and Dyson.
What it does well:
- Modern, clean interface designed for non-academics
- Automatic transcription and meeting import
- Good integrations with tools like Zoom, Slack
- AI-powered tagging and sentiment analysis
AI features: Dovetail offers AI-powered highlights, automatic theme detection, and sentiment analysis. Their “Channels” feature uses ML to continuously classify themes in large datasets like support tickets.
Limitations:
- User reviews are mixed on AI quality: “AI features feel tacked on”, “making us only marginally more efficient”
- Per-user pricing makes team collaboration expensive
- Tailored to UX research use case only
- Some researchers find it lacks the rigour needed for academic works
- Pricing has increased significantly with AI features
Pricing: Starts free, but paid tiers required for most features. Enterprise pricing not publicly disclosed but reported as expensive for teams.
Best for: Product and UX teams who need quick insights from customer feedback and already use modern SaaS tools.
AILYZE
AILYZE is a newer entrant (founded 2023) that uses LLMs to automate qualitative analysis. The founder has a background at MIT and Wharton.
What it does well:
- Fully automated thematic analysis
- Works in any language
- Very fast initial results
- Includes an AI interviewer feature for autonomous data collection
Limitations:
- Limited document counts on lower tiers (3 docs on free, 15 on Pro)
- Black-box analysis - less transparency in how themes are derived
- Limited user control over the analysis process
Pricing: Free tier for up to 3 documents. Pro at €45/month for up to 50 documents.
Best for: Researchers who want speed over methodological rigour, or for initial exploration before deeper analysis.
The Problem with Path 2
Tool taking this path (like Dovetail and AILYZE) add AI, but it’s often bolted onto an ad-hoc workflow. You upload documents, the AI suggests themes or summaries, but the underlying approach is “ask the AI a question and see what comes back”. This is essentially the RAG (Retrieval-Augmented Generation) approach that works nicely in demos but falls short in serious analysis.
The result is fast but not rigorous. You get summaries, but you can’t easily trace back to see if the AI considered all relevant passages. You lose the transparency that makes qualitative research credible. Analyses using these tools are simply not passable to any academic publication, but also the quality can be too low for serious business or journalistic analysis.
Path 3: AI-Native Structured Analysis
This is the approach we took with Skimle. Rather than adding AI to a traditional workflow, we asked: what would qualitative analysis look like if designed for AI from the start - while maintaining the transparency and rigour of academic methods?
Skimle
Skimle is an AI-native qualitative analysis tool combining academic rigour with the speed of AI. It can be used by researchers, analysts, legal professionals, consultants and anyone else wanting to make sense of large sets of qualitative data.
How it works: Rather than using RAG to retrieve relevant chunks at query time, Skimle processes each document systematically during upload using hundreds of atomic LLM calls. This mirrors what a human expert does - reading each section, understanding what it says, and assigning it to categories - but at AI speed.
The result is a structured dataset you can explore, edit, and query. Every insight links back to specific quotes. You can merge categories, split them, add your own, and see exactly which passages support each theme.
What it does well:
- Fast: what takes weeks manually takes minutes
- Two-way transparency: every categorisation traces to source quotes, and source quotes are traceable to categories
- Editable: you control the categories, not just the AI
- Rigorous: maintains the structure of academic thematic analysis
- Works in any language and domain
Limitations:
- Newer product with smaller user base than established tools
- May be overkill for very small projects (3-5 interviews)
- Not designed for mixed methods (qualitative + quantitative integration)
Pricing: Free tier available for up to 600 pages of analysis. Paid plans for larger projects.
Best for: Consultants, researchers, and analysts who need speed without sacrificing the ability to defend their findings. Particularly strong for interview analysis, policy consultation responses, and document-heavy due diligence.
Choosing Your Approach
Choose Path 1 (NVivo/MAXQDA/ATLAS.ti etc.) if:
- Your institution provides licenses
- You have months for analysis
- You need mixed methods (qual + quant)
- Methodological purity and sticking to older workflows is non-negotiable
- Best fit for participatory research where researcher holds knowledge beyond the textual data
Choose Path 2 (AI-native generic tools or bolted-on AI features) if:
- Quick insights matter more than rigorous methodology
- You’re working with a tool that already has the AI analysis inbuilt and don't need deeper insights
- You don’t need to defend methodology in detail
- You are happy with "just chatting" with your documents, no need for comperehensive analysis
Choose Path 3 (Skimle) if:
- You need speed AND rigour
- You want two way transparency, being able to go from categorties to original data and see how each document has been coded.
- You want to understand your dataset and develop expertise, not just outsource one-off analysis to an AI
- You’re a consultant, analyst, or researcher with deadlines and a high quality bar
- You want AI to handle the mechanical work while you focus on developing insights
The tools you choose should match your actual needs, not just feature lists. A PhD student spending three years on a dissertation has different needs than a consultant with a two-week deadline. Both are valid - they just require different approaches.
Frequently Asked Questions
Which tool is best for beginners? For learning qualitative methods, MAXQDA has the gentlest learning curve among traditional tools. For getting quick results without methodology training, Skimle or AILYZE are more accessible.
Can I use AI tools for academic research? Yes, but with documentation. Be transparent about your methodology. If using AI-assisted coding, explain how you validated the results. Also check publication and your institutions guidelines. Tools like Skimle that maintain full traceability make it more defenable to use AI than black-box LLM calls.
What about data privacy? Check each tool’s data handling policies. ATLAS.ti offers data residency options (US/EU). Cloud tools like Dedoose and Dovetail process data on their servers. Skimle stores data in EU based clouds and also has insitutional hosting options available. For sensitive data, desktop-only tools may be preferable.
How long does analysis actually take? With manual tools: months for a substantial project. With AI-native tools like Skimle: hours to days. But remember that analysis is thinking, not just coding - AI speeds up the mechanical work, not the interpretation and theory building.
If you want to try an AI-native approach that maintains academic rigour, you can try Skimle for free. See how structured AI analysis compares to both manual coding and ad-hoc AI tools.
Olli and Henri from the Skimle team