Most consulting research guides focus on interviews. But a significant proportion of the evidence base in a consulting engagement comes from documents: earnings call transcripts, regulatory filings, industry reports, patent data, press coverage, court documents, expert papers, and internal client data. A large commercial due diligence might involve 200-300 documents alongside 40-50 interviews. A regulatory strategy engagement might involve 500+ pages of filings and submissions.
Analysing these documents systematically, rather than reading a selection and relying on memory and impressions, requires a different approach from interview analysis. This guide covers that approach.
Why document analysis is harder than it looks
Documents feel easier to analyse than interviews because they are already written down. But systematic analysis of a large document corpus is a distinct discipline, with its own failure modes.
Selection bias. When you have 200 documents and limited time, you read the ones that look most relevant. What looks most relevant is usually what confirms your existing hypothesis. The documents that challenge your view or surface unexpected findings often look, at first glance, less relevant. Systematic analysis of the full corpus is the only guard against this.
Memory-based synthesis. Even when analysts read a large corpus diligently, the synthesis stage often relies on what they remember most vividly rather than what the full set of documents actually shows. What is most vivid is not necessarily what is most common or most significant. A single striking example from one document can anchor the synthesis more strongly than 20 moderate examples from 20 documents.
Missing the language of the data. Documents from different sources (competitors, regulators, analysts, customers) use different language to describe the same phenomena. A systematic analysis that searches for specific terms will miss findings from documents that use different vocabulary for the same concept. Thematic coding, which groups by meaning rather than by word, is more reliable.
Losing context. A quote extracted from a document without its context can be misleading. Regulatory language, for example, often contains caveats and conditions that make the apparent finding much narrower than it first appears. Systematic analysis requires reading passages in context, not just extracting surface statements.
Types of qualitative documents consultants analyse
Understanding the document type shapes how you approach the analysis.
Earnings call transcripts and investor materials
Q&A sessions in earnings calls are among the most analytically valuable qualitative data sources in commercial research. Executive responses to analyst questions are:
- Less scripted than prepared statements
- Under more scrutiny, so more likely to be accurate about material facts
- Revealing about what executives consider to be the key drivers of performance
- A signal of where the market's attention is focused (what questions analysts are asking)
For competitive intelligence, systematic analysis of a competitor's earnings calls over 8-12 quarters reveals strategic narrative shifts, emerging problems, and performance drivers that are not visible in the quantitative financials alone. See earnings call transcript analysis for the full analytical approach.
Regulatory filings and submissions
Regulatory documents are underused as a data source in consulting research. They contain:
- Risk factor disclosures that are legally reviewed and therefore relatively accurate statements of what the company believes its real risks to be
- Capital allocation and investment rationale in annual reports and proxy statements
- Industry-specific technical information in sector-specific filings (FDA submissions, planning applications, utility filings)
- Lobbying positions and regulatory strategy in public consultation submissions
The challenge is that regulatory language is dense and requires domain expertise to interpret correctly. The analysis must distinguish between legally mandated boilerplate and material disclosures.
Industry and analyst reports
Published reports from industry associations (ESOMAR, Gartner, Forrester, sector trade bodies) and equity analysts contain:
- Market size and growth estimates (with varying reliability)
- Competitive landscape assessments
- Technology and trend analysis
- Industry participant views on the direction of the market
The key analytical task with these sources is provenance tracking: who produced the report, what was their data source, and what conflicts of interest might affect the analysis? Analyst reports on companies they have a banking relationship with, or association reports funded by industry members, require a higher discount rate.
Customer and expert documents
For customer-facing consulting work, qualitative documents from customers include:
- Customer support tickets and complaint records
- NPS verbatims and survey open-ended responses
- Social media posts and review platform content
- User forum discussions
These sources require different analytical approaches from formal documents. The language is informal, the credibility of individual statements varies, and the volume can be very large. AI-assisted analysis is particularly well suited here: the corpus is too large for manual reading but the patterns are accessible through systematic thematic analysis.
The systematic document analysis workflow
Phase 1: Corpus definition
Before reading any documents, define the corpus explicitly:
- What document types are included, and why?
- What is the time boundary (documents from what period)?
- What sources are included (which competitors, which regulatory bodies, which analyst firms)?
- What is the threshold for inclusion (minimum relevance criteria)?
Write this down. The corpus definition is part of the methodology and determines what claims your analysis can support. A finding based on documents from the past 12 months cannot support a conclusion about long-term trends. A finding based on public documents only cannot speak to what is in private communications or internal strategy documents.
Document the corpus in your final output: "Analysis based on [X] documents including [source types], covering the period [dates]."
Phase 2: Coding scheme development
Develop your coding scheme before reading the documents in depth. The scheme should map to your issue tree:
- Each branch of the issue tree becomes a code or code category
- Codes should be specific enough to be useful (not just "competitive dynamics" but "competitor pricing strategy," "competitor product development pace," "competitor channel strategy")
- Include a code for unexpected findings: material that does not fit the issue tree but appears significant
For a first-pass analysis of a large corpus, you can use a two-level scheme: broad categories (aligned to issue tree branches) and specific codes within each category. This gives you enough structure to aggregate findings without overspecifying in advance.
Phase 3: Systematic coverage
Every document in the corpus should be read and coded, not a sample. This is the principle of coverage that distinguishes systematic analysis from impressionistic reading.
For large corpora (100+ documents), full manual coverage is impractical within consulting timelines. AI-assisted analysis extends what is achievable: Skimle processes a corpus of PDF, Word, and text documents, applies the coding scheme, and identifies thematic patterns across the full set. The analyst then reviews the AI's output rather than reading each document individually.
The review stage is essential. For documents with significant analytical weight (a key competitor's most recent annual report, a critical regulatory submission), read the full document yourself and verify that the AI's coding captured what matters. Spot-check a sample of coded passages to confirm the coding is accurate.
Phase 4: Cross-source triangulation
The most valuable findings in document analysis are those that appear consistently across multiple source types. A finding that appears in:
- A competitor's public statements (management view)
- Analyst reports (outside observer view)
- Customer reviews (end-user view)
- Regulatory filings (legally attested view)
...is far more credible than a finding that appears in only one of these sources. Cross-source triangulation is the primary quality mechanism in document analysis, compensating for the bias and limitation of any single source.
Build triangulation into the synthesis: for each key finding, identify which source types it appears in and where the sources diverge.
Phase 5: Integration with interview data
In most consulting engagements, document analysis runs alongside an interview programme rather than separately. The integration of document and interview findings is where the most powerful insights emerge.
Document analysis tells you what is publicly stated and legally attested. Interview data tells you how insiders and informed observers actually interpret those statements and what is happening beneath the surface. The combination allows you to:
- Use document findings to frame interview questions ("You mention in your annual report that integration complexity is a top risk. How are you thinking about addressing it?")
- Use interview findings to interpret document patterns ("Several analysts noted a decline in gross margin, and our interviews with former employees suggest this is partly attributable to a shift in the sales mix rather than input cost pressure")
- Identify the gap between stated strategy (documents) and reported experience (interviews)
This gap analysis is often the most analytically productive dimension of research that combines both methods. For the full approach to combining document and interview data in a strategic research context, see evidence-based strategy: making sense of unstructured data.
AI-assisted document analysis: where it helps and where it does not
AI-assisted analysis of document corpora addresses the primary practical constraint: analyst time. For a corpus of 200 documents, full manual reading and coding requires 80-150 hours of analyst time. AI processing of the same corpus takes hours, with the analyst reviewing the output rather than reading every document.
What AI does well in document analysis:
- Coverage: every document is processed, not a sample
- Consistency: the same coding framework applied to document 200 as to document 1
- Pattern identification: themes that emerge across a large corpus are surfaced even when they would not be salient in any individual document
- Quote extraction: relevant passages are pulled with their context, ready for analyst review
What AI does less well:
- Interpreting dense technical or legal language that requires domain expertise
- Distinguishing between boilerplate and material disclosure in regulatory documents
- Understanding implied meaning or strategic subtext in executive communications
- Making the inferential leap from what documents say to what they imply about competitive position or strategic intent
The analyst's role in AI-assisted document analysis is to provide this interpretation layer: reviewing the AI's pattern identification, challenging its categorisation where the context requires it, and making the inferential connections that require domain knowledge and analytical judgement.
For the practical setup of an AI-assisted document analysis workflow in Skimle, the process is the same as for interview analysis: upload documents, set metadata (document type, source, date, subject), run the analysis, review the thematic output. For guidance, see how to find themes across a large set of interviews, which covers the same analytical steps applied to a document corpus.
Synthesising findings from large qualitative corpora
The synthesis of a large document corpus follows the same structure as interview synthesis: issue tree first, then evidence, then the gaps.
The specific challenge in document synthesis is maintaining the connection between the finding and the evidence across a large corpus. When your evidence base is 200 documents rather than 30 interviews, the risk of orphaned assertions (findings that cannot be reconnected to specific supporting text when a client pushes back) is higher.
Practical disciplines that prevent this:
Cite documents in the synthesis draft. As you write each finding, note the source documents that support it. In the final client output you will not show every citation, but having them in your working draft means you can find the evidence quickly if challenged.
Keep verbatim extracts. For your 5-6 most important findings, have 3-5 verbatim extracts ready to share. These are the evidence you will produce if a client says "where did you get that?"
Note the document distribution. For a finding that appears across 50 documents, you do not need to cite all 50. But stating "this theme appeared in 14 of 18 earnings call transcripts reviewed, across competitors A, B, and C" is far more defensible than "multiple sources suggest."
Frequently asked questions
How do you handle documents in different languages?
For multilingual document corpora, the options are: translate before analysis, or analyse natively. Translation introduces distortion (especially for regulatory and technical language) and adds cost. AI-assisted analysis with multilingual capability avoids this: Skimle processes documents in over 100 languages and analyses them together, identifying patterns across the multilingual corpus without requiring translation. This is particularly relevant for European market research, where source documents may be in German, French, Dutch, and English within the same study.
Can document analysis replace expert interviews?
Rarely, and not as a general rule. Documents tell you what was publicly stated and, in the case of legal or regulatory filings, what the author was prepared to attest to. Expert interviews tell you the interpretation, the context, and the things that are known but not written. The combination is standard in consulting research: documents provide coverage and an audit trail; interviews provide depth and interpretive context. For a research programme that must choose one for budget reasons, the choice depends on the question. Strategic intent and competitive positioning are better served by interviews; compliance, regulatory posture, and historical performance are better served by documents.
What is the right sample size for document analysis?
Unlike interview research, document analysis does not have a "saturation" concept in the same way. The relevant question is: have you covered the sources that are most likely to contain the evidence relevant to your question? For a competitor analysis covering 3-4 competitors over 3-5 years, this might mean 50-80 documents. For an industry trend analysis covering 10-15 firms over 5-10 years, it might mean 300-500. Define the corpus based on the scope of the question, not based on convenience.
How do you cite document analysis findings in a client presentation?
In a slide deck, in-text citation is typically replaced by a footnote or an appendix. "Expert interviews and document analysis (see appendix)" is the standard. For findings that will face the most scrutiny, prepare a one-page evidence summary for the relevant appendix section: the key finding, the supporting quotes from documents, the source types represented, and the date range of the evidence.
Facing a large document corpus and need to process it systematically within your project timeline? Try Skimle for free and see how AI-assisted analysis handles your document corpus, from upload to thematic output with full traceability.
Related reading: Evidence-based strategy: making sense of unstructured data | Consultant's guide to summarising interviews and finding themes | Voice-of-market research: building a complete picture
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
