From Raw Verbatims to a Report: The Full Text Analytics Workflow¶
It's easy to think of Text Analytics as just "generate categories, then classify" - and that's the core of it - but the useful version of the workflow includes a few steps on either side that make the difference between a rushed first pass and results you'd actually stand behind in a report.
1. Collect Enough Responses to See Real Patterns¶
Category generation works on whatever responses exist at the time you run it - see How Many Responses Do You Need Before Category Generation Is Worth Running? for how to judge timing. Generating too early on a thin response set risks missing themes that later respondents would have surfaced.
2. Generate Categories¶
Point Text Analytics at your open-ended question and let it draft a category set from the responses collected so far. This is a starting point, not a finished product - see How Text Analytics Works: From Open-Ends to Categories.
3. Review and Edit the Category Set¶
Read through the drafted categories the way you'd review a colleague's first pass - rename, merge, split, add, or delete as needed. See Editing and Refining Categories in Text Analytics.
4. Classify¶
Run classification once you're satisfied with your categories - see Editing and Refining Categories in Text Analytics for why sequencing this after your category edits matters, since a later re-run resets any manual corrections made before it.
5. Audit and Correct¶
Don't treat the first classification pass as final. Use confidence scores to focus your review (see Understanding AI Confidence Scores in Text Analytics), correct anything that's clearly wrong (see Reviewing and Correcting Individual AI Classifications), and for anything going into a client-facing deliverable, consider a more formal audit pass (see Auditing AI Classification Accuracy on Your Own Data).
6. Build the Report¶
Once classification is finalized, bring the results into Reporting - a standalone category breakdown (see Choosing the Right Widget for Your Report), or cross-tabbed against a structured question to see who raised which themes (see Combining Text Analytics Results with Structured Questions in One Report).
7. Export or Share¶
Finish by exporting the underlying classified data if you need it outside Opionate (see Exporting Classified Responses with Confidence Scores), or export the report itself to a presentation-ready deck (see Exporting Your Report to PowerPoint), or share a live report link (see Sharing a Report Publicly).
Where Time Is Actually Saved¶
The steps that used to take the most unbillable hours - reading every response, drafting a codebook from scratch, and manually tagging each one - are the ones Text Analytics compresses into a first-draft pass. Your time goes into review and judgment calls instead of transcription-style coding work.
FAQ¶
Do I have to do every step in this order?
Roughly yes for steps 2 through 4 - editing categories before classifying, and classifying before your final manual review, is the sequence that avoids resetting work you've already done. Steps 6 and 7 can happen in whatever order fits your project.
Can I skip the audit step for a low-stakes internal report?
It's reasonable to do a lighter review for internal exploration than for a client-facing deliverable - see Auditing AI Classification Accuracy on Your Own Data for how to scale the depth of review to the stakes.
Does every step in this workflow cost credits?
No - category generation and classification are credit-based; editing, reviewing, correcting, building reports, and exporting are not. See What Actually Consumes Your Credits (and What Doesn't).
What if I need to add more responses after I've already built a report?
Re-run classification to bring new responses into your existing categories, then your report widgets will reflect the updated data automatically since they read live results.
Ready to start? See How Text Analytics Works: From Open-Ends to Categories.