Second-Pass Classification: Drilling Deeper Into One Category (2026)

AI-Powered Analysis
Tutorial
Updated Sep 02, 2026

Knowing that 34% of survey responses fall into a "pricing" category is a finding, but it's a shallow one - it groups together someone who thinks the product is overpriced across the board, someone who thinks one specific plan tier is a bad deal, and someone comparing your price directly against a named competitor's, as though all three were saying the same thing. They're not. A first-pass classification is built to answer a broad question well; it isn't built to answer the narrower, more specific question that often turns out to matter more once you know which broad category deserves the closest look. That's what a second pass, scoped to just the responses inside one category, is for.

Table of Contents

  1. What a Second Pass Actually Does
  2. When It's Worth the Extra Step
  3. Scoping a Second Pass Well
  4. How Deep Is Too Deep
  5. A Worked Example
  6. FAQ

What a Second Pass Actually Does

A second-pass classification takes the responses already sorted into one category from an earlier pass and runs a fresh classification on just that subset, with its own goal and its own set of sub-categories, without touching or re-sorting anything outside that original category. The result is a category that used to be a single number - "34% pricing" - broken into its own internal structure: maybe 40% of those pricing responses are about overall price level, 35% are about a specific plan's value, and 25% are direct competitor comparisons. Nothing about the original, broader classification changes; the second pass is purely additive, adding a layer of detail underneath one category rather than replacing or redoing the first pass.

When It's Worth the Extra Step

A second pass earns its place specifically when a category is both large and actionable in a way that a single number can't fully support - if "pricing" is your biggest category and pricing decisions are actually on the table, knowing the internal breakdown genuinely changes what gets proposed. It's less useful for a small category that isn't going anywhere strategically regardless of its internal structure, or for a category where the first pass already answered the specific question you needed answered. A good rule of thumb: if you find yourself wanting to read through a category's responses by hand to understand what's really going on inside it, that's usually a sign a structured second pass would save time and produce a more consistent, reportable answer than an informal read-through would.

It's also the right move whenever a category turns out to be a genuine grab-bag - a "shipping" category that on closer reading contains complaints about speed, damage, and communication as three genuinely distinct problems calling for three different fixes is exactly the kind of situation a second pass is built to untangle, rather than trying to redefine the original category boundaries and rerun the whole first-pass classification from scratch.

Scoping a Second Pass Well

The same discipline that makes a first-pass classification goal work well - a clear angle, a sense of granularity, anything already known worth anchoring against, covered in more depth in our guide on writing a good classification goal - applies just as directly to a second pass, with one addition: it's worth being explicit about what you're hoping the sub-categories will reveal that the parent category didn't. If the goal for drilling into "pricing" is simply "find sub-themes in the pricing responses," you're likely to get a similarly generic split as an underspecified first-pass goal would produce. A goal like "split by whether the complaint is about overall price level, a specific plan's value, or a named competitor comparison" gives the second pass the same kind of clear, actionable target a first pass benefits from.

How Deep Is Too Deep

A second pass is usually as deep as it's worth going, and a third pass - drilling into a sub-category of a sub-category - is worth a moment's pause before running. Each additional layer shrinks the underlying response count further, and a sub-sub-category built on forty or fifty responses starts running into the same small-sample caution that applies to any segment-level finding - see our guide on why segment-level results are noisier than the topline for the underlying reasoning. If a category is large enough that a third layer of drilling would still leave each resulting group with a reasonably sized sample, it can be worth it; if it would leave you reading percentages built on a couple dozen responses each, the honest move is treating that layer as a qualitative read of individual responses rather than a percentage breakdown that implies more precision than the sample size actually supports.

A Worked Example

A subscription meal-kit company's first-pass classification of 800 cancellation-survey responses shows "too expensive" as the largest category at 29%, roughly 230 responses. Rather than reporting that number alone, the team runs a second pass scoped to just those 230 responses, with a goal of splitting by the specific nature of the price complaint: is it about the price relative to grocery store alternatives, the price relative to a specific promotional rate that expired, or the price relative to portion size or quality received. The results split roughly 45% grocery-alternative comparisons, 33% expired-promotion complaints, and 22% portion-or-quality-relative-to-price complaints - three meaningfully different findings hiding inside the original single "too expensive" number, each pointing toward a different lever: competitive positioning messaging, promotional pricing strategy, and portion sizing, respectively. The team prioritizes the expired-promotion segment first, since it's the most directly and cheaply addressable of the three, a decision the original single-category finding wouldn't have been specific enough to support on its own.

FAQ

Does a second pass cost more or take longer than a first pass?
It's scoped to a smaller subset of responses, so it's typically faster to run than the original first pass, though the setup work - writing a good goal and reviewing the resulting sub-categories - takes real time regardless of the subset's size.

Can I run a second pass on more than one category from the same first-pass classification?
Yes - there's no restriction to just one; it's common to drill into your two or three largest or most strategically relevant categories separately, each with its own tailored second-pass goal.

Should sub-category percentages be reported as a share of the whole dataset or just the parent category?
As a share of the parent category, generally - "45% of pricing complaints were about grocery-store comparisons" is a clearer, more useful claim than converting that back into a share of all 800 original responses, which would understate how concentrated the finding actually is within the group it's describing.

How do I know if a category is a genuine grab-bag worth splitting versus a properly coherent single theme?
Read fifteen or twenty responses from the category together. If they clearly cluster into two or three distinct sub-stories on a quick read, it's a grab-bag worth a second pass; if they all genuinely feel like variations on one consistent theme, the category is probably fine as it stands.


For the full classification workflow, see Introduction to Text Analytics and Writing a Good Classification Goal.

second pass classification drill down survey data nested classification sub-category analysis

Related Articles

The Ethics of Letting AI Read Your Customers' or Employees' Words (2026)

Running open-ended feedback through an AI classifier is a practical, increasingly ordinary choice - and it's also a choice that involves someone else's words, often written under an assumption of who or what would actually be reading them. This guide covers the genuine ethical considerations worth thinking through before adopting AI-assisted analysis of customer or employee feedback: consent and expectation, anonymity, and what respondents were actually told.

How Many Human-Coded Responses Do You Need to Validate an AI Classifier? (2026)

Checking whether an AI classifier is trustworthy means hand-coding a sample and comparing it to the AI's output - and the obvious next question is how big that sample needs to be. Too small, and the check itself is unreliable; too large, and you've spent more effort validating than the original classification saved you. This guide covers what research on validation set sizing actually shows, and a practical range for everyday business use.

Prompt Engineering for Qualitative Research: A Non-Technical Introduction (2026)

\"Prompt engineering\" sounds like a technical skill for people who write code, and for the purposes of qualitative research, it's closer to a writing and thinking skill - the same instinct that makes someone a clear research brief writer translates almost directly into getting better results from an AI tool. This guide introduces the core ideas in plain language, for researchers and analysts who've never written a line of code and don't need to.

Sentiment Analysis and Thematic Analysis Are Not the Same Thing (2026)

\"We did sentiment analysis on the feedback\" and \"we did thematic analysis on the feedback\" get used almost interchangeably in casual conversation, and they describe two different questions with two different kinds of answers. One tells you how people felt. The other tells you what they were talking about. Confusing the two - or assuming one substitutes for the other - is a quietly common source of thin, unconvincing findings from open-ended data.

AI vs. Manual Coding: How to Decide Which One Your Project Needs (2026)

Neither AI-assisted coding nor fully manual coding is the universally correct choice - they trade off speed, cost, auditability, and nuance differently, and the right pick depends on what your specific project actually needs from its analysis. This guide covers a practical decision framework: the questions worth asking about your stakes, your timeline, and your audience before choosing a method, plus the hybrid approach most real projects actually end up using.

We value your privacy

We use cookies and similar technologies to improve your experience, analyze site traffic, and personalize content. Learn more