This guide shows you how to build a repeatable Voice of Customer (VoC) research workflow — from collecting raw feedback to surfacing insights your marketing and product teams can act on. Most teams can set up the core system in one afternoon and start generating usable insights within a week.

What You'll Build

  • A structured feedback collection system across at least three customer touchpoints
  • A centralised tagging and categorisation layer that processes responses automatically
  • A sentiment analysis pipeline using a modern AI tool (tested with Claude 3.5 Sonnet and GPT-4o, as of mid-2026)
  • A weekly insight digest delivered to your Slack or email automatically
  • A living insight repository your team can query before writing copy, launching campaigns, or updating positioning

Prerequisites

  • Access to at least one customer feedback source (support tickets, review platforms, post-purchase surveys)
  • A free or paid account on Typeform, Tally, or a comparable survey tool
  • A Make (formerly Integromat) account — the free tier is sufficient to start
  • A Google Sheets or Airtable workspace for your insight repository
  • Basic familiarity with prompt engineering — no coding required

Step 1: Define the Questions Your Business Actually Needs Answered

Most VoC programmes fail before they start because they collect everything and learn nothing. Begin with three focused research questions.

How do you choose the right research questions?

Map your current business priority to one of three VoC categories: acquisition (why do customers choose you?), retention (why do they stay or leave?), or expansion (what would make them spend more?). Pick the category that matches your biggest revenue lever right now. Write one primary question per category you select, then add two supporting questions beneath each.

Example for an Australian SaaS company targeting retention:

  • Primary: "What nearly made you cancel in the last 90 days?"
  • Supporting: "What feature do you rely on most?" and "What would you tell a colleague about us?"

Common pitfall: Don't ask customers what they want. Ask them what they struggled with, feared, or celebrated. Behavioural framing produces 60–70% more specific responses than feature-request framing, based on established Jobs-to-be-Done research methodology.

Step 2: Set Up Three Feedback Collection Channels

Relying on a single channel introduces selection bias. Build three parallel streams from the start.

Channel 1 — Post-interaction surveys

Use Typeform or Tally to deploy a 3-question survey triggered after key moments: a completed purchase, a support ticket closed, or a free trial expiry. Keep the survey under 90 seconds. Embed it in your transactional email using a direct link — do not gate it behind a login. Aim for a 15–25% completion rate as a healthy benchmark.

Channel 2 — Review platform scraping

Use Phantombuster (free tier) or Browse AI to pull reviews weekly from Google Business Profile, Trustpilot, Capterra, or G2 depending on your industry. Export to a Google Sheet with columns: date, source, star rating, review text, reviewer type. This gives you unsolicited, unfiltered language — the most valuable input for copywriting.

Channel 3 — Customer interview recordings

Conduct four to six 20-minute customer interviews per quarter using Zoom or Google Meet. Record with permission. Use Otter.ai or Fireflies.ai to generate transcripts automatically. Store transcripts in a dedicated Google Drive folder. Even four interviews per quarter produce enough signal to validate or challenge what your surveys reveal.

Pro tip: If your team runs social media, untagged brand mentions contain some of the most honest customer language. Pair this workflow with a structured content planning system — Lenka Studio's free social media toolkit includes a content calendar that can help you track and align VoC insights with your publishing schedule.

Step 3: Build Your Centralised Insight Repository

Raw feedback is not an insight. You need a structured home before you start processing anything.

What should the repository structure look like?

Create an Airtable base (or Google Sheet) with the following columns:

  • Source — survey, review, interview, social
  • Date collected
  • Raw quote — verbatim, never paraphrased
  • Sentiment — positive, neutral, negative (auto-filled in Step 4)
  • Theme tag — pricing, onboarding, support, feature gap, competitor mention
  • Urgency — high, medium, low
  • Assigned to — marketing, product, support

Create a locked "Raw Quotes" view and a filtered "Actioned Insights" view. The separation prevents teams from accidentally editing source data.

Step 4: Automate Sentiment Analysis and Tagging with AI

Manual tagging does not scale. Automate it using Make and an AI model via API.

How do you set up the Make automation?

Build a Make scenario with this flow:

  1. Trigger: New row added to your Google Sheet or Airtable base
  2. Action 1: HTTP module sends the raw quote to the OpenAI API (GPT-4o mini works well here — it costs roughly $0.15 per 1M input tokens as of mid-2026 and is fast enough for batch processing)
  3. Action 2: Parse the JSON response and write sentiment + theme tag back to the repository row
  4. Action 3: If sentiment is "negative" and urgency is "high", post the quote to a dedicated Slack channel (#voc-alerts) immediately

Use this prompt in your HTTP module. Adjust the theme list to match your business:

You are a customer research analyst. Analyse the following customer feedback quote.
Return a JSON object with exactly three keys:
- "sentiment": one of ["positive", "neutral", "negative"]
- "theme": one of ["pricing", "onboarding", "support", "feature_gap", "competitor", "general"]
- "urgency": one of ["high", "medium", "low"]

Do not return anything except valid JSON.

Feedback: "{{1.raw_quote}}"

Expected result: Every new piece of feedback is tagged within 10–15 seconds of entering your repository, with zero manual effort.

Common pitfall: If the AI returns inconsistent tags, add a "valid values" constraint to your prompt and instruct it to default to "general" when uncertain. Do not let ambiguous tags enter your dataset unchecked.

Step 5: Build a Weekly Insight Digest

Tagged data sitting in a spreadsheet changes nothing. The digest turns data into decisions.

When should the digest run?

Schedule a Make scenario to run every Monday at 8:00 AM in your team's timezone. The scenario should:

  1. Query your repository for all rows from the past seven days
  2. Group results by theme tag and count occurrences
  3. Pull the top three verbatim quotes per dominant theme
  4. Send a formatted summary to a Slack channel or email using a simple template

Your digest template should answer five questions for the reader:

  • What was the most common theme this week?
  • What's the sentiment split (positive / neutral / negative)?
  • What are three exact customer quotes that illustrate the dominant theme?
  • Has this theme appeared for more than two consecutive weeks?
  • Which team owns the next action?

Pro tip: Keep the digest under 300 words. If it takes more than three minutes to read, teams stop reading it within a month.

Step 6: Translate Insights Into Marketing Assets

This is where VoC pays for itself. Customer language is your most underused copywriting resource.

How do you turn a VoC insight into usable copy?

Take the verbatim quote from your repository. Identify the specific phrase that describes a problem, outcome, or fear. Lift that phrase — nearly unchanged — into your headlines, ad copy, email subject lines, or landing page headers.

Example: A Canadian fintech SMB running this workflow found the phrase "I didn't realise it until the invoice was already late" appearing in six separate interviews. They rewrote their homepage hero from "Smart invoicing for small teams" to "Stop finding out about late invoices after it's too late." Their trial-to-paid conversion rate improved by approximately 18% over the following quarter.

Apply the same principle to:

  • Google Ads headlines: Mirror the exact anxiety your customers describe
  • Email subject lines: Use customer phrasing for the outcome they want
  • FAQ sections: Use real objections from negative-sentiment quotes
  • Case study framing: Use the "before" language customers use to describe their problem

If your brand positioning feels misaligned with what VoC is surfacing, run a quick audit using Lenka Studio's free brand health score assessment — it helps identify where your perceived value drifts from your communicated value.

Step 7: Review and Evolve the Workflow Monthly

A VoC workflow degrades if you don't maintain it. Spend 30 minutes each month on a quick review.

Check these four things every month:

  • Is your survey completion rate above 12%? If not, shorten the survey or change the trigger timing.
  • Are any theme tags becoming a catch-all? Rename or split overloaded tags.
  • Has a new competitor name appeared in quotes more than five times? Add it as a standalone theme.
  • Are insights being actioned or just read? If no "assigned to" column is being filled, run a 15-minute team review instead of async digest delivery.

Frequently Asked Questions

How many customer responses do I need before VoC insights are reliable?

Qualitative themes tend to stabilise after 15–20 responses for a focused question. For quantitative sentiment splits, aim for at least 50 responses per time period before drawing conclusions. More volume helps, but consistent collection matters more than hitting a specific number quickly.

Can I run this workflow without paying for any tools?

Yes, for a basic version. Use Google Forms instead of Typeform, the free tier of Make (1,000 operations per month), and Google Sheets as your repository. For AI analysis, OpenAI's API is pay-as-you-go with no monthly minimum — low-volume teams typically spend under $5 per month at GPT-4o mini rates.

Is this different from running a standard NPS survey?

NPS gives you a score. VoC gives you language. NPS tells you a customer is unhappy; VoC tells you exactly what they said when they were unhappy and what words they used. The two complement each other — NPS is a metric, VoC is a qualitative research system.

How is this different from social listening tools like Brandwatch or Sprout Social?

Social listening captures public mentions at scale but misses private feedback (support tickets, interviews, post-purchase surveys). This workflow covers channels social listening tools cannot access. Use both if budget allows; prioritise VoC first if you're choosing one.

What if customers don't respond to our surveys?

Survey fatigue is real. Try three fixes: send the survey within 15 minutes of the triggering event (response rates drop sharply after 24 hours), limit the survey to two or three questions maximum, and personalise the sender name to a real team member rather than a generic company address. These changes typically lift completion rates by 10–20 percentage points.

Next Steps

Your VoC workflow will be most valuable when it feeds directly into your marketing calendar, campaign briefs, and product roadmap. Start with one channel this week — a post-purchase survey is the fastest to launch. Add the AI tagging automation in week two once you have your first 10–15 responses to test against.

If you'd like help building this workflow into a broader marketing system — or connecting VoC insights to your content and paid media strategy — the team at Lenka Studio works with SMBs across Australia, Singapore, Canada, and the US to design and implement marketing systems that scale. Get in touch to talk through what your business needs.