Surveys are cheap, fast, and familiar. They answer exactly one question well: 'How many?' Every other question — why, driven by what, under which conditions — a survey only appears to answer. This guide shows when to stay with numbers, when a conversation is needed, and how to plan the move without losing your existing trackings.
Why this matters
Many insight programs keep measuring what they have always measured — because the time series matters. That's right. What's wrong is forcing every new question into the same template. The moment a decision depends on a 'why,' a number becomes an excuse: it shows something is happening but doesn't replace understanding.
Seven steps for the transition
- Diagnose the decision. Write down the specific decision the data has to serve. No decision, no method choice.
- Classify the question type. Counting questions ('How many?', 'How often?') belong in a survey. Understanding questions ('Why?', 'Driven by what?', 'When does it tip?') belong in a conversation.
- Choose the sampling logic. For understanding questions, heterogeneity matters more than representativeness — cover the relevant contrasts (roles, sites, tenure).
- Design the guide. Five to seven core questions with clear probing rules. No scales, no multiple-choice anchors.
- Pilot. Run five to ten conversations first. Check whether probing actually deepens — and whether the guide stays under twelve minutes.
- Migrate existing trackings. Keep your core KPI as a quantitative anchor (e.g. eNPS, CSAT). Add a conversation program for every why-question. The time series stays intact.
- Governance sign-off. Involve privacy, works council, and legal — before rollout, not after.
Decision matrix
| Decision depends on … | Method | Why |
|---|---|---|
| Frequency or distribution | Survey | Numbers are the clean tool for shares. |
| Cause or condition | Conversation | Why-questions require probing. |
| Prioritising a backlog | Both | Survey ranks, conversation explains outliers. |
| Outcome for the board | Conversation | Boards need voice, not just percentages. |
Common pitfalls
- 'Let's just ask more': open-text fields in surveys produce shallow answers, not depth.
- Double measurement: running both in parallel without clear roles confuses stakeholders.
- No anchor: dropping the core KPI kills comparability across time.
- Guides that are too long: past twelve minutes completion drops.
Sign-off checklist
- Decision documented
- Question type classified
- Sampling logic justified
- Guide under twelve minutes
- Pilot with five to ten people completed
- Core KPI preserved as time series
- Privacy, works council, legal signed off
Frequently asked questions
Do we have to drop our survey when we introduce conversations?
No. In most cases the core KPI stays as a quantitative anchor; conversations complement it wherever numbers cannot explain a cause.
How large does the sample for a conversation program need to be?
Not large, but diverse. Rule of thumb: at least three to five conversations per relevant contrast, typically 40 to 120 in total.
When does the switch pay off?
As soon as a decision depends on a why and you no longer want to rely on a number alone. For purely descriptive metrics a survey remains the cheaper option.