TL;DR
- Voice of customer across business units works when the group fixes three things (the metric definition, the follow-up promise to unhappy customers, the reporting format) and leaves wording, channel and case ownership to the units.
- A blended group score can stay flat while every unit declines, because the mix of responses between units changes (Simpson's paradox). Never publish the group number without the unit numbers next to it.
- Weight the roll-up by revenue or account value, not by response count. The unit that answers most is rarely the unit that matters most.
- Counterintuitive: do not start by making the questionnaires identical. Start by making the follow-up identical. A shared response promise is worth more than shared wording.
Contents7 sections
What does running voice of customer across business units mean?
Running voice of customer across business units means one company with several units, brands or countries, and one listening discipline that produces numbers you can compare without fooling yourself. It does not mean sending the same survey to everyone, and it does not mean letting every unit do what it likes. The model that holds up is one governance layer and local execution. The group owns the metric definition, the promise made to unhappy customers and the way results are reported. The units own the wording, the channel and the cases.
This article is for the stage after the first programme works. If you are still building the basics, start with the voice of customer guide. If your question is who in the organisation should own the whole thing, read CX governance first.
Our position is simple and a little uncomfortable: the group score is the least useful number a multi-unit programme produces. You still need it, because the board asks for one number. The point of building it properly is to be able to open it up.
Why does one blended score hide more than it shows?
Because a group NPS is an average of units, and an average changes when the mix changes, not only when the units change. Statisticians call this Simpson's paradox: a trend in the aggregate can differ from, or even reverse, the trend in every subgroup. In a VoC programme it shows up the moment one unit starts answering more than another.
Here is a hypothetical example for Nordika A/S, with two units. The group NPS is the response-weighted average of the unit scores, which is exactly how a single pooled NPS is calculated. The figures are SurveyGauge's own calculation.
| Unit | Responses, year 1 | NPS, year 1 | Responses, year 2 | NPS, year 2 |
|---|---|---|---|---|
| Unit A (Nordic) | 100 | 50 | 200 | 45 |
| Unit B (Export) | 300 | 20 | 200 | 10 |
| Group, weighted by responses | 400 | 27.5 | 400 | 27.5 |
The group score did not move. Unit A fell by 5 points and Unit B fell by 10. The only thing that changed is that Unit A, the healthier unit, went from a quarter of the responses to half of them. A dashboard that shows 27.5 both years tells the board that nothing happened.
Now assume Unit A represents 30% of revenue and Unit B 70%. Weighted by revenue, the group score falls from 29 to 20.5. That is the number that describes what happened to the business. Same data, opposite message, and the only difference is the weighting choice nobody wrote down.
The rule that follows is cheap to enforce: the group number never appears without the unit numbers beside it, and the lowest unit score appears in the same row. If the group has to report one figure to the board, report the weighted score and the floor together.
- Year 1
- Year 2
- Unit A (Nordic)5045
- Unit B (Export)2010
- Group, weighted by responses27.527.5
- Group, weighted by revenue2920.5
What should be centralized, and what should stay local?
Centralize what has to be identical for the numbers to mean the same thing, and what customers experience as the company's promise. Keep local what depends on language, market and relationship.
| Element | Group owns | Unit owns |
|---|---|---|
| Metric definition (question, scale, which respondent roles count) | Yes | No |
| Follow-up promise (who contacts a detractor, how fast, how it is closed) | Yes | No |
| Reporting cadence and format, always with unit numbers shown | Yes | No |
| Weighting rule for the roll-up | Yes | No |
| Survey wording, translated faithfully and reviewed by a native speaker | Reviews | Yes |
| Channel mix (email, in-app, phone, meeting) | No | Yes |
| Case ownership and the follow-up conversation | No | Yes |
The counterintuitive part is the order. Most programmes start with the questionnaire, because it is visible, and spend a year arguing about wording. Start with the follow-up instead. A group that agrees on a shared closing-the-loop standard and a shared detractor recovery playbook gets consistent customer treatment on day one, even while the units still ask slightly different questions. A group with identical wording and no shared response promise gets comparable scores for customers nobody calls.
How should you weight the group roll-up?
By what the business cares about, stated in advance, and with coverage shown next to the score. Weighting by response count is the default in most tools, and it is the one that lets the most survey-active unit write the group story.
| Weighting basis | What the group number answers | Main risk |
|---|---|---|
| Response count | How do the people who answered feel? | The loudest unit dominates |
| Customer count | How does the average customer feel? | A small account counts as much as a key account |
| Revenue or account value | How do the customers that carry the business feel? | Hides small accounts that could grow |
| Revenue, plus the lowest unit score shown beside it | How do paying customers feel, and where is the weakest spot? | Needs a clear definition of revenue |
Our recommendation is the last row. Choose revenue or contract value, write the rule down once, and publish the response coverage with every report: how many customers, how much revenue and how many roles inside each account the score actually represents. CustomerGauge's NPS Benchmark Report Decade Edition, based on 340 B2B CX programmes surveyed between 6 May and 13 July 2026, and a CustomerGauge blog post of 9 October 2026 by Ian Luck, point at the same problem: roughly 40% of respondents account for 60 to 70% of the responses, and 60% of programmes do not know how much revenue their feedback covers. Those are vendor figures about programmes in general, not about multi-unit groups, but the mechanism is exactly the one that distorts a group roll-up. For the account-level side of this, see account-based CX, and for what small response counts do to a unit score, see B2B survey sampling with a small n.
One more rule that costs nothing: do not rank units in a league table before you have checked response counts. A unit with 30 responses and a unit with 300 do not have scores of equal precision, and a league table makes that invisible.
- Share of responses, year 2
- Share of revenue
- Unit A50%30%
- Unit B50%70%
What changes with language and country?
More than the translation. The survey wording has to be faithful and reviewed by a native speaker, but the bigger differences sit in the follow-up conversation and in how the score is read.
On language, the best-known study is CSA Research's "Can't Read, Won't Buy" from 2020, which surveyed 8,709 consumers in 29 countries. It found that 76% prefer to buy products with information in their own language, and that 75% are more likely to buy the same brand again if customer care is in their language. These are consumers, not B2B buyers, so read them as a direction rather than a B2B benchmark. The direction is still useful: a detractor call that happens in a second language is a weaker recovery than one that happens in the customer's own.
On scores, our experience is that customers in different countries use rating scales differently, so raw scores from different countries should not be compared before each unit has been compared with its own history. The group question is "is each unit moving in the right direction", not "which country is happiest".
On data, running a programme across countries means personal data moves. Group reporting should be built on minimum data: scores, case status and response times travel to the group, free-text notes and contact details stay with the unit that owns the relationship. The principles are covered in GDPR and customer feedback in B2B. If the group platform or a sub-processor sits outside the EU, check the transfer basis. The EU-US Data Privacy Framework was upheld by the EU General Court on 3 September 2025, but the challenger appealed to the Court of Justice on 31 October 2025, according to a WilmerHale analysis of 1 December 2025. Keep standard contractual clauses as a fallback and check the current status of the appeal before you rely on the framework alone.
How does the Autorola Group run voice of customer across units?
The Autorola Group runs voice of customer across three business units in 19 countries with one governance layer and local execution. The group provides the governance and the units execute locally. That is the same principle as in the table above, and it is the reason the case is useful: it shows that the model works at a scale where "just standardize everything" would be impractical and "let every unit decide" would be unreadable at group level.
What generalises is the division of labour, not the exact setup, which is described in the Autorola Group case itself. Your units will have different channels, different languages and different owners. What they cannot have is a different definition of a detractor or a different promise about what happens next.
Where does SurveyGauge fit?
SurveyGauge enforces one metric definition and one follow-up standard across units, while each unit keeps its own channels, language and case ownership. The group sees the weighted roll-up with the unit scores and response coverage beside it, and the units see their own cases. Because the platform comes with strategic CX advisory in the same subscription, the governance decisions in this article, such as the weighting rule, the follow-up promise and the floor reporting, are worked out with you rather than left as settings to figure out alone.
If you have only one unit running, do not buy governance yet. Get one programme working first, and bring the second unit in once you can show the first one closes its loop.
Frequently asked questions
What is voice of customer across business units?
It is running one voice of customer programme across several business units, brands or countries so that scores mean the same thing everywhere and customers get the same follow-up. The group fixes the metric definition, the follow-up promise and the reporting format, and the units keep wording, channels and case ownership.
Should every business unit use identical survey questions?
The metric question, scale and respondent roles should be identical, otherwise scores cannot be compared. Everything around it, such as translated wording, channel and timing, can differ. If you must choose where to be strict, be strict about the follow-up process before the wording.
How do you stop one business unit from dominating the group score?
Weight the roll-up by revenue or account value instead of response count, and publish response coverage with every report. Show the lowest unit score next to the group figure so a strong, survey-active unit cannot hide a weak one.
What is Simpson's paradox in a VoC programme?
It is when the group score stays flat or rises while every unit declines, because the share of responses coming from each unit changes. In a hypothetical Nordika A/S example, both units fall by 5 and 10 points while the group NPS stays at 27.5 because the healthier unit supplies a larger share of the responses.
Do you need one platform or one tool per business unit?
One platform is the practical choice when you want one metric definition, one follow-up standard and a roll-up you can trust, because the rules are enforced in one place. Separate tools per unit tend to produce separate definitions that cannot be combined later.
Who should own a multi-unit voice of customer programme?
The group owns the definition, the standard and the reporting, usually through a named CX owner at group level, while each unit has an owner for its own cases and follow-up. The ownership question in detail is covered in the article on CX governance.
How many business units justify dedicated VoC governance?
Two units are enough once their scores are reported together, because the mix problem appears as soon as there is a group number. Heavy governance is not needed at two units, but a written metric definition, a follow-up standard and a weighting rule are.
Written by
SurveyGauge Team
The advisors and developers behind SurveyGauge. We run customer satisfaction programmes for B2B companies, from 20 key accounts to global banks.

