Key driver analysis

Key driver analysis finds which factors move an overall score, such as NPS, satisfaction or a star rating, and ranks them by how much each one moves it: not by how often it comes up or how well it is rated. The drivers can be rating questions in the survey, or the themes customers raise in their comments.

What key driver analysis is

Every key driver analysis has two parts: one outcome, the score you report, and several possible drivers, the things that might push it up or down. The analysis looks at all responses together and asks how much the score changes with each driver, with the others held apart.

The result is a ranked table: the drivers that move the score most at the top. “Driver analysis” and “key driver analysis” mean the same thing.

OutcomeThe overall score: likelihood to recommend from 0 to 10, overall satisfaction, a star rating.

DriversWhat might explain it: ratings of price, support or delivery, or themes such as “waited too long” in the comments.

EffectHow much higher or lower the score is with each driver. The key drivers are the ones with the largest effect on the score overall.

Two kinds: rating questions and comments

The classic version needs a survey built for it: an overall question plus a block of rating questions, one per driver. The text version works on the survey you already run, as long as it has an open comment next to the score.

	Rating questions	Themes in comments
Drivers	One rating question per factor, chosen before the survey	Themes customers raise in their own words
Survey length	Long: every respondent rates every factor	Short: a score and one open question
Blind spots	Factors nobody thought to ask about	Factors customers rarely put into words
Evidence	Ratings	The comments behind every number

When it helps

Key driver analysis earns its keep when you have one score, many possible reasons and limited time to fix them.

Choosing what to fix firstA ranked list turns a pile of complaints into an order of work.

Explaining a move in the scoreWhen the score drops, the drivers show which themes grew or got more costly.

Settling arguments about anecdotesA complaint that feels loud and one raised quietly by hundreds can be compared on the same scale.

Less useful with very few responsesWith a few dozen comments, most drivers rest on too few mentions to rank. Read the comments instead.

How to read a driver table

A useful driver table answers four questions per row: how big is the effect, how sure is it, how often does it come up, and what does it cost the score overall.

In the sample report of 4,000 clothing reviews, fit or size is the most mentioned complaint, in 32 % of reviews, at 0.30 stars less per mention. “Doesn’t look flattering” comes up in 15 %, at 0.94 stars less. Multiply effect by frequency and the second costs the average more: 0.14 stars against 0.09. The method page explains each column.

Effect per mentionHow much lower or higher the score is when the driver comes up. Large means each case hurts.

Likely rangeWhere the true effect most likely lies. A range that crosses zero means the data cannot tell whether the driver helps or hurts.

CountHow many responses the number rests on. Few mentions give a wide range.

Cost to the averageEffect times how often it comes up. Rank by this column to decide what to fix first.

Pitfalls

A driver table looks precise. These are the ways it can mislead.

Drivers that come togetherUnhappy customers rate everything low and raise several complaints at once. Compared one at a time, each driver takes credit for the others. In our test of a chat model on restaurant reviews, staff complaints came out at −2.67 and −2.5 stars this way, against about −1.5 on their own. See the test.

Small numbersA driver with a dozen mentions can show a huge effect by chance. Look at the range and the count before the effect. Leverboard marks themes with fewer than 30 comments as low data.

Cause and associationA driver analysis shows which factors go with lower scores, not that they cause them. Something else can sit behind both. Use the ranking to decide where to look, and the comments to understand why.

Frequent is not expensiveRanking by mentions answers a different question. The most common complaint is often not the one that costs the most.

Asking customers what mattersAsked directly, customers often rate most factors as important. A driver analysis measures what moves the score instead of asking.

Questions

What is driver analysis?

Driver analysis, or key driver analysis, finds out which factors go with higher or lower values of an outcome such as NPS, satisfaction or a star rating, and ranks them by how much each one moves it.

What does key driver mean?

A key driver is one of the few factors that move the overall score the most. Fixing a key driver is expected to change the score more than fixing a factor that customers mention often but that costs little.

How many responses do I need for a key driver analysis?

It depends on how often each driver comes up, not only on the total. In a Leverboard report, themes with fewer than 30 comments are marked low data. A theme raised in 10 % of comments reaches 30 mentions at about 300 responses with comments.

Is key driver analysis the same as correlation?

No. A correlation looks at one driver at a time, so drivers that come up together take credit for each other. A key driver analysis looks at all drivers together and gives each one its separate effect.

Can I run a key driver analysis on NPS comments?

Yes. The themes in the comments become the drivers and the 0 to 10 score is the outcome. You need an export with a score and a comment per response.

Run a driver analysis on your comments

Your score says how many customers are unhappy. Your comments say why, and Leverboard measures which reason costs the most. Upload a CSV or Excel export with a score and a comment per response, from any survey tool, and get the ranked report by email within one working day.

Try it on your data See a sample report

First analysis free. No card.