Driver Analysis

Suppose a customer satisfaction survey asked people to rate price, service quality, staff friendliness, ease of use and product quality. If price satisfaction came out lowest, should the company fix price first?

Not necessarily. An item customers rate poorly may not be strongly related to overall satisfaction. Conversely, an item customers are reasonably happy with may still be an important factor in explaining overall satisfaction.

So to improve customer satisfaction, you need to look beyond each item's score and check which factors are closely tied to overall satisfaction, and how those factors are currently rated.

Two methods help here: driver analysis and importance-performance analysis (IPA).

Driver analysis looks for the key factors related to an outcome such as overall satisfaction or loyalty. IPA compares each item's importance with its current rating to show improvement priorities visually. Combined, the two let you find items that are strongly related to satisfaction but currently rated low.

In this article we'll cover the basic ideas behind driver analysis and IPA, how importance is calculated, a worked example, how to read the IPA quadrants, and what to watch out for when setting improvement priorities.

For example, suppose 500 customers were surveyed on overall service satisfaction along with service quality, price satisfaction, staff friendliness, ease of use and problem-resolution speed.

You can then treat overall satisfaction as the outcome variable and analyze how the other items relate to it. This shows which items are also rated highly by more satisfied customers, and which factors account for relatively more of the differences in satisfaction.

Driver analysis is not so much the name of a single statistical test as an analytic approach that, depending on the research goal, draws on methods such as regression, relative importance analysis and SHAP.

Also, the word "driver" does not necessarily mean a causal factor. Relationships found in observational data such as surveys are generally best interpreted as factors associated with satisfaction.

Driver analysis, by contrast, looks at how strongly each item is related to overall satisfaction. Suppose, for example, that ease of use has a mean score of 4.5 and price satisfaction 3.2. A simple satisfaction analysis might conclude that price needs work because its score is low.

But a driver analysis might show that price satisfaction is only weakly related to overall satisfaction, while ease of use is strongly related. In that case, mean scores alone can't tell you how much each item needs improving.

The two analyses don't replace each other. To decide where to actually improve, it's best to look at both each item's rating and its importance.

That importance may be the importance customers rated directly in a survey question, or it may be estimated with a statistical model. The two differ in meaning and in how they are computed, so always keep them apart.

Stated importance (explicit importance)

You ask customers directly how important each item is. For example, a question such as "How important is price when you use the service?" could be answered on a 1-to-5 scale. Other items such as staff friendliness, quality and ease of use are rated the same way.

Stated importance has the advantage of showing how important respondents themselves think each item is. But respondents may rate everything as important, or answer in ways that differ from how they actually behave.

Statistical importance (derived importance)

Importance is estimated by statistically analyzing the relationship between the items customers rated and overall satisfaction. For example, you can run a multiple regression with overall satisfaction as the dependent variable and service quality, price satisfaction and the other items as independent variables.

You can then use the regression coefficients or relative importance measures to see how strongly each item is related to overall satisfaction.

A large regression coefficient, however, does not necessarily mean that improving that item would have the biggest real effect. And when the variables are highly correlated with each other, it can be hard to compare importance using individual coefficients alone.

For example, staff friendliness, problem-solving ability and response speed are all part of the service quality customers experience. Putting these variables into one regression model can produce high correlations among them.

Comparing importance using standardized regression coefficients alone can then give unstable or hard-to-interpret results. Relative importance analysis takes this problem into account and estimates how much each predictor contributes to explaining the variation in the outcome.

Common methods include relative weight analysis, dominance analysis and Shapley value–based importance. Some methods divide the model's explanatory power among the variables and rescale it so the total is 100%.

Suppose, for example, you got the following results.

Here, a relative importance of 30% for service quality means that, under the importance method used, 30% of the explained variance was allocated to service quality.

Typically you draw a four-quadrant chart with importance on the horizontal axis and satisfaction or performance on the vertical axis, or the other way around.

Items that are important but currently rated low can be treated as the first candidates for improvement. Items high on both importance and satisfaction can be read as current strengths to maintain.

Some people use the term performance instead of satisfaction here. Be clear about what each axis means, depending on whether you use satisfaction scores customers gave or actual operational performance metrics.

IPA has the advantage of summarizing complex results visually, which makes it handy for decision making. But a quadrant position alone does not settle budget allocation or the actual effect of an improvement.

① High importance · High satisfaction — Maintain strengths

Items strongly related to customers' overall satisfaction that are also rated well now. This is where the company is doing well, so the key is to maintain the current level and keep quality from slipping.

② High importance · Low satisfaction — Concentrate here

Items that are important but currently rated low. In IPA this is the first region to consider for improvement, and it's worth looking closely at the causes of the problem and how much it can be improved.

③ Low importance · Low satisfaction — Low priority

Items rated low now but also relatively weakly related to overall satisfaction. When allocating limited resources, you can give them lower priority than the other regions. However, items tied to legal obligations, safety or basic service quality may need to be fixed even if their statistical importance is low.

④ Low importance · High satisfaction — Keep & monitor

Items with high current satisfaction but low relative importance. Further investment may have limited effect, so you can keep the current level while reviewing how resources are allocated.

This region is often described as "possible overkill," but the fact that importance is low and satisfaction is high does not by itself show that money is actually being wasted.

The survey asked about overall service satisfaction along with five specific items. Satisfaction with each item was collected on a 5-point scale, and statistical importance was calculated with relative importance analysis.

Driver analysis and satisfaction survey results

In these results, service quality has the highest relative importance at 30% and also high satisfaction at 4.3. Resolution speed, on the other hand, has the second-highest importance at 25% but the lowest satisfaction at 3.0. Price satisfaction was also low at 3.2, but its relative importance was 15%.

Placing these results on an IPA chart makes it easier to consider which item to improve first.

Depending on the purpose of the analysis, a cutoff can be the mean, a target value set in advance, or some other criterion. Here we use these two values for ease of explanation.

Each item's position can be interpreted as follows.

In this example, resolution speed is the item to consider improving first. It is fairly strongly related to overall satisfaction, but customers currently rate it low.

Service quality, by contrast, is high on both importance and satisfaction, so it is an existing strength to maintain. Price satisfaction is low, but its importance is relatively small, so rather than making it the top priority just because its score is low, it's better to weigh its likely payoff against the other items.

An item that sits exactly on a cutoff line, like ease of use, should be handled with a classification rule set in advance or flagged separately as a borderline item.

Even with the same data, the quadrant an item falls into can change depending on where you set the importance and satisfaction cutoffs. Common ways of setting them are:

For example, at a company with high average satisfaction, splitting the quadrants at the mean across all items can put items with high absolute satisfaction into the relatively low region. So "low satisfaction" in IPA does not necessarily mean absolute dissatisfaction.

Also, if relative importance has been rescaled to total 100%, its mean depends on the number of items. For example, the mean importance of 5 items is 20%, but with 10 items it is 10%.

So explain exactly what the importance cutoff means, and when comparing quadrant results across different analyses, check that the item sets and cutoffs are the same.

But the importance customers state may not fully reflect the statistical relationship with actual satisfaction. For example, respondents may say price is very important when ease of use or their experience of getting problems resolved is actually more closely tied to overall satisfaction.

With driver analysis, you can use statistically estimated importance on the IPA importance axis. The process generally goes like this:

1. Collect overall satisfaction and the specific rated items.
2. Calculate the mean satisfaction for each item.
3. Identify the key related factors with regression or relative importance analysis.
4. Plot the calculated importance and mean satisfaction on the IPA chart.
5. Review improvement tasks, focusing on items that are important but rated low.

Set up this way, you can explore where to improve by focusing on items whose relationship with overall satisfaction shows up strongly in the data, rather than items customers say are important. Using statistically estimated importance in IPA, however, does not mean the actual effect of an improvement has been verified.

Regression is a statistical method for estimating the relationship between independent and dependent variables. Driver analysis focuses on an analytic goal: identifying, among several candidate factors, the key ones related to a particular outcome. IPA goes one step further and combines importance with current ratings to visualize improvement priorities.

For example, regression can confirm that resolution speed is positively related to overall satisfaction.

Driver analysis can then examine how important resolution speed is compared with the other rated items. After that, IPA can compare resolution speed's importance with its current satisfaction to help decide improvement priorities.

In this way, the three methods answer different questions and can be linked within a single analysis.

For example, new customers may care most about ease of use, while long-term customers may be more sensitive to resolution speed or service reliability. To check for such differences, you can run driver analysis separately for each customer group.

In these results, ease of use has the highest relative importance for new customers, while service quality and resolution speed are relatively more important for long-term customers. But the fact that the two groups' importance values differ does not by itself let you conclude that there is a statistically significant group difference.

To test differences between groups, you need to consider additional analyses such as an interaction analysis or a test of the difference in coefficients. You should also check that each group's sample size is adequate and that the same measures were used.

Importance ranking chart

Shows each item's relative importance as a horizontal bar chart. This shows which items are relatively closely tied to overall satisfaction. Importance is usually sorted from highest to lowest, and it's good to state the calculation method alongside.

IPA quadrant chart

Plots each item's importance and satisfaction as coordinates. Drawing the cutoff lines on both axes and marking the four regions makes it easy to spot items to consider improving and strengths to maintain. Especially when there are many items, it helps to label the points directly with item names or to show details on mouse hover.

IPA comparison across groups

You can also compare IPA results across groups, such as new versus long-term customers, customers by region or customers by age group. Using the same axis ranges and cutoffs makes differences between groups easier to compare.

However, if the groups use different importance methods or include different items, a simple comparison of positions can be misleading, so keep the analysis conditions the same.

High importance does not mean a large improvement effect.

Just because an item is closely tied to overall satisfaction doesn't mean improving it will raise satisfaction by the same proportion. For example, a 30% relative importance for service quality doesn't mean that improving quality alone can change 30% of overall satisfaction.

Confirming the actual effect of an improvement requires further checks, such as before-and-after comparisons or experiments.

Check the correlations among the rated items.

Service quality, staff friendliness, resolution speed and similar items can be closely related to one another. If high correlations among these variables aren't properly accounted for, the importance of individual factors may be estimated unstably. So beyond comparing simple regression coefficients, consider methods such as relative importance analysis.

Items with low satisfaction don't automatically need fixing.

An item that shows low satisfaction in IPA is not necessarily the most important target for improvement. Along with importance, consider the actual cost of improvement, the impact on customers, the company's strategy and legal or safety requirements. In particular, there may be a baseline level of service that must be met even if its statistical importance is low.

Results can change with the quadrant cutoffs.

Each item's position can change depending on whether the IPA cutoffs are set at the overall mean or at an absolute target level. So when reporting the results, state the importance and satisfaction cutoffs clearly.

Survey results can contain common method bias.

When the same respondents rate both the specific items and overall satisfaction at the same time, response tendencies or the measurement method can make the relationships between variables look stronger than they really are. So where possible, check the results against objective behavioral data or measurements taken at a different time.

Importance can differ across customer groups.

An item that is highly important for customers overall is not necessarily equally important in every customer group. So, as needed, you can run separate analyses for groups such as new customers, long-term customers and key accounts and check for differences.

Using the online service company example from earlier, you could write it up like this:

In an actual report, state which method was used to calculate relative importance. And rather than presenting only the IPA quadrants, providing the original satisfaction scores and importance measures as well makes it easier to see what the results are based on.

When making the final call on improvement tasks, it's good to also review how feasible each improvement is, its expected cost and its impact on the customer experience.

Customer satisfaction surveys (CSAT)

Analyze the relationship between specific items such as service quality, price, staff friendliness and ease of use and overall satisfaction, and review improvement priorities.

Brand and marketing research

Examine how brand image, product quality, price competitiveness, ease of purchase and so on relate to overall brand ratings or purchase intent.

Product and service improvement

Analyze ratings of user experience, feature satisfaction, experience with errors, processing speed and so on to inform product improvements.

Employee satisfaction surveys

Analyze how the work environment, pay, growth opportunities, organizational culture, manager support and so on relate to overall job satisfaction.

Evaluating educational services

Examine how teaching quality, course materials, learning support, assessment methods and so on relate to overall satisfaction with the education.

Evaluating healthcare and public services

Compare the importance and satisfaction of several rated aspects together, such as access to services, wait times, staff interactions, quality of guidance and facilities. In this way, driver analysis and IPA don't stop at rating many items; they help you decide where to put limited resources first.

With regression or relative importance analysis, you can see how strongly several rated items are related to the outcome variable. IPA compares this importance with current satisfaction or performance to show improvement priorities visually.

Using the two together takes you beyond simply finding items with low satisfaction scores: you can prioritize items that are strongly tied to overall satisfaction and currently rated low.

But statistically estimated importance does not guarantee an actual improvement effect. So alongside the results, consider the cost of improvement, feasibility, customer characteristics and the company's strategic goals.