{"id":3593,"date":"2026-08-24T00:00:00","date_gmt":"2026-08-23T22:00:00","guid":{"rendered":"https:\/\/humes.pl\/key-driver-analysis-what-truly-drives-satisfaction-and-nps-and-what-merely-correlates-with-them\/"},"modified":"2026-08-25T15:31:45","modified_gmt":"2026-08-25T13:31:45","slug":"key-driver-analysis-what-truly-drives-satisfaction-and-nps-and-what-merely-correlates-with-them","status":"publish","type":"post","link":"https:\/\/humes.pl\/en\/key-driver-analysis-what-truly-drives-satisfaction-and-nps-and-what-merely-correlates-with-them\/","title":{"rendered":"Key driver analysis: what truly drives satisfaction and NPS, and what merely correlates with them"},"content":{"rendered":"<p>Your customers complain about dozens of things at once &#8211; wait times, prices, the interface, the tone of communication, and the availability of consultants. The problem is that there is only one improvement budget, while the list of demands is endless. <strong>Key driver analysis<\/strong> makes it possible to separate what is most strongly associated with ratings and recommendations from what merely co-occurs with them &#8211; and thus direct efforts where they are most likely to move the metric.<\/p>\n<h2>What distinguishes a satisfaction driver from a simple correlation?<\/h2>\n<p>When a customer gives a low rating while also complaining about five different things, intuition suggests that all five need to be fixed. This is a mistaken assumption. Some of these complaints may be an effect rather than a cause &#8211; a customer dissatisfied with one key aspect may transfer their negative assessment to secondary features that, in isolation, would not discourage them at all. <strong>Key driver analysis<\/strong> answers the question of which service or product features most strongly predict overall ratings and willingness to recommend after accounting for other variables, and which merely co-occur with them in simple correlations.<\/p>\n<p>This distinction has a direct impact on resource allocation. If a company responds to every stated pain point with the same level of effort, it spreads its budget across areas that may not change its NPS score or stated satisfaction. Attribute importance analysis brings order to this chaos by assigning each attribute a weight based not on how often customers talk about it, but on how strongly it predicts their final rating.<\/p>\n<p>The key distinction here is between <strong>stated importance<\/strong> and <strong>derived importance<\/strong>. Stated importance is the answer to a direct question such as, &#8220;How important is this aspect to you?&#8221; The problem is that respondents often overestimate socially expected features (such as safety and price) and underestimate emotional factors that actually drive their behavior. Derived importance is calculated statistically &#8211; based on how ratings of individual features explain the variance in the overall rating. This is the basis of reliable key driver analysis.<\/p>\n<h2>How is key driver analysis conducted step by step?<\/h2>\n<p>Key driver analysis is based on linking ratings of multiple detailed features to one outcome variable &#8211; most often overall satisfaction or willingness to recommend measured using the NPS question. The process consists of several structured stages.<\/p>\n<p>Before applying statistical methods, the right data must be collected. The following steps describe the typical course of a key driver analysis project:<\/p>\n<ul>\n<li><strong>Defining the outcome variable<\/strong> &#8211; determining whether the analysis explains overall satisfaction, NPS, or another metric, such as repurchase intention.<\/li>\n<li><strong>Building a list of detailed features<\/strong> &#8211; attributes describing the customer experience, previously identified through qualitative research, so that the list does not omit important factors or include unnecessary ones.<\/li>\n<li><strong>Measuring ratings on consistent scales<\/strong> &#8211; collecting ratings for each feature and the outcome variable in one questionnaire, using comparable scales.<\/li>\n<li><strong>Calculating derived importance<\/strong> &#8211; applying a statistical model that assigns each feature a strength of association with the outcome variable.<\/li>\n<li><strong>Plotting the results on a map<\/strong> &#8211; comparing the importance of each feature with its current rating, which organizes areas according to their potential.<\/li>\n<\/ul>\n<p>The most commonly used tool is <strong>regression analysis in research<\/strong> &#8211; either linear or logistic regression. A regression model estimates how much the overall rating changes when the rating of a given feature increases by one unit, while holding the other features constant. This &#8220;all else being equal&#8221; clause helps limit the confusion of simple co-occurrence with the independent contribution of a predictor, although it does not in itself prove causality. When features are strongly correlated with one another, standard regression can be unstable, so methods robust to multicollinearity are used, such as relative importance analysis based on variance decomposition, which distributes the contribution of individual predictors more fairly.<\/p>\n<p>The result is usually presented as a <strong>priority map<\/strong> &#8211; a two-dimensional layout in which one axis represents the derived importance of a feature and the other its current rating. Features with high importance and low ratings form the area with the greatest potential for improvement. Important features that are rated well should be maintained. Features of low importance, regardless of their rating, do not deserve priority &#8211; and this is usually the most valuable conclusion for a manager working with a limited budget.<\/p>\n<p>As Hume&#8217;s Institute experts point out, one pattern recurs in research practice: customers complain about many things at once, but only a few of them are usually most strongly associated with their final rating &#8211; the rest may be noise that consumes the improvement budget without producing a visible return in the metric. This observation changes how results should be read: the number of mentions of a given problem is not a measure of its impact.<\/p>\n<p>In Hume&#8217;s Institute projects, the factors driving NPS are often surprisingly different from those that customers themselves identify as most important. An aspect rarely mentioned in open-ended comments can have the highest derived importance &#8211; and conversely, a loudly voiced complaint may merely be a rationalization of dissatisfaction originating elsewhere.<\/p>\n<h2>What mistakes most often distort attribute importance analysis?<\/h2>\n<p>Key driver analysis is a useful method, but it is sensitive to how the study is designed and interpreted. Several pitfalls recur particularly often, and it is worth knowing them before the results reach a decision-making meeting.<\/p>\n<p>Below are the most common errors that undermine the credibility of the results:<\/p>\n<ol>\n<li><strong>Confusing correlation with impact<\/strong> &#8211; two features may increase together because both depend on a third, unaccounted-for factor. Without a model that controls for other variables, it is easy to assign independent importance to an attribute that merely co-occurs with the outcome.<\/li>\n<li><strong>Multicollinearity among predictors<\/strong> &#8211; when features are strongly related to one another, standard regression distributes their contribution unstably. Ignoring this phenomenon leads to weights that could look entirely different if the study were repeated.<\/li>\n<li><strong>Omitting an important feature<\/strong> &#8211; if the questionnaire lacks a factor that truly drives the rating, its relationship with the outcome may &#8220;spill over&#8221; into other variables, artificially inflating their importance. This is why the list of attributes should be based on an earlier qualitative stage.<\/li>\n<li><strong>Treating the relationship as linear<\/strong> &#8211; some features have threshold effects. Below a certain level, they cause strong dissatisfaction, while above it, further improvement no longer raises the rating. A linear model will not capture such effects without additional analysis.<\/li>\n<li><strong>Interpreting weights as a prescription<\/strong> &#8211; high feature importance indicates that it is worth addressing, but it does not say how or at what cost. Importance is an input for an operational discussion, not a ready-made decision.<\/li>\n<\/ol>\n<p>It is also worth remembering the method&#8217;s limitations. Key driver analysis indicates which factors are most strongly associated with NPS and satisfaction at a given point in time and in a given sample. It is not an immutable law &#8211; the importance structure changes along with the market, competitors&#8217; offers, and customer expectations. This is why the analysis should be repeated and treated as part of monitoring rather than as a one-time measurement.<\/p>\n<p>Machine learning methods, such as random forests, can serve as an alternative and complement to regression, as they handle nonlinearities and interactions between features better than a linear model. Their disadvantage is lower interpretive transparency &#8211; it is more difficult to state directly that &#8220;improving a feature by one unit increases the rating by this much.&#8221; In research practice, both approaches are often used in parallel to compare the stability of the findings. Regression remains the starting point because of its clarity, while more complex methods serve as a validation tool.<\/p>\n<h2>When is it worth using key driver analysis?<\/h2>\n<p>Not every research project requires key driver analysis. The method delivers the greatest value in specific decision-making situations where operational priorities need to be organized based on data rather than on the volume of complaints.<\/p>\n<p>Key driver analysis is particularly useful in the following cases:<\/p>\n<ul>\n<li>when the list of possible improvements is long, while the budget and team are limited;<\/li>\n<li>when the NPS score is declining or stagnating and the cause is not obvious;<\/li>\n<li>when different departments advocate for the priority of their own areas and a shared, measurable point of reference is needed;<\/li>\n<li>when a company wants to verify whether customers&#8217; stated priorities align with what actually drives their ratings;<\/li>\n<li>when satisfaction is measured on a recurring basis and an explanatory layer is needed beyond the metric itself.<\/li>\n<\/ul>\n<p>What these situations have in common is tension between many possible actions and limited resources. Where that tension does not exist &#8211; because the priority is obvious or the list of actions is short &#8211; attribute importance analysis may be excessive. Its role is to provide an objective, statistical basis for discussing priorities, not to replace that discussion.<\/p>\n<h2>Frequently asked questions<\/h2>\n<h3>What is key driver analysis?<\/h3>\n<p>Key driver analysis is an analytical method that identifies which product or service features are most strongly associated with overall satisfaction or willingness to recommend. It links ratings of multiple detailed attributes to a single outcome variable and assigns each one a statistically derived weight. This helps distinguish factors that may drive ratings from those that merely co-occur with them.<\/p>\n<h3>How can a driver be distinguished from a coincidental correlation?<\/h3>\n<p>The key is a model that estimates the relationship between each feature and the outcome while holding the others constant &#8211; this is what regression and related methods do. Correlation alone shows that two quantities change together, but it does not identify the cause; their relationship may result from a third, hidden factor. Additional safeguards include controlling for multicollinearity and comparing the results of several methods to determine whether the findings are stable.<\/p>\n<h3>Which factors should be improved first?<\/h3>\n<p>Priority should be given to features that combine high derived importance with low current ratings &#8211; these are the areas with the greatest potential to improve the metric. Important features that are rated well should be maintained, while features of low importance should be deprioritized regardless of their ratings. The final order should also take into account the cost and feasibility of change, which the analysis itself does not determine.<\/p>\n<p>Want to know which aspects truly drive your customers&#8217; ratings and which merely consume budget? <strong>Ask Hume&#8217;s Institute about an analysis of the factors driving your customers&#8217; satisfaction<\/strong> &#8211; <a href=\"https:\/\/humes.pl\/en\/contact\/\">get in touch<\/a> to discuss a research scope tailored to your metrics.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Your customers complain about dozens of things at once &#8211; wait times, prices, the interface, the tone of communication, and the availability of consultants. The problem is that there is only one improvement budget, while the list of demands is endless. Key driver analysis makes it possible to separate what is most strongly associated with [&hellip;]<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":""},"categories":[912],"tags":[],"slowa_kluczowe":[],"class_list":["post-3593","post","type-post","status-publish","format-standard","hentry","category-badania-i-analizy"],"acf":[],"_wp_attached_file":null,"_wp_attachment_metadata":null,"wpml_media_processed":null,"_wpml_media_usage_in_posts":null,"_wp_attachment_context":null,"_oembed_35c905c64c03156f243b94f18c4eb80f":null,"_wp_attachment_image_alt":null,"rank_math_description":"Key driver analysis separates true drivers of satisfaction and NPS from simple correlation, contrasting stated importance with derived importance.","rank_math_focus_keyword":"key driver analysis","rank_math_contentai_score":null,"_wpml_post_translation_editor_native":null,"_menu_item_type":null,"_menu_item_menu_item_parent":null,"_menu_item_object_id":null,"_menu_item_object":null,"_menu_item_target":null,"_menu_item_classes":null,"_menu_item_xfn":null,"_menu_item_url":null,"_wp_page_template":null,"rank_math_og_content_image":null,"_wp_trash_meta_status":null,"_wp_trash_meta_time":null,"_wp_desired_post_slug":null,"rank_math_primary_category":null,"_acf_changed":null,"wp_pattern_sync_status":null,"_form":null,"_mail":null,"_mail_2":null,"_messages":null,"_additional_settings":null,"_locale":null,"_hash":null,"_config_validation":null,"_wp_old_slug":null,"rank_math_internal_links_processed":"1","_top_nav_excluded":null,"_cms_nav_minihome":null,"_thumbnail_id":null,"_last_translation_edit_mode":null,"_wpml_word_count":"1913","_dp_original":null,"_edit_last":null,"_edit_lock":null,"rank_math_seo_score":null,"_wpml_location_migration_done":null,"_wpml_media_duplicate":null,"_wpml_media_featured":null,"_wp_old_date":"2026-08-25","copied_media_ids":[],"referenced_media_ids":[],"rank_math_title":"Key driver analysis: stated vs derived | Hume's Institute","job_department":null,"_job_department":null,"job_location":null,"_job_location":null,"job_offer_external_link":null,"_job_offer_external_link":null,"footnotes":null,"inline_featured_image":null,"blog_podtytul":null,"_blog_podtytul":null,"blog_czas_czytania":null,"_blog_czas_czytania":null,"blog_dalsza_lektura":null,"_blog_dalsza_lektura":null,"slownik_krotka_definicja":null,"_slownik_krotka_definicja":null,"slownik_cytat":null,"_slownik_cytat":null,"slownik_na_stronie_glownej":"1","_slownik_na_stronie_glownej":null,"slownik_slowa_kluczowe":null,"_slownik_slowa_kluczowe":null,"slownik_w_praktyce":null,"_slownik_w_praktyce":null,"slownik_powiazane":null,"_slownik_powiazane":null,"slownik_kluczowe_punkty":null,"_slownik_kluczowe_punkty":null,"lang":"en","translations":{"en":3593},"pll_sync_post":{},"_links":{"self":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts\/3593","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/comments?post=3593"}],"version-history":[{"count":1,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts\/3593\/revisions"}],"predecessor-version":[{"id":3594,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts\/3593\/revisions\/3594"}],"wp:attachment":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/media?parent=3593"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/categories?post=3593"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/tags?post=3593"},{"taxonomy":"slowa_kluczowe","embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/slowa_kluczowe?post=3593"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}