Your NPS file in, a driver report out
Upload the score and the verbatims. cxCatalyst codes the comments, runs Catalyst and sends back the quadrant, the driver table and a confidence verdict on what the data can support.
Catalyst · driver analysis
Customers tell you what matters. Catalyst shows what moves the score.
Catalyst reads your survey data and separates two things most driver analysis blurs: how strongly each attribute is linked to your outcome, and how much room there is to move the outcome by improving it. You get a short list of where to invest, where to hold, and where to stop spending.
Personas built on what drives each person
Catalyst reads the drivers for every respondent, groups people whose drivers look alike, then reads importance and impact again inside each persona.
Any outcome worth predicting
Satisfaction, main-bank status, retention, tenure, purchase intent or a composite index. If the survey measures the outcome and the candidate drivers, Catalyst can read it.
Worked example · Wrenfield Bank
They say fees. They stay for control.
Ask customers what would make a bank their main bank and they put fees, interest rates and the app at the top. Read the same customers through Catalyst, linking each attribute to how strong their main-bank relationship actually is, and the order turns over.
Twelve attributes, ranked two waysMeasured Modelled
Customers' top threeCatalyst's top threeThe rest
Fees and interest rates come out 10th and 12th of twelve on impact. Feeling in control of my money, guidance and being recognised as a customer, all near the bottom of what customers say, lead the model. Across all twelve the two orders run in opposite directions (rank correlation −0.73).
Illustrative study. Simulated customers for Wrenfield Bank, a fictional bank, n = 2,500. What customers say matters is measured in the study (importance, 1 to 7); the impact ranking comes from Catalyst and is labelled Modelled.
Importance and impact
Two numbers, four decisions
Importance says how strongly an attribute is linked to the outcome. Impact says how much the outcome moves when the attribute improves. An attribute can be important and already saturated, or modest on importance with plenty of room to move. Plotting the two against each other turns a ranked list into decisions. Pick a group to see how the picture changes.
Across all customers, the largest impact comes from feeling in control of my money. 4 drivers sit in priority investment. The drivers explain 58% of the variation in the relationship score.
Priority investment
Linked to the outcome, with room to move it. Act here first.
1. Feeling in control of my money
2. Guidance and nudges
5. A person when it matters
6. App and online easeProtect and maintain
Linked to the outcome, little room to move. Hold the line.
7. Flexibility when money is tight
9. Speed of fixing problems
10. Fees and valueInvestigate before investing
Room to move, weaker link. Pilot before scaling.
3. Being recognised as a customer
4. Responsible bankingLow priority
Weak link, little room. Reallocate.
8. Security and fraud protection
11. Rewards and perks
12. Interest rates
Illustrative study. Same simulated bank. Both scores run 0 to 100 within each model, so compare positions within a group rather than across groups. The quadrants are split at the midpoint for this illustration. Labelled Modelled.
What Catalyst reports
Three layers for every driver
Each driver gets a score on three separate questions. Keeping them apart is what stops a budget going to the attribute that is already as good as it is going to get.
How strongly is it linked to the outcome?
If this driver were scrambled, how much worse would the model predict the outcome?
Measured with a machine-learning model that picks up non-linear patterns a correlation misses. Scored 0 to 100.
How much does the outcome move if it improves?
If this driver got better and nothing else changed, how far would the outcome shift?
Read from a separate model that steps each driver up while holding the others steady. Steep means room to move; flat means diminishing returns. Scored 0 to 100.
Where do you stand on it today?
Is this already a strength, or a gap?
Current performance on each driver, so a high-impact gap reads as an opportunity and a high-importance strength reads as something to protect.
Two further reads
Direction shows whether improving a driver lifts the score or whether it mainly hurts when it slips. Incidence, for coded comments, shows how many people raised it, so a vivid but rare complaint doesn't crowd out a common one.
Two patterns performance reveals
Hidden gems look modest on importance but carry high impact because performance is low today. Saturated drivers matter, but you already perform well, so more spend will not move the score.
See it explained
Why importance and impact are not the same number
Five minutes on the two scores, the four quadrants, and what separating them changes about where a budget goes.
Checked on real people
Groups built on drivers beat groups built on demographics
We tested the approach in public, on a dataset of 2,058 Americans released by Columbia Business School. Catalyst read the drivers for each person, people were grouped on those drivers, and the groups were tested on respondents the model had never seen.
3×Driver-based personas separated unseen people about three times better than segments built from age, income and region (20.3% against 6.5% of the difference between people).
102%Two weeks later, predicting a fresh set of answers from the same people, the personas worked as well as on day one.
HalfDemographics were more predictive for an individual, yet produced half the separation once used to group people. The gain is in how the information is organised.
Public data, code that runs
The Broken Ruler study
The method was written down before each test, then run on data anyone can download. Read the full study →
Used on a named study
Intrum European Consumer Payment Report
20,000 consumers across 20 European markets, grouped into four Money Manager personas from their individual driver profiles. Read the case study →
What Catalyst does not claim
Catalyst ranks where improvement is most likely to move your outcome, given how the drivers and the outcome relate in your data. It does not prove that changing a driver will cause the change, so the biggest moves are worth testing in market before they are scaled. The banking example on this page is simulated; the Broken Ruler figures come from real respondents.
How it works
From your data to a short list of decisions
Catalyst runs on respondent-level data you already hold, or on a study designed around the outcome you care about.
-
Step one
Agree the outcome and the drivers
The outcome can be a single score such as NPS or a composite index. The drivers are the attributes, experiences or coded comments you think move it.
-
Step two
Run the models
Catalyst scores importance, impact and performance for every driver, for the total sample and for any group you need, with the base shown on every figure.
-
Step three
Turn it into decisions
Every driver lands in a quadrant with a plain action: invest, protect, investigate or deprioritise, plus the gems and saturated drivers performance reveals.
What lands on your desk
A driver map you can take straight to a budget discussion
- The importance-by-impact quadrant for the total and each group you asked for
- A driver table with importance, impact, performance, direction and, for comments, incidence
- A plain recommendation per driver, and a note where the data cannot support one
- Respondent-level driver profiles, ready for personas if you want to go further
What is the Catalyst framework?
Knowsis driver analysis. For any outcome a survey measures, such as NPS, satisfaction or main-bank status, Catalyst scores every candidate driver on importance (how strongly it is linked to the outcome), impact (how much the outcome moves when it improves) and performance (where you stand today), then places each driver in a quadrant with a plain action.
Why are importance and impact different numbers?
Because an attribute can be strongly linked to the outcome and still offer little room to move it, usually because you already perform well on it. Most driver analysis reports one number and sends budget to attributes that are already saturated. Catalyst measures the two separately, with two different models.
Why not just ask customers what matters?
Stated importance and what actually moves the outcome often disagree. People rate the things that are easy to name, such as price, highly. In the simulated banking example on this page the two orders run in opposite directions. Asking still has a place: it shows what customers notice, which is useful for messaging.
What data does Catalyst need?
Respondent-level data with an outcome measure and the attributes you think drive it. Rating batteries, yes-or-no items and coded open-ended comments all work. We show the base on every figure and say when a group is too small to read.
Does Catalyst prove cause and effect?
No. It ranks where improvement is most likely to move your outcome, given how the drivers and the outcome relate in your data. The biggest moves are worth testing in market before they are scaled.
How does Catalyst relate to cxCatalyst and the Persona Engine?
cxCatalyst is the self-serve version for NPS and verbatims: upload a file, get the driver report. The Persona Engine uses Catalyst at respondent level to group people by what drives them, then runs Catalyst again inside each persona.
Next step
Find out which drivers would actually move your score
Bring a recent study or the outcome you are trying to shift. In 30 minutes we'll show what Catalyst would read from it, which groups are worth splitting out, and what the driver map would look like.
The booking calendar is provided by Cal.com. Opening it loads content from Cal.com and may set third-party cookies.
30 minutes · Video call · With Greg Streatfield, founder of Knowsis
Prefer NPS and comments, self-serve? Try cxCatalyst · Or send us a message