The CFO asked what the CX program returned
Her CFO asked what the CX program returned. She had a score that went up, and she knew that was not an answer. So she counted whether the customers who had a better care call went on to renew, and by how much more often.
This article is about CX ROI: what a CX lead can defend when her CFO asks what the program returned. The scene below is illustrative, set in the 2026 budget season, but the question in it lands on most people who run a CX program sooner or later.
The question
A CX lead at a mobile operator gets a short message from her CFO. It is budget season. He wants to know what the program returned last year.
She has something ready, in a sense. Touchpoint scores are up across most of the journey. Care contact is still the weak stage, but even that moved a few points.
It is the wrong deck for his question. A score that went up says customers had a better time. He asked what the company got for that. She knows the difference before she opens the deck.
The question that has an answer
She looks for a direct line from the program to money. There is none.
What is the next best thing she can give him? Nobody can price an experience, but what people did after one can be counted.
So she trades. She sets down the question she was asked and picks up one her own data can settle. The customers who rated a touchpoint favorably: did they come back and do the next thing more often than the customers who rated it unfavorably? And if so, how much more often?
She knows what she is giving up. He asked for money and she is bringing him behavior. She thinks it is the better trade, because whatever comes back will be true of her own customers rather than of an industry average, and he will be able to check the counts behind it.
She starts with care contact, because that is the stage everyone argues about.
One touchpoint first
- Take every customer who called care last year and answered the survey afterwards.
- Split them by how they rated the call. The favorable end of the scale is one group. The unfavorable end is the other.
- Open a window after the call. She set it at 90 days for care contact.
- Inside that window, watch for one thing: did this customer renew.
The middle is left out on purpose, and she makes a note of that. It makes the gap wider than a comparison of the favorable group against everyone else would be.
The number
Her figures, illustrative here, come out like this. 24 percent of the customers in the favorable group renewed inside the window. 3 percent of the unfavorable group did.
24 divided by 3 is eight. That ratio is what Numr calls the multiple. Customers in the favorable group were eight times more likely to renew than customers in the unfavorable group.
Nothing was forecast. It is two shares and a division.
Across the whole journey
Then she runs the same arithmetic on all six stages of the mobile journey. In Numr CXM the view is called Journey Outcome, and it puts the six multiples side by side.
The figure bands them. Two stages sit in the high band, one in the mid band, and three in the low band. Care contact is the biggest number on the figure, which is why it is drawn in red. It also carries the lowest touchpoint score of the six.
The stage where the experience is worst is also where it is most tied to what customers do next.
Two other numbers sit on the widget face. The touchpoint score is whatever the main question at that stage scores, worked out the normal way for that metric. A recommendation question comes through as an NPS score. A satisfaction question may come through as an average rating. Numr does not compute anything on top of it. The 7,800 respondents is the total across all six touchpoints, not a count per stage.
Three things she had to decide
The platform did the counting. Three of the inputs were hers to set.
The outcome. For care contact she chose renewal. At another stage the natural outcome might be:
- a purchase
- a repeat order
- a return for service
There is no standard list. Somebody at the operator had to decide what the business counts as the outcome at each touchpoint, and that took a meeting with the commercial team and a second one with billing.
The window. She set it per touchpoint. The sensible wait after a care call is a different length from the sensible wait after a device delivery, so one window for the whole journey would have been wrong at most stages.
The two groups. These were hers as well. Any question works, so long as it sorts people into a favorable end and an unfavorable end and is asked consistently at that touchpoint. Any of these can produce the scale:
- a recommendation question, which gives the familiar promoter and detractor split
- a Net Easy question, which does the same job
- a satisfaction question, likewise
Where she drew the line between the two ends was her decision too.
One Numr client configured an outcome they called recommendation. The survey asked whether the customer would refer somebody, and the platform counted whether that person actually came in. A converted referral was their outcome, and the multiple ran on that. That is one client's setup, not a rule. There is no general definition of a recommendation outcome.
What IT had to build
It turned out to be nothing.
Every event that generates a survey is already in the platform, because sending the event is what triggers the survey. The care call that asked the customer how it went is on the same feed that shows, months later, whether that customer renewed. The program runs continuously, so the feed is still arriving. Journey Outcome reads further along the feed the operator was already sending.
So there was:
- no new data feed
- no export
- no integration project
The view was switched on.
One condition sits under that. The view reads the transactional record of the later event alongside the survey responses. Where a client's setup holds responses only, there is nothing to count.
What she has to tell him straight
Strictly, it is not ROI. An investment in the accounting sense has a dollar figure on both sides, and neither is in her number. What she has is the behavioral half of CX ROI. The value of a renewal is a figure finance already holds, and the cost of the program is on her own budget line. She will put the multiple in front of him and let his team do that arithmetic, because doing it herself would mean assuming his margins.
She does not have to police the edges of it. Where a group has nobody in it there is nothing to divide, so the platform hides that touchpoint. Nothing is estimated or filled in.
What the number does not tell her
The multiple says care contact carries weight. What is wrong with the calls is a different question.
A separate driver analysis answers that one. That analysis does use regression, a method that estimates which things move with the score. For any touchpoint it sets out:
- the key drivers of the experience
- how the operator performs on each
- how many people each rests on
Why the usual CX ROI answer does not hold up
There is a more usual way to answer a CX ROI question, and she went over it the night before. Model it. Take a retention lift from an industry study, multiply by the customer base, then by average revenue per user.
That produces a dollar figure. It is exactly what he asked for, which is why people build one.
Every input is an assumption. There is nothing dishonest in that. Finance builds models like that all the time.
But the lift comes from somebody else's customers. So the CX ROI it produces describes customers in general, and he asked about hers.
What settled it is what he could do with it. His team can check a count. With an assumption, the most anyone can do is argue. She wanted to bring him a number he could take apart, and a model is not built for that.
The meeting
She sends him the care contact multiple before the deck, with the two counts behind it.
In the room he asks one question. How does she know the customers in the favorable group were not already the loyal ones?
She cannot rule it out. She says so. The two groups may differ in tenure, plan, device, and how many problems they had before this one. That is why the slide says "more likely to" and not "caused". The multiple is a measured difference in behavior between two groups. It is not a controlled test.
He can check the two counts himself.
He takes the point. Then he asks what it would take to fix care contact. That is a different analysis, she tells him. The driver analysis uses regression, and it is on another slide.
Frequently asked questions
Is this a regression model?
No. It is one rate over another. Nothing is fitted. The regression lives in the driver analysis, a different view that answers a different question: what shaped the experience at a touchpoint, not what followed it.
Does it prove the experience caused the renewal?
No, and the slide in the story does not say so. Two reasons:
- Nobody assigned customers to the two groups at random.
- They sorted themselves by how they rated the call, and they may differ in other ways too.
What the multiple shows is how much more often one group took the next step than the other, inside the window that was set. That is the whole claim.
