Two Questions a Spend Ranking Can't Answer

Take two customers of the same luxury house.

The first spent €40,000 in her first three months, over four visits. She has not replied to a message, and nobody in the store could tell you what she thinks of the brand.

The second has spent about €3,000 a year for eleven years. She buys from a photo when her client advisor suggests something, and she kept buying through two price increases. Twice she has brought her sister in.

Sort the list by twelve-month spend and the first woman is near the top. The second is far below the line for an event invitation.

Bain and Altagamma estimate that customers who spend more than €20,000 a year made up a little over 46% of personal luxury goods spending in 2025, up from 30% in 2019.¹1Bain & Company and Altagamma, Finding a New Longevity for Luxury, December 2025. With that much of the money at the top, ranking by money feels like enough.

BCG and Altagamma have estimated what such a ranking misses. In their 2024 study, most brands pick out their very important customers, VICs, with a single variable: spend above a threshold over 12 to 18 months. Those brands, they estimate, might miss 70% of the potential VICs.²2BCG and Altagamma, True-Luxury Global Consumer Insights, 10th edition, July 2024.

You can do better with the same receipts. A model fitted on how often and how recently each customer buys, the kind an earlier piece here is about, will notice that the second woman always comes back.³3Wei, River of Time, TP Infinity Lab Journal. It still works only from purchase records, and those do not show whether she trusts the house. So ask three questions about each customer and keep the answers apart.

Spend, stage, closeness

Spend: what she has spent. On record already, usually as tiers.

Stage: where she is in the life of the relationship. New, settled, or drifting. Read it from her purchase record. Years on the books will not give it to you: an eleven-year customer can be drifting.

Closeness: how far she trusts the house. Would she take its word on something she cannot check? Would she stay through a bad season? That is a working definition. Expect the true answer for most customers, at first, to be "not known yet."

Three bands on each axis make a cube of 27 cells, with the not-known-yet customers waiting outside it. The cube is the core of what I call the Galaxy Relationship Model. The first woman sits in high, new and, on thin evidence, distant. The second sits in low, settled, close.

Which of them is the better customer depends on what you want a customer for, and the cube makes you say so first. An evening to reward the clients closest to the house: invite the second woman. An evening to start a relationship with a big new spender: invite the first.

Where the parts come from

Almost none of the parts are new. The grid is the oldest. Direct marketers have sorted customers by recency, frequency and money for decades.⁴4Blattberg, Kim and Neslin, Database Marketing, Springer, 2008, chapter 12. Three bands on each give the same 27 cells, and all three measures come from receipts.

Others have added a relationship or an attitude to what the receipts say. A consultancy called rpc publishes a Customer Value Cube: customer value, customer potential, customer relationship.⁵5rpc (The Retail Performance Company), white paper on the Customer Value Cube, undated. In 2010 Worthington, Russell-Bennett and Härtel proposed auditing brand loyalty on a matrix of three dimensions, emotional, cognitive and behavioral, each cut high and low.⁶6Worthington, Russell-Bennett and Härtel, Journal of Brand Management 17(4), 2010. Neither has a stage.

A patent assigned to Accenture, filed in 2002 and granted in 2010, describes crossing separate segmentations of behavior and value with others, such as brand attachment or life stage, in a matrix of three or more dimensions. Its banking example includes segments for new and entrenched customers. The user picks the other data types from a list, and the stated aim is more precise targeting of offers.⁷7U.S. Patent 7,698,163, assigned to Accenture, filed 2002, granted 2010. Gainsight, software that business-to-business companies use to manage their accounts, keeps recurring revenue, lifecycle stage and a health score as separate fields on each company record.⁸8Gainsight, Company object documentation.

I have not found spend, stage and closeness, with closeness meaning trust, named together for the individual customers of a consumer brand. The search was partial. I have read only the abstract of a 2007 paper that classifies a securities broker's customers on three dimensions, and what brands do in-house is mostly unpublished. So I claim only a choice: these three axes for luxury clients, defined this way, with the rules and the test that follow.

Two rules for reading it

Keep the three answers apart on the advisor's screen, with no total. A loyalty program may need one score to set its thresholds. The advisor needs the three answers, because "high, new, not known yet" tells her where to start. I expect advisors to do better with three answers than with one score, though nobody has tested it.

Spend is on record, but closeness is mostly inferred, so each piece of evidence about closeness should carry one of four labels: she told us, we saw it, we inferred it, we confirmed it. An inference must never go into a customer's file as a fact about her. "Distant", for the first woman, is an inference from silence, and her file should say so.

From position to action

Two more parts of the model decide whether to do anything about where a customer sits.

One is the persona: a kind of customer, grouped by how she behaves, whatever her tier. In the cube a persona is a cloud, with members in many cells. Line up what those members did by time since each one started, and you get the usual path for that kind of customer, with a band of ordinary variation around it.

The other is journey orchestration. Compare each customer with that band and with her own past. Act only on a difference that lasts.

The reason to wait is regression to the mean. People who buy often in one period buy less often in the next, on average. Jerome Greene called this "a real fact of life" in 1982.⁹9Greene, Consumer Behavior Models for Non-Statisticians: The River of Time, Praeger, 1982. In one retailer's file, published by Fader, Hardie and Ross, 24% of the customers who bought in both of two years stayed in the same profit decile.¹⁰10Fader, Hardie and Ross, The Customer-Base Audit, Wharton School Press, 2022. Three bands will hold steadier than ten, though some customers will still move between them.

The second woman has eleven years of her own history, and she buys most years. Say that for her one missed year is ordinary and two in a row is a gap. Skipped years are common: at the same retailer, of the customers acquired in the first quarter of 2016 who bought in 2018, 39% had bought nothing in 2017.¹⁰10Fader, Hardie and Ross, 2022.

For the first woman there is no history yet, so compare her with other big early spenders. If most of them do not reply in their first months either, her silence is ordinary. If most of them have asked for something by now, it is a gap.

Inside the band, carry on. Above it is a cue to go deeper. Below it for long enough, find the cause before anyone sends her an offer. If the second woman's advisor has left, the move is a repair, and the first message should ask nothing of her. A customer who never fit her band may be filed under the wrong persona.

An upgrade is one possible goal among several. The brand sets the goal. The advisor chooses how to pursue it, with more room the better she can show she knows the client. Then keep watching, and keep a randomly chosen group you leave alone, so you can tell whether the action did anything.

I do not expect orchestration to move a persona's cloud. Where a house's customers sit in the cube follows decisions made far from the sales floor, such as pricing. In the earlier piece I called the difference between one customer and the next fate, and the movement inside one customer fortune. Clienteling can work on fortune and should go along with fate, which in Chinese I call 顺命. When the cube goes on a slide and somebody adds an arrow, everyone spending more and getting closer, take it off. My own slides have had that arrow.

What each house has to test

Each house has to set its own baselines, its own threshold for "lasts", and its own rule for when two comparisons disagree. It also has to check whether the customers with gaps are the ones an action can move. In two field experiments on retention campaigns, Eva Ascarza found that the customers most likely to leave and the customers the campaign was estimated to move overlapped about as much as chance would predict.¹¹11Ascarza, Journal of Marketing Research 55(1), 2018.

The likeliest failure is that the third axis repeats the first. One of Byron Sharp's laws points that way: attitudes reflect behavioral loyalty, and people know and say more about the brands they use.¹²12Sharp, How Brands Grow, Oxford University Press, 2010. In a 2015 meta-analysis, Watson and colleagues found that behavioral loyalty carried the effect on business performance (a path coefficient of 0.27) and attitudinal loyalty added almost nothing (0.02, not significant), although measures combining the two did better than either.¹³13Watson, Beck, Henderson and Palmatier, Journal of the Academy of Marketing Science 43(6), 2015. Evidence about closeness will lean the same way, because customers who buy more leave more chats and more notes.

Suppose the axis survives. Close customers will then turn out to spend more, and that will prove little: the people who warm to a house may be the ones who were going to buy anyway. In Dutch grocery retail, Leenheer and colleagues estimated that loyalty-program membership raised share of wallet by 4.1 percentage points once they corrected for who chooses to join. Uncorrected, the figure was 29.8.¹⁴14Leenheer, van Heerde, Bijmolt and Smidts, International Journal of Research in Marketing 24(1), 2007. I have not found that correction run on luxury clients. The Watson analysis does credit attitude with word of mouth: 0.41, against 0.19 for behavior.¹³13Watson and colleagues, 2015. On that evidence, expect closeness to show up in what she says about the house to other people more than in what she buys herself.

The closeness may also belong to someone else. When the second woman buys from a photo her advisor sends, is that trust in the house, or in the advisor? Palmatier, Scheer and Steenkamp studied 362 pairs of business buyers and salespeople, and only the loyalty a buyer felt toward the salesperson directly affected sales growth.¹⁵15Palmatier, Scheer and Steenkamp, Journal of Marketing Research 44(2), 2007. In the BCG and Altagamma survey, 68% of VICs said they would follow their client advisor to another brand.²2BCG and Altagamma, 2024. A survey can ask about the two apart, as Sirdeshmukh, Singh and Sabol did in 2002 with trust in frontline staff and trust in management's policies and practices.¹⁶16Sirdeshmukh, Singh and Sabol, Journal of Marketing 66(1), 2002.

And reading service chats and advisor notes to score closeness may be a purpose the customer was never told about. In mainland China, Article 14 of the Personal Information Protection Law requires fresh consent when the purpose changes, and Article 24 on automated decision-making may apply if software then decides who gets which treatment.¹⁷17Personal Information Protection Law of the PRC, 2021, Articles 14 and 24. Ask counsel before building anything.

The first test is small. Use two years of one house's customers. Read spend and stage from the transaction file. For closeness, draw a random sample across every spend tier and ask them, with consent, then ask part of the sample the same questions again a few weeks later. Fix the cut points and the pass mark before you look.

People must give much the same answer the second time. If their answers wobble between the two rounds, the closeness questions are not yet measuring anything. Then the cells where spend and closeness disagree must hold more customers than the wobble alone would put there. If they hold no more, the third axis is redundant and the cube is really a grid of spend by stage.

Customers like the first woman, who answer nothing, form a group of their own. Report the result once with all of them counted as distant and once with all of them counted as close, so that their silence cannot decide it.

Until a house has that count for its own customers, every chart of its cube keeps the footnote: illustrative, pending validation.

Daniel Wei is COO, TP Infinity, the consumer engagement practice within TP. The Galaxy Relationship Model is part of the practice's CRM and clienteling work across Asia Pacific and beyond.

The two customers are invented.

Sources

1. Bain & Company and Altagamma, Finding a New Longevity for Luxury, December 2025.
2. BCG and Altagamma, True-Luxury Global Consumer Insights, 10th edition, July 2024.
3. Wei, River of Time, TP Infinity Lab Journal.
4. Blattberg, Kim and Neslin, Database Marketing: Analyzing and Managing Customers, Springer, 2008, chapter 12.
5. rpc (The Retail Performance Company), "What Is the Value of Customer Relations?" White paper on the Customer Value Cube, undated.
6. Worthington, Russell-Bennett and Härtel, "A tri-dimensional approach for auditing brand loyalty," Journal of Brand Management 17(4), 2010.
7. Reed, Hariharan and Saito, U.S. Patent 7,698,163, "Multi-dimensional segmentation for use in a customer interaction," assigned to Accenture, filed 2002, granted 2010.
8. Gainsight, Company object documentation.
9. Greene, Consumer Behavior Models for Non-Statisticians: The River of Time, Praeger, 1982.
10. Fader, Hardie and Ross, The Customer-Base Audit, Wharton School Press, 2022.
11. Ascarza, "Retention Futility: Targeting High-Risk Customers Might Be Ineffective," Journal of Marketing Research 55(1), 2018.
12. Sharp, How Brands Grow, Oxford University Press, 2010.
13. Watson, Beck, Henderson and Palmatier, "Building, Measuring, and Profiting from Customer Loyalty," Journal of the Academy of Marketing Science 43(6), 2015.
14. Leenheer, van Heerde, Bijmolt and Smidts, "Do Loyalty Programs Really Enhance Behavioral Loyalty? An Empirical Analysis Accounting for Self-Selecting Members," International Journal of Research in Marketing 24(1), 2007. Figures read from the ERIM working-paper version.
15. Palmatier, Scheer and Steenkamp, "Customer Loyalty to Whom? Managing the Benefits and Risks of Salesperson-Owned Loyalty," Journal of Marketing Research 44(2), 2007.
16. Sirdeshmukh, Singh and Sabol, "Consumer Trust, Value, and Loyalty in Relational Exchanges," Journal of Marketing 66(1), 2002.
17. Personal Information Protection Law of the People's Republic of China, 2021, Articles 14 and 24. Official text, in Chinese.

← Back to The Journal