Loyalty Ladder demo: Retail Fuel Loyalty Study

Demonstration built on a Latenta retail-fuel loyalty study (MENA, fielded late 2023 - early 2024). Mixed-methods design: 8 focus groups (×6) followed by a CAWI behavioural survey - 12,224 surveyed, 2,350 completes, 2,030 final clean sample (effective ≈1,988, 98%); IR 16.6%, LOI 35 min; weighted to 70% men / 30% women; significance by False Discovery Rate (FDR, p=0.05). Loyalty, cross-selling and frequency modelled with XGBoost classifiers; SOW (share of wallet) by regression; interpreted with SHAP. All figures are real study results; brand identifiers are disguised and money is shown in USD equivalents.
Slice widths:
The through-line: climbing the rungs, patronage, fuel & non-fuel spend, share of wallet, satisfaction, Loyalty app usage and openness to cross-sell all rise, while demand for improvements and orientation to short-term gain fall. The strategic message: a personalised, tiered loyalty programme that graduates each rung upward - tactical goal “+1 visit per month per customer.”
The Loyalty Ladder - the theory behind this view

What it is. The loyalty ladder is a relationship-marketing model popularised in the early 1990s (Christopher, Payne & Ballantyne’s Relationship Marketing, building on Raphel’s ladder). Its premise: a customer base is not one mass but a sequence of rungs of increasing attachment, and marketing’s job is to move people up one rung at a time.

The rungs. Prospect - aware but has not yet bought. Customer - has bought once or occasionally; no bond. Client - buys repeatedly but transactionally, often price-led and open to competitors. Supporter - likes the brand and buys habitually, yet stays passive. Advocate - actively recommends the brand and defends it; some versions add Partner above. Each transition has its own mechanism - trial, habit, trust, identity, voice - and asking a rung for behaviour two rungs up (e.g. asking a Customer to advocate) typically backfires.

Why climbing matters. Retention is cheaper than acquisition, spend and share of wallet rise with attachment, and Advocates recruit new customers at zero media cost - so the ladder shifts budget from perpetual acquisition to graduation mechanics: onboarding, habit loops, loyalty programmes, recognition and referral schemes.

How this study uses it. An XGBoost classifier assigns each of the 2,030 respondents a rung from surveyed behaviour (frequency, spend, Loyalty app engagement, attitudes). The theory’s prediction holds in the data: patronage, spend, share of wallet, satisfaction and cross-sell openness all rise with the rung, while improvement demands and short-termism fall. The engagement’s tactical translation is one number: +1 visit per month per customer. Prospects were excluded by design and are recommended as the next study.

Dashed grey outline = NET, the total sample (all 2,030 respondents) on the same axes - where the rung bulges past it, it over-indexes; inside it, it under-indexesAxes normalised to the strongest rung on each metric · hover an axis label for the source question
Share of wallet: out of each $100 spent on fuel, how much goes to PetroX (and the non-fuel equivalent). Below, the leakage picture: the share of each rung that also fuelled at a competitor in the last three months. The RFM logic is visible throughout - SOW rises with frequency and spend.
Which promotional hooks move which rung, and where non-fuel demand concentrates. Positive cross-sell drivers (SHAP): satisfaction, long-distance travel, non-fuel spend and loyalty itself.
Colour shows the relation to the total sample (NET): blue = above NET, orange = below NET; an arrow marks a difference of ≥ 8 points vs NET (FDR p=0.05 applied in the technical report). Hover a row label for the source question.
Reading the dots: the dashed hollow circle is the same attribute for the total sample (NET); the solid dot is the selected rung, and the thin line shows how far the rung deviates from NET. x-axis: improvement demand (Q25, “dramatic + considerable improvement needed”) · y-axis: satisfaction (Q42, top-2-box). Six analyst-matched attribute pairs; “fix first” = high demand, low satisfaction. Clients and Customers light up the fix-first quadrant. Base n = 2,030 · effective ≈1,988 (98%) · FDR p=0.05.
Compare: vs
An illustrative deterministic what-if on real inputs - measured rung sizes, visit frequencies, spend levels and shares of wallet - not a forecast. Revenue is shown per 1,000 PetroX customers per month; fuel spend per visit uses analyst point estimates within each rung's measured band.
Baseline mix: Advocate 23.7% · Supporter 50.6% · Client 22.6% · Customer 3.1% (n = 2,030, weighted). Fuel spend/visit points: $25 / $15 / $10 / $15 within measured bands; non-fuel ticket from Q40 means ($39.2 / $35.9 / $29.8 / $39.7). Customer figures rest on n = 62 - treat with caution.
Models trained on the full clean sample - base n = 2,030 · effective ≈ 1,988 (98%) · significance by FDR p = 0.05. Scores shown honestly against naive baselines; Customer rung (n = 62) carries limited signal.
The consumer knowledge graph: how the study hangs together. Every line is a measured or modelled relationship carrying a number - nothing decorative. Click any node to isolate its evidence; click again (or tap the canvas) to release. The side panel links straight to the view where each relationship lives.
Show: Link strength: all
Edge weights: SHAP relative importance (per model, top driver = 100), measured deltas between adjacent rungs, and model scores vs baselines. Base n = 2,030 · effective ≈1,988 (98%) · FDR p = 0.05 · Customer figures directional (n = 62).
Answer five quick questions to see where you would sit on the ladder. A playful heuristic mirror of the real classifier's top SHAP drivers - not the production model.
Drive a customer up the ladder. Each station is a rung - to pull away you must name the mechanism that graduates it and back it with the evidence. Right answers refuel the tank; wrong ones cost fuel but show the real finding. Everything on this road is measured in this study.
The study in seven slides - use Next or the arrow keys, or skip straight to the dashboard. Every figure is a real study result; money figures are modelled USD equivalents (not forecasts).
Every insight in the deliverable, ranked by weight. Filter by act or rung; each entry cites its evidence (hover the chips for source questions).