A 32-hypothesis exploration of Falcon Consumer's warehouse surfaces three tensions the current dashboards don't expose:
Below are the ten most consequential findings, each with the evidence and implication spelled out. Full methodology and hypothesis log is in the exploration journal; charts are in the evidence dashboard.
The simplest story you can tell about any loyalty program is that members spend more than non-members. In Falcon Consumer's book, that story is true for LuxeStyle Online (+3.4%), LuxeStyle Mobile App (+3.5%), and LuxeStyle Outlet (+11.9%). In the other four BUs — Urban Thread, Bijou Accessories, Maison Luxe, and SoleStep Footwear — loyalty members have lower lifetime value than non-members. SoleStep is the worst case: loyalty LTV is 12.5% below non-member LTV.
The pattern fits a specific interpretation: loyalty is selecting for price-sensitive shoppers. They opt in for discounts, and then their lifetime spend settles at a lower plateau than comparable non-members who weren't discount-shopping in the first place. This is most vivid at Outlet (where the entire brand is price-sensitive, and loyalty compounds the effect positively) versus SoleStep and Maison Luxe (where loyalty appears to pull down full-price baseline spend).
The program is being run as if it has uniform ROI. It doesn't. Two interventions to test: (1) kill or restructure loyalty for SoleStep and Maison Luxe specifically; (2) redesign benefits so they don't train customers to wait for discounts — tier or badge-based instead of discount-based, for the four net-negative BUs.
Of the 1,200 active customers in the book, 543 first transacted in 2020, 627 in 2021, and only 30 combined across 2022, 2023, and 2024. The cliff is dramatic and it isn't explained by data latency — the warehouse has full 2024 transaction data, and active transaction volumes grew from $23.5M to $27.1M across those years.
What's also interesting: the 2020 cohort still active in 2024 has the highest LTV in the book — $20,839 vs a $17,623 baseline. Deep-retention of old customers is working. But new acquisition has stopped.
Revenue growth of 8% YoY is being driven by a shrinking pool of increasingly valuable customers. At a conservative 5% annual attrition on the 2020-21 cohorts (roughly 60 customers/year) and current acquisition of 10-15/year, the active base shrinks 40% over five years. Every other finding in this memo is secondary to fixing this.
The biggest LTV variance isn't by state or by BU separately — it's in specific state × home-BU crosses. The standout is 10 customers in Colorado who have chosen Bijou Accessories as their home BU. Bijou has retail stores in Arizona, California, and Washington — nothing in Colorado. Yet these 10 customers have 90% loyalty penetration and average $33,025 lifetime value, 87% above the book average.
These customers are self-selecting into BUs whose stores they can't easily visit. Either they've become aware of these brands through digital (email, social, paid) or they're business/vacation shoppers. In any case, they're outperforming their home-state averages by wide margins, and they're concentrated enough that the pattern isn't noise.
Interview these specific customers. What drove them to Bijou in Colorado? Paid search? A friend? A media placement? Whatever it is, it's repeatable — and it produces the highest-LTV cohort in the book. This is a named list of ~80 customers worth a qualitative study.
LuxeStyle Mobile App's virtual "HQ" address in the warehouse is a NYC location. 12 of the 63 active New York customers have LuxeStyle Mobile App as their home BU. Of those 12, only one is a loyalty member — an 8.3% enrolment rate when the book average is 60%. Their average LTV is $11,312, 36% below the overall average.
This could be many things: a bug in the way new-to-brand NYC customers get assigned to Mobile as their home BU, legitimate discontent with the mobile experience among NYC customers, or customers routed through the Mobile App who wouldn't naturally have picked it. Whatever the cause, the combination — lowest-loyalty, lowest-LTV pocket — sits at the BU's own administrative home address.
First check: is this a data artifact? If the 12 customers all got assigned to Mobile because of NYC store lookups defaulting somewhere, the fix is at the assignment layer. If not, there's a product/experience gap in the NYC Mobile App cohort specifically — worth a product-team deep dive.
When you segment the customer file by LTV tier and ask how many are in the loyalty program, there's a counter-intuitive pattern: the top tier has the lowest loyalty rate. Customers with $25K+ lifetime spend are 57% enrolled. Middle tier ($15-25K) runs at 66%. So the program is catching mid-value customers disproportionately and missing the high-value ones.
137 of the top-tier customers are not loyalty members. They're already spending $25K+ on lifetime value without any program benefits. If even half of them could be moved into the program at current enrolment terms, that's 68 additional high-value members — roughly a 10% lift in program population from a cohort that's already proven its spend.
This is the most concrete opportunity in the book: a named target list of 137 high-spending non-members. A direct, high-touch invite (concierge call, not email blast) makes sense given their value.
The LuxeStyle Outlet store at Jersey Gardens recorded 286 transactions in 2022, 268 in 2023, and 317 in 2024 — growing. But when you look up which customers in V_CURRENT_CUSTOMER are New Jersey residents, the answer is zero. Every single transaction at this outlet is from an out-of-state customer.
Jersey Gardens is a high-traffic outlet mall about 30 minutes from Manhattan, popular with bus tours from the NYC hospitality industry. The data is consistent with that behavior: it's a destination outlet, not a local retail store. Customers come in from everywhere but NJ.
This outlet should be operated as a tourism channel — the marketing plan is hotel partnerships, airport signage, bus-tour outreach, not New Jersey-targeted geo campaigns. Loyalty enrollment at checkout becomes critical: these customers go home to 30+ other states and the only way to reach them again is if we captured them at the register.
Segmenting the customer base by household income band and comparing loyalty-member LTV to non-member LTV produces a surprisingly clean U-shape. The loyalty program drives a premium for the lowest-income bucket (<$30K, +16%) and the highest-income buckets ($150K+, +14-19%). In between — in the $30K-$75K range — loyalty members have lower LTV than non-members.
One plausible interpretation: at the low end, the program's discounts actually move behaviour (more trips, more units). At the high end, loyalty drives aspirational or collector behaviour (badges, exclusives). In the middle, discounts aren't compelling enough to change behaviour but they do anchor spend ceilings.
The program needs income-sensitive benefit design. A flat discount structure is training middle-income customers to discount-shop without moving their overall spend. Either differentiate benefits by tier (experiential/status for $100K+, discount+volume for sub-$30K), or lean out middle-income acquisition altogether.
Two structural facts about the Private Label Credit Card (PLCC):
This is unusual. PLCC programs are typically positioned as premium upsells — the card is supposed to capture higher-spenders and lock them in with payment convenience and card-exclusive rewards. Here, PLCC holders spend meaningfully less than peer loyalty members without a card.
Two possibilities: (a) PLCC enables payment-plan spreading that dampens total spend, or (b) PLCC is being cross-sold to the wrong loyalty cohort — perhaps customers who are budget-constrained rather than high-spend. Check the PLCC underwriting criteria and the promotional channels for card acquisition; if we're offering PLCC at the discount-shopper threshold, reconsider.
Across all 28 marketing-channel × order-source combinations in FACT_ORDER_TRANSACTION, Email attributed to a Web order generates 2.10 orders per customer — 26% more than the runner-up (Paid Search on Web, 1.67). Email on Mobile App is second at 1.87.
Every cell in the top tier is digital (Web or Mobile App), and every cell in the top tier is email or paid/organic search — not Social or SMS. Email's lead over Paid Search is 26%, which is a large gap for a channel that's also much cheaper per impression.
Given the acquisition cliff (Finding 2), the highest-ROI play in the book right now isn't net-new acquisition — it's re-activation. Email to the 714 loyalty members with a tailored offer to Web is the single channel-source combination most likely to generate a second/third order. Budget should concentrate here.
Breaking customers out by six age bands and comparing loyalty-member LTV to non-member LTV shows a single band where loyalty pays a positive premium: 35-44 year-olds (+$1,757, +10.3%). In every other band, loyalty members have flat or lower LTV.
The 35-44 band corresponds to prime earning years, often with school-age children in the household, high discretionary spend, and habituated brand loyalty. The program clicks for this cohort specifically. Younger customers (18-24, 25-34) are potentially being trained by loyalty discounts to wait for sales rather than build baseline spend. Older customers (65+) may be fixed-income and using loyalty to moderate spend rather than unlock premium behaviour.
If loyalty's "golden band" is 35-44, that's who marketing should be targeting with program acquisition. Acquiring 18-24 year-olds into loyalty may be counterproductive — a model fit to this data would predict they'd spend more outside the program than inside it.
32 hypotheses were tested across 7 warehouse tables: V_CURRENT_CUSTOMER (1,200 rows), FACT_CUSTOMER_PERFORMANCE (filtered to GOAL_TYPE_KEY=5 to avoid 10× overcounting), FACT_ORDER_TRANSACTION (12,540 orders), FACT_SALES_TRANSACTION (16,680 transactions), DIM_LOCATION, DIM_HOUSEHOLD, DIM_BUSINESS_UNIT. Every numeric claim above comes from a specific query result; the hypothesis log and individual SQL are captured in the exploration journal.
Findings are ranked by a combination of statistical surprise (how much the pattern diverges from what a reasonable operator would predict), sample size (findings based on <20 observations are excluded), and business consequence (which would change decisions if acted on).
Several plausible patterns tested null:
A few data gaps limit further analysis and are worth flagging:
In priority order, based on reversibility (small experiments beat big programs) and expected impact: