BIRCH RESERVE

Insights · September 11, 2026 · Measurement and customer value

Customer value is an empirical question, not a feeling

Birch Reserve Editorial

Customer value is an empirical question

“This customer came through a trusted introduction” is a useful description. It is not yet a measurement result. Customer value is easy to narrate and surprisingly easy to misattribute: people who accept an introduction may already be more engaged, better matched, or more likely to return. A serious growth loop therefore needs a design that can separate the story from the counterfactual.

One important reference is Schmitt, Skiera, and Van den Bulte's 2011 Journal of Marketing study [1]. The authors tracked approximately 10,000 customers of a leading German bank for almost three years. Compared with nonreferred customers with similar demographics and time of acquisition, referred customers had an average value at least 16% higher. The study also reports differences in contribution margin and retention, with the margin difference eroding over time while the retention difference persisted [1].

That is empirical evidence, but it is not a universal referral multiplier. The study was not a randomized advertising experiment. Referral and nonreferral customers can differ in motivation or unobserved ways even after matching. It concerns a German bank, not health and wellness services, display media, a Birch Reserve placement, or a curated private network. The result supports a measurement question; it does not answer ours.

Define the value before collecting the data

Write the outcome in advance. “Customer value” might mean contribution margin through a defined window, retained revenue, repeat booking, qualified downstream purchase, or another approved business measure. State the currency, cost treatment, observation window, and whether refunds, cancellations, service costs, and incentives are included. Do not quietly replace an outcome with a convenient proxy such as clicks.

Define the unit and cohort next. Is the unit a person, household, account, booking, or business? A partner may have multiple bookings and a customer may see multiple contexts. Choose an index event, freeze the eligibility rules, and label the first-exposure date. The same rules must be applied to introduced and comparison groups.

Make the counterfactual credible

The cleanest test randomly withholds an eligible introduction or assigns a comparable control path, subject to consent and operational safeguards. If randomization is not possible, document why and use a defensible comparison: matched cohorts, difference-in-differences, or another pre-specified design. Measure baseline activity, seasonality, price, partner, and channel. A matched comparison is better than a raw average, but it still cannot remove every unobserved difference.

Attribution should be narrow. Record the host, context, timestamp, consent state, creative or offer version, and the next event. Use a defined lookback and look-forward window. Deduplicate exposures. If someone would have booked without the introduction, crediting the introduction for the booking overstates value. Report assisted paths separately from last-touch paths rather than blending them.

Report uncertainty and failure modes

Publish the denominator: eligible opportunities, assigned controls, exposed customers, observed outcomes, exclusions, and missing records. Show interval estimates or sensitivity ranges where appropriate. Keep partner-level results separate when volumes or operating models differ. Pre-specify the primary outcome, then label exploratory cuts as exploratory. A small or incomplete sample should stay small or incomplete in the report.

For Birch Reserve, the initial hypothesis is that a relevant introduction after a purchase or booking may lower acquisition cost or raise customer value by improving attribution and purchase conviction. That is a hypothesis. The test should also be able to find no effect, a negative effect, or a benefit limited to one context. Privacy review, consent, and customer experience are part of the measurement design, not footnotes after the result.

Source limits

The Schmitt, Skiera, and Van den Bulte study is a valuable peer-reviewed reference for how customer value can be examined longitudinally. Its 16% figure is a finding about matched referred and nonreferred German bank customers, not a randomized ad result and not a Birch Reserve outcome. Generalizability and selection remain open questions. The right response is disciplined measurement, not a stronger claim.

Sources