Measuring Revenue Lift: Baselines, Controls, and a Worked Cohort - Enterprise Digital Marketing

A measurement protocol for revenue lift in multi-location marketing: same-store baselines, control cohorts, and the leading indicators that move before revenue does. Includes a worked nine-store cohort (+9.9 percent year over year) and the control comparison that limits what we claim from it.

August 17, 2026 · 5 min read · By Nick Martinelli

Revenue lift is the correct verdict metric for multi-location marketing: the change in a store's total revenue against a defined baseline, drawn from the POS the operator already trusts. We made the case for it as the primary KPI in an earlier paper. This one is about the harder half of the subject: how to measure lift so the number survives an adversarial reading, including your own.

The reason to care is that revenue lift is simultaneously the most trusted number in the room and the easiest to fake, because every methodological shortcut moves it in the flattering direction. What follows is the protocol we run on our own book of 31 automotive service locations, with the actual results, including the parts that limit our own claims.

The protocol

Same stores, same months, year over year. The baseline is each store against itself, matched by calendar month, in the year before the program started. Seasonality in automotive service is large enough to swamp program effects; any pre/post comparison across different months is measuring winter versus summer.

Define the cohort before reading the results. Which stores count, from what start date, is decided by onboarding date, not by outcome. A cohort assembled after the fact converges on the winners by natural selection.

Run a control. Compare the cohort's growth against comparable stores outside it over the identical window. This is the step almost no marketing report includes, because it is the step that most often shrinks the claim.

Separate the verdict metric from the leading indicators. Revenue answers "is this working" on a two-to-three-quarter delay. Something has to answer it sooner, and it has to be something marketing can legitimately move.

The worked cohort

The onboarded-store cohort in our book, measured against the same months of the prior year:

Nine-point-nine percent, eight of nine stores growing, twelve months, seven figures of lift. That is the number a marketing deck would stop at.

The control comparison: over the same windows, comparable stores in the book outside the cohort grew at essentially the same rate. One onboarding wave grew 12.2 percent against a control of 12.5; a second wave grew 6.9 against 6.8. The cohort grew in line with the rest of a good automotive year.

An honest reading, then, is narrower than the headline: the cohort is real, the dollars are real, and nothing in the data shows the program caused the 9.9 percent, because the tide came in for everyone. What the program can defensibly claim sits in the leading indicators below and in the traced, closed-loop revenue covered elsewhere. Separately, the mature-book measurement using the same same-store discipline (stores onboarded six months or longer, matched half-year windows) showed a 13 percent lift, roughly $3M against ~$500K of media cost; that figure survives partly because its cohort and window were fixed in advance, which is the point of the protocol.

We publish the control result anyway, because a lift methodology you only trust when it flatters you is not a methodology.

What moves before revenue, and what does not

Across ten marketing takeovers, the first-90-day telemetry sorts cleanly:

Moves, and marketing can claim it: unique inbound callers (up at six of ten stores, mean +10.6 percent) and first-time callers (five of ten, mean +8.7 percent). Demand arrives at the phone before it arrives in the ledger.

Does not move systematically: average repair order. Pooled across ten stores, first-60-day average ticket went from $703 to $710, a 0.9 percent change, with six of ten up: statistically indistinguishable from a coin flip. Individual stores posted large ticket-quality gains (one at +43 percent on new-customer tickets), but a program-wide early ARO lift is not in our data, and we would treat one in anyone else's with suspicion.

Moves in the wrong direction if unmanaged: pipeline close rate (sold jobs as a share of sold plus declined) fell at six of eight measurable stores, mean minus 4.7 points, as new demand met an unprepared front counter. Answered rate did not improve on its own at any store where nobody managed it. More demand at a fixed conversion capability produces exactly this signature.

Moves spectacularly and means nothing: clicks. A campaign rebuild produced +431 percent clicks on 20 percent less spend at one store. Activity metrics inflate at every transition because transitions are rebuilds.

The practical use of this table: 90 days in, the honest scoreboard is caller growth and whether the store converted it, not revenue (too early) and not platform metrics (not evidence).

The three ways lift numbers lie

Every inflated lift claim we have deconstructed used one of three mechanisms:

1. Baseline shopping. Pre/post across unmatched months, or a cohort assembled after outcomes were known. Fix: calendar-matched, cohort-fixed-in-advance, control-compared.
2. Scope changes mid-series. A conversion or revenue definition that changed inside the trend. One shop's "6x conversion growth" across an agency transition decomposed to 80-85 percent counting-scope change. Fix: re-base any trend that crosses a definitional boundary.
3. Population matchback. Matching a vendor's contact list against the customer file and claiming every match, with no window and no exclusion of customers who were returning anyway. One vendor's matchback claimed 53 percent of a shop's entire annual revenue; the shop's own desk tags put the channel at 7.8 percent. Fix: per-identity joins with conversion windows, per the closed-loop paper.

The common thread is that none of these require bad faith, only unexamined defaults. The protocol exists so the defaults get examined.

Sequencing the two measurement systems

Revenue lift and closed-loop attribution are complements on a timeline, not substitutes. Lift needs no integrations and produces a store-level verdict from day one, at the cost of causal precision (as the control comparison above demonstrates). Closed-loop feeds take months to build across silos but produce channel-level traced revenue: the 7.1x ROAS floor and the six-figure uploaded revenue feeds described in the closed-loop paper. Run lift from the first month, build the loop in parallel, and retire neither: lift keeps the loop honest at the whole-store level, and the loop explains which channels earned the lift.