DogWood — Home Boarding & Daycare

Customer-acquisition experiment plan

How we turn a purchased prospect list into a learning machine — finding the few high-LTV, repeat households a limited-capacity home boarder can actually serve, and proving which data signals are worth paying for.

VERSION  0.1 — DRAFT
DATE  JUN 2026
OWNER  DAN & AUDREY
HORIZON  ROUND 1
00

The reframe

Start here — it changes everything downstream.

Premise

We are not testing "is there demand?" — there is. We are testing which purchased signals predict the handful of households we can serve. So we buy the list as a structured test matrix, not one big blast. Same postage — engineered to teach us which overlays earn their cost.

01

The loop

Five stages, run sequentially. Each round retires one assumption and feeds the next list buy.

01

Hypothesize

Write each belief as falsifiable: "[signal] predicts [behavior]; right if [metric] beats [threshold]." Rank by risk.

02

Test

Mail a structured matrix. Each cell isolates one variable, carries a unique trace code, against a no-overlay holdout.

03

Validate

Measure each cell against thresholds set before mailing — response, booking, repeat, payback. No moving goalposts.

04

Pivot / persevere

Pre-committed triggers decide. Kill signals that don't beat their cost; double down on the ones that do.

05

Learn

Codify results into a weighted signal scorecard. Every future buy is priced by proven lift, not intuition.

02

The hypotheses, ranked by risk

If the riskiest ones fail, the cheaper ones don't matter — so we test them first.

H1
Data quality
Verified pet-ownership converts materially better than modeled — enough to justify its higher cost-per-record.
Riskiest
H2
Demand driver
The frequent-traveler flag predicts boarding; dual-income predicts daycare. Different signals, different service.
High
H3
Lifetime value
Multi-dog households book both services and repeat more often — our highest contribution margin per acquired customer.
Medium
H4
Geography
Drive-time targeting beats radius on inquiry-to-booking — proximity to the door matters more than map distance.
Medium
H5
Segment offer
The two buyer profiles — 25–44 working families vs 55–70 active retirees — need different messages, not one postcard.
Medium
03

The test matrix

One variable per cell. A unique trace (promo code / landing URL / phone extension) makes every booking attributable.

Cell Isolates Variable A Control / Variable B Trace
A Data quality · H1 Verified pet-owner Modeled pet-owner PAW-A1 / A2
B Boarding driver · H2 Frequent-traveler ON Traveler flag OFF PAW-B1 / B2
C Lifetime value · H3 Multi-dog household Single-dog household PAW-C1 / C2
D Geography · H4 Drive-time ≤ 15 min Radius ≤ 6 mi PAW-D1 / D2
E Segment offer · H5 "First dog" → 25–44 "You travel often" → 55–70 PAW-E1 / E2
Baseline holdout Generic $100k+ list, no overlays Proves the overlays actually lift response PAW-00
04

Validation thresholds

Set before the first piece mails. A cell only "wins" if it clears its gate — judged on numbers, not hope.

Mail → inquiry
2%+
Baseline cold mail runs ~0.5–1%. A winning overlay cell roughly doubles it.
Inquiry → booking
30%+
A strong meet-and-greet close on qualified, well-matched households.
Repeat ≤ 90 days
40%+
The real prize for limited capacity — fill slots with returners, not churn.
CAC payback
1stay
First booking recovers list + postage. Blended CAC under one stay's margin.
05

Pivot triggers

Decided in advance so live results aren't argued with. When the condition fires, the action is automatic.

If

Verified doesn't beat modeled by enough to cover its premium → stop paying for verified; mail more modeled records for the same spend.

If

Traveler is the strongest single predictor → make it a required overlay on every future buy and shift budget toward it.

If

Drive-time cells out-convert radius → switch all future geo-targeting to drive-time and drop radius selects entirely.

If

New-movers inquire but rarely book → they're shopping, not ready; route them to a nurture sequence instead of re-mailing cold.

06

What we keep — and where to start

The output of every round is a weighted scorecard. The input to Round 1 is discipline about scale.

Output · the signal scorecard

Each overlay earns a measured lift

Results become coefficients. Future buys are priced and weighted by proven lift — the part vendors can't sell your competitors. (illustrative)

SignalLiftCost
Frequent traveler+1.8× respbase
Multi-dog household+2.4× LTVbase
Verified pet-owner+1.3× conv2× / rec
Drive-time ≤ 15 min+1.5× book+ fee
New mover (12mo)+1.2× respbase

Input · the honest constraint

At boutique scale you can't power every cell at once.

Cold mail runs ~1% response — a 1,000-piece cell yields ~10 inquiries, far too noisy to trust. So we don't test everything in parallel. Round 1 funds only the two riskiest variables, with cells big enough to read a real signal. Prove those, then test the next two.

ROUND 1 → H1 · verified vs modeled ROUND 1 → H2 · traveler flag
07

RFQ cover-emails

Four lines apiece, to top each request for quote. Same core selects across all three so quotes are comparable. Swap the [ bracketed ] fields.