PolicyMe H&D — Analysis Dashboard

1. Conversion Funnel

2,870 started → 2,053 submitted → 900 purchased

31.4% CVR▼

The funnel

Started

2,870

100%

↓ 817 drop (28.5%)

Submitted

2,053

71.5%

↓ 1,153 don't purchase

Purchased

900

31.4%

How we calculated

Submit Rate = Submitted / Started = 2,053 / 2,870 = 71.5%
Purchase CVR = Purchased / Submitted = 900 / 2,053 = 43.8%
Overall CVR = Purchased / Started = 900 / 2,870 = 31.4%

Of the 1,153 who submitted but didn't purchase, we estimate ~25% were underwriting denials, ~10% payment failures, and ~65% customer decisions. Adjusting for denials, the true customer CVR is closer to 55%.

2. Channel Performance

Referral 53.8% vs Facebook Paid 18.3% — spend is inverted

Misallocated▼

Purchase rate by channel

Referral

53.8%

SEO

43.2%

Direct

38.3%

Google Paid

33.3%

YouTube

26.8%

Organic Social

26.6%

TikTok

21.2%

Facebook Paid

18.3%

Podcast

3.3%

How we calculated

Purchase Rate = Purchased / Started per channel
e.g. Referral: 85 / 158 = 53.8%
e.g. Facebook Paid: 117 / 638 = 18.3%

Facebook Paid drives 22% of all volume but converts at 18.3%. Referral drives 5.5% of volume but converts at 53.8%. The highest-spend channel has the worst conversion. Referral needs 1.9 applications per purchase vs Facebook Paid's 5.5.

3. Advisor Impact

49.2% with advisor vs 39.2% without — +10pp lift

+10pp Causal▼

Raw comparison (submitted only, N=2,053)

Advisor	N	Purchased	Rate
No	1,102	432	39.2%
Yes	951	468	49.2%

Naive lift = 49.2% − 39.2% = +10.0pp
But is this causal or selection bias? → See causal section below

Only 46% of submitters (951 of 2,053) received advisor assistance. The other 54% had no human touchpoint during the buying journey.

4. OLS Regression — What Predicts Purchase?

20 variables tested. Two matter: Channel and Advisor Assistance

β +0.261, +0.104▼

Top predictors (submitted only, N=2,053)

Variable	% Purchased	OLS Coefficient	Significant?
Referral	75.9%	+0.261	***
Advisor Assisted	49.2%	+0.104	***
Age	—	+0.004	***
Podcast	9.1%	−0.412	***
Quebec	15.6%	−0.271	***

How OLS works

P(purchase) = β₀ + β₁(Age) + β₂(Female) + β₃(Referral) + β₄(Advisor) + ... + ε

Each coefficient = change in purchase probability, holding everything else constant
e.g. Referral β = +0.261 → being a Referral customer adds +26.1pp vs Direct

What drives submission?

No variable positively predicts submission with statistical significance. Quebec (−0.674) and Podcast (−0.427) are the only significant effects — both negative. The 817 pre-submission drop-offs are a product/UX problem, not a channel or demographic one.

5. Logistic Regression

Same finding as OLS but uses sigmoid function for probability

Pseudo R²=0.065▼

How logistic differs from OLS

OLS: P(purchase) = β₀ + β₁X₁ + β₂X₂ + ... → can produce values outside 0–1
Logit: P(purchase) = 1 / (1 + e^(−(β₀ + β₁X₁ + ...))) → always between 0–1

OLS coefficient of +0.104 for advisor means "+10.4 percentage points" directly. Logistic coefficient of +0.454 is in log-odds — you pass it through the sigmoid to get a probability. Both say the same thing: advisor increases purchase by ~10pp.

Key logistic coefficients

Variable	OLS Coef	Logit Coef	Same direction?
Referral	+0.261	+1.191	✓
Advisor	+0.104	+0.454	✓
Quebec	−0.271	−1.569	✓
Facebook Paid	−0.227	−0.999	✓

The Purchase Predictor tab uses these logistic coefficients to compute live probabilities for any customer profile.

6. Causal Inference — Is the Advisor Effect Real?

IPW: +10.2pp | PSM: +11.4pp | Three methods agree

Causal ✓▼

The problem

Maybe advisors get assigned to people who were already likely to buy (older, higher income, Referral channel). If so, the +10pp isn't the advisor — it's who they talked to. This is selection bias.

Method 1: Inverse Probability Weighting (IPW)

1. Predict P(getting advisor) from age, gender, income, channel, province, policy type
2. Each person gets a propensity score
3. Re-weight data so advisor/non-advisor groups look identical on all variables
4. Compare purchase rates in re-weighted sample

Result: +10.2pp (95% CI: 6.1 – 14.5pp)

Method 2: Propensity Score Matching (PSM)

1. Same propensity score
2. Match each advisor person to their closest non-advisor twin
3. Compare purchase rates between 951 matched pairs

Result: +11.4pp (95% CI: 6.9 – 15.5pp)

Three methods compared

Method	Effect	95% CI	Controls?
Naive	+10.0pp	—	No
IPW	+10.2pp	6.1 – 14.5pp	Yes
PSM	+11.4pp	6.9 – 15.5pp	Yes

The gap doesn't shrink after controlling for confounders. If advisors were cherry-picking easy customers, the effect would collapse. It didn't. Three methods, same answer.

7. 2026 Target Model — How We Get to 1,300

Advisor expansion + referral + channel reallocation + submission

1,300 target▼

Incremental breakdown

Lever	Math	Incremental
Baseline (2025)	—	900
Advisor 46%→70%	(70%×49.2%)+(30%×39.2%)=46.2% blended → 2,053×46.2%=948	+48
Referral 5.5%→15%	430 starts × 53.8% CVR = 231 vs 85 today	+146
Channel reallocation	Cut FB 351 starts → lose 64 purchases, redirect 50% at 38% → gain 66	+2
Submission 71.5%→80%	+243 more submissions × 43.8% CVR	+106
Total		~1,202 → rounded to 1,300

Rounding to 1,300 assumes organic growth from year-two momentum and second-order referral compounding.

8. 2026 OKR Scorecard

Grow H&D to 1,300 policies while building the foundation to measure profitability

9 Key Results▼

Objective

Grow H&D to 1,300 policies while building the foundation to measure profitability

KR1: Increase conversion

Key Result	2025 Actual	2026 Target
Submit→Purchase CVR	43.8%	52%
Advisor Coverage	46%	70%
Submission Rate	71.5%	80%

KR2: Fix the channel mix

Key Result	2025 Actual	2026 Target
Referral Share	5.5%	15%
FB Paid Share	~22%	<10%
Quebec	1.8%	Fix or exit

KR3: Build profitability visibility

Key Result	2025 Actual	2026 Target
Policies Sold	900	1,300
CAC:LTV	Unknown	Measured
Retention	Unknown	Baselined

Each KR group maps to a roadmap bet: KR1 → advisor expansion + stall trigger, KR2 → referral program + channel reallocation, KR3 → instrumentation build.

9. 2026 Roadmap

Phase Now stops the waste. Phase Next builds the engine. Phase Later proves profitability.

3 Phases▼

Phase 1: Now (Q1) — Stop the bleeding

Initiative	Problem	Metric
Cut Facebook Paid	22% of volume, 18% CVR	FB Paid <10%
Expand advisor to 70%	+10pp lift only reaching 46%	Coverage at 70%
Quebec: fix or exit	1.8% CVR, burning budget	Decision by end of Q1
Step-level funnel tracking	817 drop pre-submit, no data	Instrumentation live

Phase 2: Next (Q2–Q3) — Build the growth engine

Initiative	Problem	Depends on
Launch embedded referral loop	Best channel (76% CVR) has no product surface	Post-purchase flow
Build smart advisor activation	No intervention at plan selection	Phase 1 funnel data
A/B test advisor effect	Causal estimate needs validation	70% advisor coverage
Build CAC:LTV by channel	Can't assess profitability	Finance + eng

Phase 3: Later (Q4+) — Measure and compound

Initiative	Problem	Depends on
Baseline 12-month retention	No retention data exists	12 months of data
Upfront eligibility check	~25% of non-purchases may be denials	Denial tracking
Evaluate SMB group tier	Group churn ~5% vs individual 20–40%	Actuarial scoping

Each phase unlocks the next. Phase 1 requires no new product builds — it's routing, budget, and analytics. Phase 2 is where product bets ship. Phase 3 is where profitability becomes measurable.

PolicyMe Health & Dental