Product Analytics

HarborMug management report

Prioritize an investigation of the activate-to-pay journey, starting with event definitions and checkout instrumentation. It has the highest proportional funnel loss: 2 of 4 activated users do not appear at pay (50%). Each transition loses two users, so checkout is not uniquely the largest loss by count. Weekly sales growth comes with equal growth in traffic and advertising spend; conversion and revenue efficiency are unchanged. These results do not justify scaling advertising or claiming that the checkout copy revision improved performance.

All supplied business data are synthetic examples; no operational change or sales lift has been measured.

Measurement basis

Use AARRR for this growth-stage business: acquisition through visit and signup, activation through the supplied activate label, retention through recorded activity, and revenue through pay reach and weekly trading totals. Referral is unmeasured. The supplied activation label still needs a business definition before it can represent customer value.

The activity observation cutoff is November 2, 2026, inclusive. Dates have day-level resolution; no timezone is supplied. Retain each distinct user once in each denominator, each distinct user/cohort/activity-date combination for activity, and each distinct user/stage pair for funnel reach. This removes one duplicate activity row (18 to 17) and one duplicate funnel row (21 to 20). Each file contains eight distinct users. Cohort dates are consistent within users, and all activity falls between its cohort anchor and the cutoff.

The two files use different user identifiers with no overlap or mapping. Do not join them or infer retention for paying users. Deduplication is by supplied ID only; cross-device or cross-account identity cannot be resolved.

Current conversion definition: distinct users reaching each supplied stage divided by users in the preceding stage, across the undated funnel extract. Stage sets are nested, but sequence and elapsed time cannot be verified. The extract has no date range; November 2 cannot be asserted as its cutoff. Thus 25% visit-to-pay is stage reach, not a verified time-bounded purchase conversion rate.

Proposed production conversion window: follow users from first eligible visit for seven elapsed days; require visit → signup → a defined activation → confirmed payment in chronological order. Count each user once, attribute to the first-visit cohort, and exclude users lacking the full seven-day follow-up from the mature denominator. This is a measurement proposal, not a recalculation possible from the current files. Confirm whether signup and activation are required purchase steps before enforcing this path; track direct-purchase paths separately if permitted.

Retention definition: use the supplied cohort_date as an anchor of unspecified business meaning, not an assumed signup or first-purchase date. W0 is anchor days 0–6, W1 days 7–13, W2 days 14–20, and so on. A user is retained if at least one activity occurs in that window. Count a window only once it has fully elapsed for the cohort. Exact-day retention is separate and requires only that day to have elapsed. Missing activity in a fully observed window counts as zero, conditional on the supplied event coverage being complete.

Funnel losses

Stage	Unique users	Conversion from preceding stage	Users lost	Loss rate	Share of visitors
Visit	8	—	—	—	100%
Signup	6	75.0%	2	25.0%	75.0%
Activate	4	66.7%	2	33.3%	50.0%
Pay	2	50.0%	2	50.0%	25.0%

Six of eight visitors do not reach pay: a cumulative 75% loss. Activate-to-pay has the largest proportional drop, while all three transitions tie in absolute loss. With only four activated users, one additional payer would change that transition rate by 25 percentage points. Treat the ranking as an investigation priority, not proof of a persistent checkout problem or an estimate of recoverable revenue.

Mature retention cohorts

Values are active users / cohort users and rate. N/A means the full window has not elapsed; it does not mean churn.

Anchor cohort	Users	W0	W1	W2	W3	W4
October 5	4	4/4 (100%)	2/4 (50%)	0/4 (0%)	0/4 (0%)	N/A
October 19	4	4/4 (100%)	2/4 (50%)	N/A	N/A	N/A

The October 5 W4 window is November 2–8, and the October 19 W2 window is November 2–8. Both are incomplete. The older cohort's W2 and W3 are complete and show no recorded activity. The newer cohort has the same W1 rate; there is no observed improvement in that comparable window. Across both mature W1 cohorts, retention is 4/8 (50%). Do not blend the older cohort's mature W2 zero with the newer cohort's unfinished W2.

Anchor cohort	D1 exact day	D7 exact day	D14 exact day	D28 exact day
October 5	2/4 (50%)	2/4 (50%)	0/4 (0%)	0/4 (0%)
October 19	3/4 (75%)	2/4 (50%)	0/4 (0%)	N/A

Both D14 observations are mature, yielding 0/8, even though the newer cohort's full W2 is immature. Older-cohort D28 is also mature on the cutoff date while its full W4 is not. Exact-day zeros do not prove absence of activity throughout a week.

The curve shows activity at anchor, a drop to 50% in W1, and disappearance in later mature windows for the older cohort. Four users per cohort are too few to establish a stable curve or a release effect. W0 is not repeat retention because every user already has anchor-day activity. Five of eight users have at least one activity on a later date, but this cumulative return measure has unequal follow-up and should not replace the mature cohort comparison.

Recorded activity is not identified as a visit, key action, or purchase. These results describe generic activity retention only. Return visits and repeated key actions require event types; repeat purchases require customer-linked, distinct paid order IDs and payment dates. Neither repeat-purchase rate nor purchase frequency is measurable here. The funnel's two pay users and weekly order totals cannot supply those missing links.

Weekly trading performance

Dollar amounts are treated as USD for this US merchant. The weekly extract labels periods only as prior/current, so their dates, equal duration and relationship to the release cannot be verified.

Metric	Prior	Current	Change
Sessions	4,000	4,800	+20%
Orders	100	120	+20%
Gross sales	$2,400	$2,880	+20%
Refund value	$120	$144	+20%
Sales less refunds	$2,280	$2,736	+20%
Advertising spend	$300	$360	+20%
Orders / sessions	2.50%	2.50%	0.00 percentage points
Gross average order value	$24	$24	0%
Refund value / gross sales	5.00%	5.00%	0.00 percentage points
Gross sales / ad spend	8.00×	8.00×	0%
Sales less refunds / ad spend	7.60×	7.60×	0%
Ad spend / order	$3	$3	0%

At the prior order/session rate, 800 additional sessions imply 20 additional orders, matching the observed increase. At unchanged $24 AOV, that accounts arithmetically for the $480 gross-sales increase. This is a decomposition, not evidence that advertising caused the increase. Sales less refunds rose $456 with $60 more ad spend, but this is not profit: fulfillment, goods, payment fees, tax and refund accounting are not fully specified at order level.

The sales/ad-spend ratios are blended store ratios, not attributed advertising ROAS. Ad spend/order is not customer acquisition cost because orders are not distinct newly acquired customers. Refund value rate is not the percentage of orders refunded. The 2.5% order/session ratio and 25% user funnel reach measure different populations and denominators and must not be reconciled as if they were one conversion metric.

The October 19 checkout copy revision has no causal evidence. The newer cohort starts on that date, but neither feature exposure nor dated funnel events nor dated trading periods are available. Seasonality and same-period-last-year comparisons are also unavailable.

Next investigation and operating dashboard

First priority — validate activate-to-pay and locate the failure. Assign the product/checkout owner, with analytics support, to define activation, confirm whether stage skipping is allowed, and reconcile the two activated non-payers against checkout starts, payment attempts, errors and confirmed orders in the source system. Obtain timestamped events with a common customer identifier, event type, order ID, release exposure, device and acquisition channel. First separate instrumentation gaps from actual abandonment, then compare mature seven-day funnels across equivalent periods and segments. These are recommended investigations; none has been performed in this analysis.

The next retention investigation should examine why the October 5 cohort has no W2/W3 activity: verify tracking coverage and identify which later visits or actions would represent customer value. Track visit retention, key-action retention and second-purchase incidence separately. For purchases, anchor on first confirmed paid order and measure a second distinct paid order within 30 days among customers with complete 30-day follow-up. A durable mug may have a longer repurchase cycle; validate the horizon against purchasing history before adopting a commercial target.

Use six executive metrics below, with weekly cohort filters and counts alongside percentages. Thresholds are proposed operating rules based on this small baseline, not established performance standards; rebaseline after obtaining comparable production coverage.

Metric	Current baseline	Proposed target / alert	Owner	Decision rule
Ordered seven-day visit-to-pay	Unavailable; undated reach 2/8 (25%)	Establish mature baseline; alert below 25% provisionally	Analytics lead	Repair coverage first; investigate a mature decline before changing acquisition spend
Activate-to-pay	Undated reach 2/4 (50%)	Above 50%; alert at or below 50%	Checkout owner	Review payment and abandonment evidence; prioritize a specific fix only after diagnosis
Mature W1 activity retention	4/8 (50%)	At least 50%; alert below 50%	Product owner	Inspect onboarding and later value actions; do not infer repurchase
Orders / sessions	2.50% in both periods	At least 2.50%; alert below 2.50%	Operations lead	Check dated comparable periods and segment mix before attributing movement
Sales less refunds / ad spend	7.60× in both periods	At least 7.60×; alert below 7.60×	Growth lead	Review acquisition mix and contribution economics before scaling
Refund value / gross sales	5.00% in both periods	At most 5%; alert above 5%	Finance lead	Reconcile refund timing and reasons before diagnosing product quality

Below the executive layer, show acquisition → activation → pay by cohort, channel and device, followed by the mature retention matrix. At feature level, add checkout-copy exposure, adoption, repeat key actions, payment errors and purchase outcomes once instrumented; those measures cannot be inferred from these inputs. Annotate the release on dated charts without attaching causal claims.

Supporting deliverables

• funnel_metrics.csv: unique stage counts, transition losses and conversion rates.
• retention_weekly.csv: cohort windows, maturity flags, counts and rates; immature cells are blank.
• retention_exact_day.csv: exact-day maturity and retention, separate from weekly windows.
• weekly_metrics.csv: source totals and calculated trading ratios.
• events_deduplicated.csv and funnel_deduplicated.csv: cleaned business records.
• verification.json: row counts, duplicate removal and consistency checks.
• calculate.py: reproducible calculations; run python3 deliverables/calculate.py from the folder containing inputs/ and deliverables/.

Rates are stored as fractions in the CSV files. Weekly formulas are orders/sessions, gross/orders, refunds/gross, gross/adspend, (gross−refunds)/adspend and adspend/orders. Supporting checks confirm nested funnel membership, consistent cohort dates and events within the observation period.