# Keep COD. Close the handover loop.

## Decision requested

Authorize a gated pilot with a **PHP2.61m ceiling**, subject to carrier, cost and enrollment feasibility before build spending is released. Lead with buyer preferences that a participating courier can accept and execute, followed by verified recovery of unsuccessful attempts. Keep standard COD available to nonresponders. Test apparel purchase confidence separately; release the prepaid test only after product/payments feasibility is approved.

The supplied case data is fictional. The recommendation is a testable operating hypothesis. Intervention effects, commercial rates, eligible volume and integration effort are planning assumptions. No firsthand app observation, partner agreement, interview or measured intervention result is claimed.

## 1. What the evidence establishes

The national planning baseline is 5m daily orders × 70% COD × 29% final failure × 30 days: **30.45m failed COD orders per month**. At the case's cost allocation, the platform bears PHP1.5225bn reverse-logistics burden and PHP1.827bn allocated cash-handling overhead. Sellers bear PHP70 forward shipping per failed order.

Allocated overhead is not automatic cash savings. Forward shipping remains a cost when an order succeeds. The commission opportunity calculated from the channel-average PHP800 basket is PHP2.436bn, compared with approximately PHP2bn stated in the narrative; the basket value of failed orders is not separately supplied. Preserve that distinction.

| Supplied signal | Quantification | Decision implication and limit |
| --- | --- | --- |
| Apparel: 28% of orders, 78% COD, 33% failure | PHP1.189bn monthly raw platform burden; FMCG PHP784m despite 22% failure | Prioritize apparel for a confidence experiment. Burden does not establish the fraction preventable by a recap. |
| NCR: 18% failure; Mindanao: 35% | PHP653m and PHP766m raw monthly platform burden | Propose serviceable lanes in two operating settings. Compare treatment with control within each setting; do not attribute the regional gap to geography alone. |
| Creator-live: 32% noncampaign failure, 38% campaign | At equal 100k COD orders, 6,000 more failures and PHP660k additional platform burden | Test order-confidence friction; do not assume campaign causality or monetize unknown campaign exposure. |
| XYZ: 26% failure at 3.2 days; A&B: 29% at 2.5 days | Same-day costs PHP120 versus PHP70 standard; same-day coverage is NCR-only | Faster service is not sufficient evidence of better performance across different carriers and customer mixes. Defer a blanket speed upgrade. |
| Newest buyers: 38% failure, versus 18% after a year | Buyer shares are available; cohort order counts are not | Instrument tenure in the pilot. Do not invent a new-buyer addressable order volume or a causal tenure effect. |
| FSV text: 15% relatively higher failure | Redemption, exposure and settlement are absent; regional figures do not reconcile | Investigate incentive-driven low intent, but defer a priced voucher redesign. |

Raw platform burden uses a 30-day month × supplied daily volume × order share × COD share × failure × PHP110. It includes allocated overhead for diagnosis, not as an intervention cash benefit. Regional and category totals do not reconcile to the national headline. Payment shares total 97%. There is no joint buyer × category × channel × region × carrier table. Keep these views separate; never add overlapping dimensions or multiply marginal shares into an observed target cohort. Full audit: `analysis/data-quality.md`.

## 2. Why test handover first?

The exhibits identify burden, not weighted failure causes. They do **not** establish that handover coordination is the largest preventable cause. Select it as the first experiment because it preserves cash access and creates specific operating actions, named owners and observable states that can be tested through one existing 3PL. This preference is conditional on actual carrier capability and commercial terms.

| Candidate | Evidence and proposed mechanism | Modeled hurdle / key uncertainty | Priority |
| --- | --- | --- | --- |
| Acknowledged handover | A buyer preference reaches an operator who accepts a supported action; an unsuccessful attempt receives an accountable recovery owner | 1.20pp paid-completion gain at base unit inputs; carrier actionability, cost and preventable failure share unknown | First operational test, after carrier feasibility |
| Apparel purchase confidence | Highest raw category burden; size/variant/total-price recap may reduce avoidable refusal | 2.22pp operating break-even at PHP450 AOV and PHP2/assigned order; effect on commitment unknown | Separate supporting test |
| First prepaid adoption | COD is 70% of national orders; funding/refund clarity and a capped offer may enable switching | 20% incremental conversion and 10% migrated failure are assumptions; payment-state integration and incentives need approval | Feasibility-gated supporting test |

These hurdles use different baskets, costs and mechanisms; they are not an apples-to-apples causal ranking. The decision favors a controllable experiment rather than the highest assumed uplift. Do not launch three nationwide initiatives.

Defer COD bans, blanket faster delivery, a new locker network and voucher-threshold redesign. They introduce access risks, broad capital or subsidy commitments, or missing commercial inputs. NCR and Mindanao are proposed operating environments, not proven optimal pilot locations. Use only lanes the participating carrier can service and enroll; redesign the geographic plan if that gate fails.

## 3. One parcel: the proposed change

The following is an illustrative design, not an observed app journey or a real customer outcome.

| Stage | Proposed experience | Accountable owner / recorded state |
| --- | --- | --- |
| Buyer preference | Buyer nominates an authorized receiver, Maria, and adds a landmark before the existing route cutoff. Only carrier-supported choices appear. | Product records order ID, consent, preference version and submission time. No input keeps standard COD. |
| Carrier acceptance | Dispatcher validates serviceability and sends the accepted receiver instruction to the rider's manifest before cutoff. | 3PL ops records route/date, operator, acknowledgment time and accepted/fallback state. API receipt alone does not count. |
| Attempt evidence | An attempt is unsuccessful because the receiver is unavailable. The existing timestamp/reason/evidence process records it; the buyer can confirm or dispute. | Carrier records evidence; platform ops routes the exception. An unsuccessful first attempt is not terminal order failure. |
| Owned recovery | A named lane exception desk selects a supported receiver/reattempt before the next cutoff, or follows the standard return process when recovery is unavailable. | Exception owner records action, timing, extra attempts, support cost, final delivery, settlement and returns. Only additional paid completions count as recovered sales. |

If a preference is unsupported or unaccepted by cutoff, show standard service as the fallback. Do not promise exact-time delivery, automatically cancel for nonresponse or remove COD access for missing acknowledgment. Apply any existing restrictions only through verified evidence and a review/appeal path. Do not reward attempted-delivery counts that do not improve final outcomes.

The country operations lead is accountable for the pilot. Product owns preference/fallback; 3PL ops owns manifest actionability and exception routing; seller ops owns the apparel recap; payments owns safe conversion and funding/refund information; analytics/finance owns assignment and realized contribution. Carrier workload, seller workload and buyer complaints are measured alongside financial outcomes.

### Before any build release

Country operations and finance must verify a sample manifest/driver view can carry the accepted instruction, identify an operator and exception owner, establish cutoff and fallback rules, obtain billing/reattempt terms and estimate the eligible enrollment and support workload. Record evidence and an agreed cost range. If these cannot be established, stop or redesign the handover experiment; do not treat the operating diagram as proof of feasibility.

### Supporting initiatives

**Apparel confidence:** use a separate eligible cohort for a short size/variant, total-price and delivery recap before seller handover. Track cancellations and delivered value across all assigned orders; preventing dispatch is not a recovered sale. The +4pp completion target and PHP2/order cost are planning inputs, yielding PHP160k contribution per 100k treated orders at PHP450 AOV.

**Prepaid:** first perform a bounded feasibility review of the proposed conversion of an already placed, eligible unshipped COD order. Product/payments must approve buyer eligibility, funding/refund explanation, incentive rules, payment settlement, carrier cash-due consistency, duplicate-charge protection and standard-COD fallback. The PHP5 reward and 20% incremental switching target are hypotheses, not validated incentive economics. Do not release prepaid development or enroll its cohort before this gate passes. If it fails, remove that cohort and recost/reapprove the portfolio; do not retain its modeled outcomes or silently move its sample into another test.

## 4. Real-world precedent and transfer limits

| Existing practice | What to adapt | What it does not establish |
| --- | --- | --- |
| Published TikTok Philippine guidance includes verification and repeat delivery attempts | Add a traceable preference → supported operator acknowledgment → owned recovery loop | Ordinary reminders, verification and retries are not novel |
| DHL Malaysia Express offers delivery-date, receiver and location controls | Offer only participating-carrier-supported choices with visible fallback | Philippine COD capability, local pricing or a transferable uplift |
| Shopee PH supports QR payment for eligible SPX COD parcels | Treat doorstep QR as a possible cash-collection bridge | It remains payment on delivery; it does not prove prepayment adoption |
| GCash publishes cash-in fees | Account for buyer funding effort in the prepaid design | Buyer cash-in fees do not validate our merchant-fee assumption |

Primary sources checked 24 October 2026: [TikTok](https://seller-ph.tiktok.com/university/essay?default_language=en&knowledge_id=7654197957248769), [DHL](https://www.dhl.com/discover/en-my/ship-with-dhl/services/optional/dhl-on-demand-delivery), [Shopee](https://help.shopee.ph/portal/4/article/81256?seo=1), [GCash](https://help.gcash.com/hc/en-us/articles/360017595774-Cash-in-fees). These are implementation precedents; no measured causal effect is transferred. The source register records dates and limits.

## 5. What must be true for the economics to work?

Use **100,000 eligible original-COD orders assigned to handover treatment**, including nonresponders and missing acknowledgments. This is a normalization, not monthly reach or the full randomized sample. Effects are additional paid completions across every assigned order. Cancelled, pending or suppressed-dispatch orders are not recovered sales.

At base planning inputs, a +3pp paid-completion gain produces 3,000 additional completed orders. Each contributes PHP50 avoided reverse charges + PHP80 commission (PHP800 × 10%) − PHP5 incremental-delivery reserve = PHP125 before programme cost. Therefore:

**3,000 × PHP125 − 100,000 × PHP1.50 = PHP225,000 contribution before setup and other unmodeled costs.** Allocated overhead is excluded.

At the 29% planning baseline, 3,000 recoveries represent approximately **10.3% of the 29,000 otherwise failed orders**. Operating break-even at base inputs is PHP1.50 / PHP125 = **1.20pp**, equivalent to approximately **4.1% of otherwise failed orders**. These are required-equivalent shares, not evidence that those failures are recoverable. Replace the national benchmark with the actual cohort baseline. With no realizable reverse-charge saving, break-even becomes **2.00pp** at the stated AOV and cost.

| Decision stress, per 100k assigned treatment orders | Completion gain | Cost/order | Recovered AOV | Avoided reverse/recovery | Contribution |
| --- | ---: | ---: | ---: | ---: | ---: |
| Base target | 3pp | PHP1.50 | PHP800 | PHP50 | PHP225k |
| Higher operating cost | 3pp | PHP2.50 | PHP800 | PHP50 | PHP125k |
| Smaller recovered baskets and higher cost | 3pp | PHP2.50 | PHP450 | PHP50 | PHP20k |
| No reverse saving and higher cost | 3pp | PHP2.50 | PHP800 | PHP0 | −PHP25k |
| Lower effect and higher cost | 0.5pp | PHP2.50 | PHP800 | PHP50 | −PHP187.5k |

These are transparent stresses, not probabilities or confidence intervals. The existing 5pp/PHP1.50 upside gives PHP475k. Finance must use actual retained commission after refunds/returns, carrier billing, reattempt costs and support; the PHP5 reserve is not proof those costs are covered.

The PHP1.50 operating target comprises PHP0.20 in-app events, PHP0.90 support (10% exceptions × 3 minutes × PHP180/hour) and PHP0.40 coordination. These are planning inputs, not quotes. Meter messages, exception frequency, handling time, coordination and extra attempts. Collect actual recovered-basket values and whether reverse charges can genuinely be avoided before using any savings in a scale decision.

For prepaid, the separate base assumptions are 20% incremental switching relative to control, 10% migrated failure, 1% merchant fee, PHP5 reward with 1.5× reward leakage and PHP5 extra-delivery reserve. They produce PHP165k per 100k offers; the migrated-failure operating break-even is 16.6%. The randomized original-COD denominator is retained. Buyer funding costs, actual merchant fees, conversion and refund costs require measurement; doorstep QR is not counted as prepayment.

PHP2.5m setup requires approximately **1.11m treated handover orders** to recover at PHP2.25 contribution/order. A six-month hurdle requires at least **185,186 treated orders/month** at that margin. These are thresholds to verify, not claims of reachable volume. Use conservative measured margin and actual setup at the scale gate.

## 6. Funding, release gates and execution

Present the pilot as an **approximately PHP2.6m learning investment**, with a PHP2.61m authorization ceiling. The detailed conditional plan totals PHP2.6044m: PHP2.5m setup plus PHP104,400 variable cost/reserve. That calculation is precise; the underlying effort and rates are not quotations.

| Setup component | Planning basis | Estimate |
| --- | --- | ---: |
| Product/engineering | 16 person-weeks × PHP75k loaded rate | PHP1.2m |
| Integration/QA | 12 person-weeks × PHP50k | PHP600k |
| Launch/training | 12 person-weeks × PHP25k | PHP300k |
| Evaluation | 4 person-weeks × PHP50k | PHP200k |
| Contingency | Planning allowance | PHP200k |

Launch/training excludes variable exception support. No partner subsidy is assumed. Release build expenditure only after carrier, billing, workload, enrollment and integration feasibility is approved. Release enrollment only after instrumentation and evaluation design pass. Prepaid has its additional approval gate. If actual quotes, capability, sample or costs differ, recost and obtain a revised decision within the ceiling, or stop/seek a revised authorization; do not reduce statistical adequacy silently to preserve the cap.

The full conditional design randomizes **100k orders**, only **50k treated**: handover 30k treatment/30k control, confidence 10k/10k, prepaid 10k/10k. The 60/20/20 split is resource allocation, not market shares. Base pilot contribution is PHP100k before setup and **−PHP2.4m after setup**. The pilot purchases evidence and does not repay setup.

| Stage | Owner and release requirement |
| --- | --- |
| Weeks 1–2: feasibility | Country ops + finance verify actionable carrier workflow, commercial costs, serviceable lanes, enrollment, payment feasibility and baseline data. Stop/redesign unsupported components. |
| Weeks 3–4: implementation | Product + 3PL ops implement one adapter and driver acknowledgment; QA fallback, recovery and event capture. Payments work requires separate approval. These are target dates, conditional on integration effort. |
| Weeks 5–8: enrollment | Analytics locks cohorts, assignment and guardrails. Start handover, then only approved supporting tests. Keep stable route assignments and full assigned-order denominators. |
| Weeks 9–12: mature/read | Analytics + finance reconcile terminal delivery, settlement and known refunds. Extend immature outcomes. No automatic rollout date. Follow 90-day repeat separately. |

## 7. Evaluation and scale decision

Before assignment, create disjoint cohorts deterministically: handover pilot routes first, then eligible apparel, then remaining approved wallet-ready offers. Enroll one first order per buyer and lock follow-up assignment. Never remove nonresponders or missing acknowledgments after randomization. Supporting results apply to their own cohorts, not automatically to handover lanes.

Randomize matched stable courier-route clusters once for handover; do not switch arms daily. The planning design is 300 clusters/arm with 100 enrolled first orders/cluster, not all route traffic. Track reassignments and cross-arm learning. At 29% planning baseline, 1.5pp minimum detectable difference, two-sided 5% alpha, 80% power and ICC 0.01, the equal-size normal approximation needs about 28,140 orders/arm; propose 30k. Before enrollment, verify that 600 stable clusters are available and re-estimate baseline, cluster-size variation, correlation and delivered-value variance. Increase/extend or redesign if necessary. This is pooled power, not independently regional, and not 80% assurance of passing an economic gate.

Use a common predeclared 30-day order observation window for terminal delivery, settlement and known returns/refunds. Extend unresolved outcomes and disclose later liabilities. Analyze all assigned eligible orders. Measure the operating funnel—preference offered, submitted, accepted, driver-visible, recovery acted on—without substituting responder-only success for the randomized result. Collect failure reasons without assigning unobserved historical cause weights.

Scale only when:

- The 95% lower confidence bound on realized contribution is positive using actual costs and retained commission.
- Delivered value per eligible buyer/order rules out a predeclared material decline, and complaints/wrongful restrictions pass predeclared limits with adequate precision.
- Actual carrier capacity and measured eligible throughput support six-month setup recovery at a conservative measured unit margin.

If feasibility fails, stop/redesign before build. If outcomes are inconclusive, extend or redesign rather than claiming success. If a supporting test is removed, update the budget, allocation and portfolio impact together. The pilot earns a scale decision; no national reach or guaranteed savings are asserted.