How long does an EU trader have to deliver? Thirty days from the contract, by default, under Article 18 of the Consumer Rights Directive, and without undue delay within that. But the default rarely governs, because almost every product page displaces it by making a promise: delivered in 2 to 4 days, ships within 24 hours, at your door by Friday. The promise becomes the agreed term. It also becomes something else that matters more to a buyer of the business: a conversion lever with a cost of goods nobody accounts for.
Outside-in screening rests on one premise: what customers report can be checked against what the seller claims. Delivery is where that check is most literal. On one side, a written promise with a number in it, published by the seller at the moment of sale. On the other, thousands of timestamped accounts of what actually arrived when. No other operational metric offers outside-in analysis a more literal like-for-like check.
The Promise Is a Conversion Lever
Stated delivery speed lifts checkout completion. The effect shows up wherever checkout testing looks for it, which is why promise inflation is a temptation rather than an accident. The mechanics resemble the manufactured signals of the GP3 series more than anyone in a fulfilment team would like. An optimistic delivery window does the same work at the checkout as an inflated reference price or a curated review score. It moves a hesitating customer over the line. The difference is that this signal generates a measurable operational debt at the moment it converts, because a promise made to a thousand customers a day is a distribution of outcomes, and the tail of that distribution has legal rights and a refund claim.
Two structurally different targets emerge:
- The honest-promise operator states a window its logistics reliably beat, converts somewhat less, and carries almost no promise-driven cost.
- The inflated-promise operator states the window its checkout tests said would convert best, ships to a slower reality, and pays for the difference downstream, in cost lines nobody connects to the product page.
Both show a delivery promise on the site and a fulfilment cost line in the P&L. The connection between the two appears nowhere.
Where the Gap Lands in GP2
A missed promise is not one cost; it is a cascade with four stages, each in a different line:
- Support contact. “Where is my order” is routinely the highest-volume inquiry class in e-commerce support. Its volume scales with the promise-performance gap, and it lands in the customer-service cost the chatbot piece examined, which means a target suppressing support access is also suppressing the clearest internal record of its own delivery problem.
- Statutory remedies. Under Article 18, a consumer facing a missed term can set an additional period, and on its failure terminate for a full refund, reimbursed without undue delay. A refund on a shipped order is the most expensive single outcome in fulfilment: the goods travel, the money returns, and the reverse logistics from the return-fee analysis run in the wrong direction at the seller’s expense.
- Chargebacks. Customers who cannot reach support do not stop; they call their card issuer. “Item not received” disputes carry fees, and a rising dispute ratio endangers the payment-processing relationship itself, a dependency with the same suspension-first dynamics as the marketplace layer. Marketplace channels add their own version: late-shipment metrics feed directly into account health.
- The repeat-purchase hole. A customer whose first order arrived a week late is measurably less likely to place a second. This cost never appears in GP2 at all; it appears as a cohort curve the acquirer will model as brand weakness.
The first three are GP2. The promise that caused them was earning its keep in GP3 the whole time. That split is why nobody owns the gap: the team paid on conversion sets the promise, the team paid on cost absorbs the outcomes, and outside a well-run minority that wires on-time delivery into its cost reviews, the P&L never introduces them to each other. Whether a target belongs to that minority is itself worth one question in the first meeting.
The Objection to Deal With First
Every experienced operator will raise the same defence, and it is half right. Delivery complaints are the most over-represented complaint class in reviews. The customer whose parcel arrived on time says nothing; the customer whose parcel was late writes a paragraph. Raw complaint counts therefore prove almost nothing, and an analysis waving screenshots of angry reviews deserves the scepticism it gets.
The honest method is relative, on two axes. First, relative to the promise: a review saying “took two weeks” is noise against a 30-day default and evidence against a “2 to 4 days” product page, and only the comparison gives the words meaning. Second, relative to peers. Complaint rates for comparable catalogues shipping comparable lanes form the complaint-rate baseline, built per screen from the public review sets of those comparables rather than from any proprietary index, and what matters is the target’s deviation from it, not its distance from zero. Selection bias inflates everyone’s delivery complaints roughly alike; it does not explain why one seller’s rate runs at a multiple of its category’s, under a faster promise. The denominator deserves the same scrutiny, because a review base diluted by aggressive solicitation, or manufactured outright, lowers any complaint rate, which is one more reason those two analyses belong in the same screen.
What Outside-In Analysis Can Detect Before the Data Room
- The promise itself, recorded per market and per channel: what the product page, checkout, and marketplace listings actually commit to, which frequently differ from each other, and where the fine print quietly resets expectations the headline set
- Review timestamps against order language: reviewers routinely state elapsed time (“ordered on the 3rd, arrived on the 19th”), and enough of them build a distribution the seller never published, windowed to the promise in force when the order was placed, since promises change and archived product pages say when
- The complaint-rate comparison described above, against both the promise and category peers, with the selection-bias correction applied symmetrically
- Marketplace seller metrics where visible: late-shipment and dispute indicators surface in seller profiles on several platforms, an independently maintained scorecard of exactly this gap
- The dispute trail: “item never arrived,” “had to claim through PayPal,” “bank refunded me” in review language, each one a chargeback that also happened somewhere in the payment data the data room will show
- Whether the promise flexes honestly: sellers with real delivery visibility show different windows by destination and stock state; a single confident number for every SKU to every country is itself an indication of how the number was chosen
As always, these are indications rather than conclusions. A promise-performance gap sized from outside scopes the confirmatory work, and the confirmatory ask is precise, since carrier scan data and refund reason codes exist in every fulfilment stack and are rarely requested.
The Pre-LOI Question Every PE Fund Should Ask
Fulfilment cost per order will be in the pack, benchmarked and tidy. The number that connects it to revenue quality will not. So ask this: what delivery window is the checkout selling, what window do the timestamps show it keeping, and which of the two paid for the conversion rate in the deck?
Where the promise and the performance match, the conversion is bought honestly and the fulfilment line is what it claims to be. Where they diverge, the target has been running a quiet loan from GP2 to GP3, conversion borrowed now and repaid in refunds, disputes, and cohorts that never returned, and the acquirer inherits both sides of the balance.
This analysis is part of Tronvik’s GP2 Fulfilment & Service Margin pillar. Nothing in this article constitutes legal advice. To initiate an outside-in delivery-performance screen on a specific acquisition target, contact info@tronvik.com.