The Same-Day Spike and the Seven-Month Tail

Half of ecommerce checkouts happen on the day the shopper first arrives. The other half arrive over the following year, carry more than a quarter of the revenue, and are mostly first-time buyers. This paper measures that distance across 25 stores, shows that the basket in front of the shopper is not what sets it, and explains why every standard instrument stops watching at about the point the second half begins.

First-party, server-side data · 25 stores · 1,214,190 checkouts · 1 June to 31 August 2026

Abstract

The average buyer does not exist

Every ecommerce dashboard reports a conversion rate, and most report an average time to purchase. Ad platforms settle the question by rule: an attribution window, chosen from a short menu and 7-day click by default on the largest platform, decides which orders count. These tools are essential for tuning a campaign inside a single platform. What they cannot tell you is how long your store's actual money takes to arrive after a buyer first shows up, and that question governs which campaigns look like they work. The answer, it turns out, is two populations that need opposite treatment: one needs a faster site today, the other needs to be remembered for seven months.

We measured it directly. For 25 ecommerce storefronts running Edgemesh's server-side analytics, we took every completed checkout over three months (1 June to 31 August 2026) and computed the lag between the buyer's first observed visit and the purchase: 1,214,190 checkouts, identified server-side, so the clock survives the cookie churn that erases the same journey from client-side tools. The paper answers four questions in order:

• How to measure lag. As a distribution, not an average, on an identifier that lives longer than the decision. Half of checkouts happen on day zero; 65% within a week; 12% take more than 90 days; one in twelve comes from a shopper first seen more than seven months earlier.
• What drives it. Price sets the pace between stores but not within one. Across stores, average order value (AOV) tracks speed closely (a correlation of −0.58), yet inside a store the buyers who take weeks spend the same as the buyers who take an afternoon. What price is measuring is the category: how considered the purchase is, and how much of the buying is repeat. One statistical dial explains 91% of the difference between stores.
• Why it is hard to measure. Every default attribution window closes before a third of checkouts arrive, and the browser cookie under most analytics tools dies at seven days, so the late buyer re-enters as a stranger.
• What to do about it. Pull your own curve, turn it into multipliers so a day-one number can be judged against a ninety-day target, and manage the same-day and the slow populations as two different programs. A Q4 2025 cut shows the visit-to-checkout clock holds steady through Black Friday, which makes it the stable reference point when the ad-to-checkout clock stretches.

Bottom line for the busy reader: half of your buyers purchase on the first day, and more than a quarter of your revenue arrives after every default attribution window has closed. Pull your own lag curve before you decide which campaigns are working.

Section 1

The clock your dashboard runs on, and the one your buyers run on

Three clocks govern how most stores read their own conversion data, and none of them belongs to the merchant.

• The platform window. Meta's default attribution setting is a 7-day click, 1-day engaged view, 1-day view; the 28-day click window is no longer available for optimization. TikTok offers 1, 7, 14 or 28-day click windows and defaults to 7. Google Ads defaults to a 30-day click window and allows up to 90.
• The analytics lookback. Google Analytics 4 (GA4) attributes acquisition events over 30 days by default and other key events over 90.
• The browser cookie. Client-side analytics runs on cookies set by script, and Safari's tracking prevention caps a script-set cookie at 7 days of life, 24 hours when the visit arrives with tracking parameters in the link. A visitor who returns in week three is reborn as a brand-new user.

Whatever your buyers are actually doing, the instruments agree to stop watching after about a week, and the most patient of them stops at three months.

Inside that regime, teams manage to summary numbers: a sitewide conversion rate, a blended marketing efficiency ratio (MER), sometimes an "average days to purchase" that lands somewhere between two and five weeks. The average is not wrong, but it's a number that describes almost nobody, because the distribution underneath it is a spike welded to a tail. Half the buyers are at zero. One in twelve is past seven months. Almost nobody is at the average.

The practical consequence shows up in budget meetings. A prospecting campaign is judged at day 7 against buyers who mostly have not bought yet. Retention gets credited for tail buyers that prospecting created, because by the time they convert their origin has been forgotten. Both errors run the same direction: fast channels look better than they are, slow ones look worse, and the store's true clock is never on the table because nothing in the stack can read it.

Section 2

The reframe: lag is a distribution with two populations

This paper works in one primitive: visit-to-checkout lag, the number of days between a buyer's first observed visit to the store and the day they complete checkout. Not session length, not cart-to-checkout time. The full courtship, from first appearance to paid order.

It is worth being precise about how this differs from the number attribution platforms report under the name conversion lag, because the two are close cousins and the paper will compare them. A platform's clock starts at an ad touch and, for click windows, restarts at every click: a 7-day click window asks whether the most recent click on that platform's ads was within a week of the purchase. Our clock starts once, at the first visit from any source, and never restarts. When a shopper's first-ever contact with the store is a click on an ad and they never click another, the two clocks read the same. They diverge in three common cases: when the shopper saw ads (impressions) before the first visit, so the ad clock started earlier; when the shopper clicked again later, so the platform's click clock restarted and now reads shorter; and when the first visit came from somewhere the platform cannot see (search, a friend, an email), so the platform's clock never started at all. We will call these the visit clock and the ad clock, and Section 7 shows where they agree and where they part.

Three regions of the distribution recur at every store we measured:

• The spike: day-zero buyers. They arrive, decide, and pay in one visit or one day.
• The long middle: days 1 through 28. Comparison shoppers, cart abandoners who come back, payday waiters.
• The tail: day 29 onward. The region your tooling treats as noise, and the subject of Section 6.

Pooled across the 25 stores, the cumulative share of checkouts and revenue reads as follows.

Pooled across 25 stores, 1,214,190 completed checkouts, 1 June to 31 August 2026. First-party, server-side data.

Half now, the rest over a year

Cumulative share of checkouts and revenue by days since first visit, 25 stores

CheckoutsRevenue

Read the revenue column against the checkout column. Revenue lags checkouts at every horizon, which means the money is slower than the people. By day 28 you have seen 77% of your eventual checkouts but only 73% of your eventual revenue. Section 6 explains why, and it is not the reason most people assume.

Section 3

The measure, defined precisely, and why it needs a first-party identifier

Everything in this paper is computed the same way. Edgemesh's server sits in front of the store and identifies each device with a first-party, server-set identifier at its first request. For every completed checkout we take:

lag = (date of completed checkout) − (date the device was first seen), in whole days

A cookie set by JavaScript on the page is what client-side analytics and pixels rely on, and Safari caps its life at seven days, regardless of what the script asks for. Server-to-server event delivery (Meta's Conversions API and its equivalents at Google and TikTok) solves a different problem: it gets the checkout event to the platform whether or not the browser blocked the pixel. It does not, by itself, lengthen the identifier that ties the checkout to an earlier visit. That match still runs on browser identifiers (Meta's _fbp and _fbc cookies, for instance), and where those are set by script the seven-day cap applies to them too. A cookie set by the server in an HTTP response is treated differently: browsers honor its stated lifetime up to a cap, currently about 400 days in Chrome. That difference is the difference between an instrument that can see the curve in Figure 1 and one that cannot. In our data, every store's curve ends at week 57, about 400 days. That cutoff is not a fact about shoppers. It is the point where the identifier expires, and it marks the longest lag any identifier on the open web can currently observe.

Three properties of the measure matter for reading the results:

• The observation window is three months, but the lag is unbounded. A buyer checking out in June may have been first seen the previous summer.
• Identity is per device. A buyer who browses on a phone and checks out on a laptop counts as a new, shorter path.
• Any identifier loss restarts the clock at zero. Across the panel, about 14% of orders from known repeat purchasers arrived on a device the identifier had never seen (range 7% to 49%): a new phone, a cleared browser, a second computer.

All three errors shorten measured lag. Every tail figure in this paper is therefore a floor.

This is the visit clock of Section 2. The ad clock has its own careful literature; Northbeam's Bryan Bumgardner has written a clear treatment of it and of why it stretches ahead of Black Friday, as shoppers see the ads and wait for the sale ("Why conversion lag can make or break your BFCM," The Media Buyer, September 2026). Where the ad clock starts at a first touch that precedes the first visit, the two are nested: ad-to-checkout lag is ad-to-visit lag plus visit-to-checkout lag, so at any horizon the share of checkouts inside the window is at least as high on the visit clock, and a catch-up multiplier computed on the ad clock should be at least as large as one computed on the visit clock. Where the ad clock restarts at a later click, it reads shorter than ours, not longer. Section 7 uses both facts.

We split each store's checkouts into first purchases and repeat purchases, which lets Section 6 test whether the tail is a loyalty artifact. Any Edgemesh customer can pull the same curve for their own store: completed checkouts by days since first seen, split by returning purchaser.

Section 4

Half the store converts today

Start with the spike, because it is the part everyone's instruments do see, and it is real: 49.8% of checkouts complete on the same day the device is first observed. Another 5.7% convert the next day. By the end of the first week, 65% of all checkouts in the panel are done.

The spike is nearly universal. Same-day share is above 43% at 22 of the 25 stores. A pet brand with orders under $100 converts 66% of its checkouts on day zero. An outdoor equipment store with orders in the $250 to $500 band converts 60%. Even the slowest stores, two apparel brands with orders over $500, convert a third of their eventual buyers on the first day.

This is the half of the distribution where site performance is the whole game. A buyer deciding in one session gives you no second chance at a slow page or a stalled checkout. We have made that argument at length elsewhere; here it is one sentence: the spike is won or lost in milliseconds.

The mistake is mistaking the spike for the entire picture.

Section 5

Same conversion rate, different clock: the cross-store spread

Two terms, used consistently from here on. A store is one merchant. A vertical is the product category we group stores by (pet, jewelry, preparedness), and when the paper says the clock belongs to the category, the vertical is what it means.

Line the 25 stores up by same-day share and the spread is 2×: from 33% at the slow end to 68% at the fast end. The share of checkouts arriving more than four weeks after first visit runs from 13% to 45%. Figure 2 draws the cumulative curve for each vertical, pooling its stores; the table beneath it gives the same picture as numbers. Where a vertical contains more than one store, each cell shows the lowest and highest value among its stores rather than a single average, so "44% to 61%" means the fastest supplement store converts 61% of checkouts on day zero and the slowest 44%.

Every vertical has the same shape and a different pace

Cumulative share of checkouts by weeks since first visit, by vertical

All storesFastest / slowest verticalsOther verticals

Each vertical pools its stores. Week 0 is the first seven days after the first visit.

Visit-to-checkout lag by vertical, sorted fastest to slowest. Where a vertical has more than one store, the cell shows the range from its fastest store to its slowest. Completed checkouts 1 June to 31 August 2026.

The ordering follows vertical lines. Grab-and-go products sit at the top: wellness, pet, household, drinkware. Considered and stock-up verticals sit at the bottom: jewelry, bags, and preparedness, where a buyer may watch the news for a season before committing. Apparel spans almost the whole panel on its own, from 62% same-day at a store in the $100 to $250 band to 33% at the two stores over $500, which is the first sign that price positioning does something inside a vertical as well as between them (Section 6). The two preparedness stores offer different products at different price points, and they land within three points of each other on every column, as they did in an earlier panel. The clock belongs to the category, not the brand.

Beyond the ordering, the curves have shapes, and the shapes are legible without any product data.

After the first week, the curves diverge

Weekly share of checkouts, weeks 1–12 since first visit, eight stores

ReplenishmentOther purchase cycles

Wellness

Week 0 79%Beyond 12 weeks 6%

Pet

Week 0 76%Beyond 12 weeks 10%

Supplements

Week 0 55%Beyond 12 weeks 19%

Outdoor equipment

Week 0 75%Beyond 12 weeks 5%

Kitchen

Week 0 64%Beyond 12 weeks 10%

Jewelry

Week 0 48%Beyond 12 weeks 20%

Preparedness

Week 0 55%Beyond 12 weeks 23%

Apparel

Week 0 43%Beyond 12 weeks 30%

• Replenishment rhythm. The pet brand has a second peak at day 30, its fourth-largest single day in the whole distribution. One supplement brand peaks again at day 25, another at day 28, a third at days 13 to 14 and again at weeks 4 and 6. The reorder cycle is written into the lag curve.
• Seasonal return. One drinkware store has a bump at weeks 49 to 56: shoppers first seen during last year's back-to-school season, buying again this year. One of the apparel stores over $500 has a cluster at weeks 36 to 43, last fall's shoppers returning. These are people the identifier held for most of a year.
• Slow consideration. The two apparel stores over $500 and both preparedness stores have the fattest tails, with a third or more of checkouts from shoppers first seen more than four weeks earlier and no rhythm to them at all. They're deciding, not reordering.

The conclusion to draw from the shapes is practical rather than statistical. Section 6 will show that the rhythm accounts for less than 1% of the difference between stores, so it does not change which store is fast or slow. What it changes is when a given store's late buyers arrive: at day 30 for the pet brand, at weeks 49 to 56 for the drinkware store, spread evenly across four months for the apparel and preparedness stores. Suppression rules, retargeting depth and retention timing are set against exactly that calendar, and no panel average can supply it. The shape is the part of the curve only your own data can give you.

Section 6

Price sets the pace between stores, not within one

The intuition is that an expensive cart takes longer to decide, and across stores the data agrees. The correlation between a store's log AOV and its same-day share is −0.58, a tight relationship for a panel of 25 businesses in fourteen verticals: the fit line in Figure 4 runs from roughly 58% same-day for stores with orders under $60 to under 40% for stores over $700, and most stores sit close to it. Price is a real driver of the clock. The question is what kind of driver, and the exceptions give the first hint: an outdoor equipment store with orders in the $250 to $500 band is one of the fastest in the panel, and a supplement brand with orders under $100 is one of the slowest.

Price pulls the same-day share down across stores

Same-day share of checkouts against average order value, 25 stores

Correlation with log average order value: −0.58

The decisive test is inside a single store, where the vertical is held constant and only the basket varies. If it were the price of the basket that slowed the decision, orders placed weeks after the first visit should be larger than same-day orders from the same store. They are not. Comparing each store's orders in weeks 1 to 12 with its own week-0 orders, the median ratio is 1.04×, and 22 of the 25 stores sit between 0.95× and 1.15×.

Inside a store, late buyers spend the same

Order value in weeks 1–12 relative to week 0, one dot per store

1.04× median within a store22 of 25 stores in the shaded band1.24× when pooled

Outside the band: wellness (0.85×), preparedness (1.16×), and outdoor equipment (1.20×). Ratios shown to two decimal places.

That pooled figure is the trap. Pool the 25 stores and average order value climbs from $176 for same-week checkouts to $254 for checkouts in months four to six, a 44% rise that looks like proof that slow buyers buy big. It is composition: the verticals with long clocks (preparedness, high-end apparel, jewelry) also carry high order values, so the tail fills up with their buyers. Pool a panel and you manufacture a trend ten times stronger than the one any single merchant experiences. This is also why revenue trails checkouts in Figure 1: not because a given store's late buyers spend more, but because the slow stores are the expensive ones. If you benchmark your own AOV-by-lag curve against a pooled industry chart, you are reading other people's vertical mix, not your buyers.

So price works at the level of the store, not the basket: it is a marker of how considered a store's vertical is, not a brake on the individual shopper. What else sets the clock? We asked the data directly, using a technique that finds the fewest underlying "dials" needed to reproduce a set of curves (principal component analysis, or "PCA"). The answer is one dial: a single component, which trades same-day checkouts against the tail and touches nothing in between, explains 91% of the difference between the 25 curves. The second component, which shifts late buyers between the first month and the far tail, explains 9%. The replenishment rhythm visible in Figure 3, though real, accounts for less than 1%.

Lag is one dial

First principal component of the 25 lag curves: what moves as a store gets slower

91% of the difference explained by this component9% explained by the next

And what moves that dial? Three store attributes, ordered by the strength of the relationship (the size of the correlation, not its sign):

Correlation of each attribute with each store's score on the first principal component, 25 stores.

The repeat mix comes first, how quickly a store's first-time buyers decide comes second, and price is third, real but weaker than either. Store size does not matter at all. Lag is a property of a store's purchase cycle; the cycle is largely set by its vertical, and price is one of the things the cycle shows up in.

Two more facts complete the picture. First, the tail is mostly strangers. About a third of the checkouts arriving more than 28 days after first visit are repeat purchases; two thirds are first purchases. Among first purchases alone, 30% happen more than a week after the first visit and 17.5% more than four weeks after. The tail is not your loyalty program talking. It is people who found you months ago, thought about it, and bought once. Second, device does not matter: at the one paid-social-heavy store where we checked, mobile shoppers converted within the first month 80% of the time and desktop shoppers 84%.

The tail is a fifth of your checkouts, more than a quarter of your revenue, and mostly first-time customers moving at the speed of their vertical. They are the most expensive buyers to misattribute, because their trails are the coldest.

Section 7

What the windows can and cannot see

If a fifth of checkouts and a quarter of revenue sit past day 28, why does no dashboard say so? Because every common instrument is structurally blind there. Read Figure 1 against the windows:

Share of checkouts arriving inside each window, measured from first visit. Platform defaults as documented in September 2026.

One important caveat: a platform window restarts on every click. A shopper who clicks another ad on day 20 is inside a 7-day window at purchase, so the curve above overstates what the window misses for shoppers who keep clicking ads, and states it exactly for shoppers who come back on their own: direct, organic, a search for your brand name. The window cannot tell you which case you are in, and in the second case the credit goes to "direct," or to whatever ad happened to be viewed in the final day.

The deeper problem sits underneath the window. The 90-day lookbacks in analytics tools help, but 12% of checkouts in this panel are past 90 days, and the identity underneath most tools is a script-set cookie that Safari expires at seven days. Under that regime the curve in Figure 1 is not merely truncated; the buyer past day 7 re-enters as a new user, their eventual purchase is credited to the last touch before it, and their original acquisition is written out of the story.

The same curve through a seven-day cookie

Cumulative checkouts by lag, with the first seven days highlighted

Full lag curveFirst seven days

You cannot fix a window problem with a longer window if the clock resets underneath it. The precondition for measuring lag at all is an identifier that outlives the decision, which in practice means first-party and set by the server. That is what makes the curve in this paper possible, and it is the reason the curve ends at 400 days rather than at 7.

The visit clock and the ad clock in Q4. The windows matter most in the quarter when the most money moves through them, and this is where the two clocks defined in Section 2 earn their keep. Northbeam's data shows the ad clock stretching as Black Friday approaches: for one apparel brand, the gap between day-one and day-14 Facebook revenue grew from 25% in October to 29% in early November and 36% to 37% in Black Friday week itself, as shoppers saw the ads, then waited for the sale. That is a real effect and a serious planning problem, because a day-one ROAS read in mid-November is measuring a smaller fraction of what the campaign will eventually return than the same read in July.

We ran the visit clock over the same season. For 21 stores in this panel with clean Q4 2025 data, Figure 8 tracks the share of each week's checkouts that came from devices first seen that same week. It does not stretch. Through October the same-week share sits at 64% to 65% of all checkouts, in line with the summer baseline of 65% within a week. In Black Friday week it is 68%. Through the first three weeks of December it is 69% to 70%, the fastest the panel gets all year, before falling to 62% in the last two weeks as holiday shopping ends. Store by store, Black Friday week moved between eleven points slower and seven points faster than the store's own October pace, and most stores stayed within five points; the typical store was essentially unchanged (66% in October, 66% in Black Friday week).

The visit clock does not stretch into Black Friday

Share of checkouts from devices first seen the same week, 21 stores, Q4 2025

All checkouts (large stores weigh more)Median store (each store counted once)

Shaded horizontal band: October’s 64–65% same-week share, near the summer 2026 baseline of 65% within a week.

Is Northbeam wrong, then? No, and the point deserves to be made carefully, because the two measurements are of different things and neither can check the other directly. Theirs is first-time revenue on one platform, timed from the first ad touch, for one brand. Ours is every checkout on every channel, timed from the first visit, for 21 stores. Where the two overlap, they agree: the four apparel stores in the Q4 cut moved between five points slower and five points faster in Black Friday week, which is consistent with a brand-specific stretch of the kind Northbeam reports and says nothing against it. Where they differ, the difference is exactly the gap between the clocks. If the ad clock lengthens into Black Friday and the visit clock does not, then the waiting happens between the ad and the visit: the shopper sees the campaign, holds off, and arrives at the store once the sale is live, then buys at the usual pace or faster. Northbeam's beauty example points the same way: same-day touchpoints fell from 81% of November's new revenue to 76% on Black Friday on the ad clock, while the same-week share on the visit clock rose. The pre-sale hesitation that attribution platforms correctly observe is not, on this evidence, hesitation on the site. We call this a reconciliation rather than a finding, because it has one test we cannot run from our side alone: a store measured on both clocks through the same Q4. We would welcome one.

What the merchant can take from it now is a division of labor. The ad clock is the one to use for judging a campaign, and in Q4 it must be re-estimated week by week, as Northbeam recommends. The visit clock is the one to use for the site itself, for suppression and retention timing, and for a ceiling on how much of the eventual revenue any window can possibly have seen; in Q4 it can be planned from the previous quarter because it does not move.

Section 8

What to do about it

Measurement is the prerequisite; the operating changes follow from your own curve, not from the panel's.

• Pull the lag distribution for your store. The store range in Section 5 is 2×, so the fleet number is not your number. The percentiles in this paper are evidence that yours is worth pulling, not a substitute for it.
• Judge prospecting at your own horizon, not day 7. For the faster half of the panel the 90th-percentile buyer arrives five to thirteen weeks after first visit; for the slower half, four months to more than a year. A campaign at a store whose beyond-7-day share is 44% has shown you a little more than half of its eventual buyers when the default window closes. Kill decisions made in week one are made on the spike alone.
• Read every reported ROAS next to your coverage number. A 7-day return on ad spend (ROAS) on a store where 44% of checkouts arrive later is reporting on half the business. Put the coverage percentage beside it and the comparison across campaigns, channels and stores becomes honest. The buyers still inside the window are not lost, they are in progress; Marking the Funnel to Market shows how to carry them on the books at what they are worth today.
• Watch the same-week share as a guardrail. The single most useful number to put on a weekly dashboard is the share of this week's checkouts that came from devices first seen this week. It is not the same as the share of first-time customers: a repeat buyer on a new phone is a new device, and about one repeat order in seven arrives that way (Section 3). It is the fast population's weight in the mix, and when it moves the cause is worth finding. A drop means more of this week's buyers came in earlier (a sale being anticipated, a new channel with a longer clock); a rise means the reverse. Northbeam uses the same guardrail on the ad clock. Figure 8 gives the panel's version on the visit clock: 64% to 70% through Q4, a range narrow enough that a store leaving it has something to explain.
• Set remarketing and suppression from the curve, not from a template. A store whose long middle runs four weeks wide is wasting suppression rules built for a seven-day cycle, and a preparedness store that retargets only 14 days deep abandons a third of its buyers mid-decision.
• Time retention to the rhythm the curve reveals. The supplement and pet stores' second peaks at days 25 to 30 are their reorder cycle, measured rather than guessed. Email and text that lands on day 22 is on time; on day 40 it is late.
• Treat day-zero performance as its own program. Half your buyers decide in the first session; page speed and checkout friction are that half's whole experience.
• Fix the identifier first. None of the above is possible on a seven-day cookie. Server-side, first-party identity is the instrument; everything else is what you do with the reading.

From the curve to a multiplier. The media-buying version of the coverage number is a catch-up multiplier: revenue inside a long window divided by revenue inside a short one. Divide the target you care about by the multiplier and you have the day-one number a campaign must hit today to reach that target eventually, so a week-old campaign can be judged without waiting ninety days. Northbeam recommends exactly this on the ad clock, re-estimated by channel and by season, and we agree with the method. On the visit clock, the pooled panel reads as follows:

Pooled across 25 stores, June to August 2026; a store's own multipliers will differ, and the slowest stores run well above these.

Two cautions. Where the ad clock starts before the first visit, these are floors for the ad-side multiplier (Section 3). And they are stable across seasons on the visit clock (Figure 8) but not on the ad clock, so the ad-side multipliers must be refreshed going into Q4 while the site-side figures can be carried forward.

Of course, patience is organizationally expensive. Holding a budget line open for ninety days on cohort evidence is a harder meeting than cutting it on a week of MER. The stores that manage the tail are the ones whose measurement lets them see it at all, which is why this is a data problem before it is a media problem.

Section 9

Conclusion: manage both populations, and stop averaging them

Visit-to-checkout lag in ecommerce is a spike welded to a tail. The spike is half your buyers and it converts in hours; the tail is a fifth of your buyers, more than a quarter of your revenue, and mostly first-time customers moving at the speed of their category. The basket in front of the shopper does not set that speed; the category's purchase cycle does, price included, and one dial describes it. Every summary statistic your tools volunteer, and every attribution window they impose, is an average over two populations that need opposite treatment: one needs a faster site today, the other needs to be remembered for seven months.

Your store converts twice: once in the first afternoon, and once over the next two quarters. Only one of those shows up in your dashboard.

There are two clocks worth running. The ad clock, which attribution platforms such as Northbeam measure well, tells you what a campaign will eventually return and, on their evidence, must be re-read every season because shoppers wait for sales. The visit clock in this paper tells you what the site does once they arrive, and on our evidence it holds through Black Friday. We could not check their measurement with ours or ours with theirs, because they time different things; what we can say is that the two are consistent, and that merchants who read both will find they agree on the important thing: the day-one number is a fraction of the truth, and the fraction is knowable.

This paper measures when the money arrives. Pricing the Funnel with Marginal Stage Value measures where in the funnel it is created, and Marking the Funnel to Market shows how to value the carts that have not closed yet. Together they are an argument for the same practice: read your own first-party distribution before renting a platform's summary of it.

About the data. Completed checkouts from 25 US ecommerce storefronts running Edgemesh server-side analytics, 1 June to 31 August 2026: 1,214,190 checkouts. Lag is the whole-day difference between a device's first server-side identification and its completed checkout; shares are pooled across stores unless labeled per store, and medians across stores are reported where one large store would otherwise dominate (one drinkware store is a quarter of all checkouts; excluding it moves the beyond-7-day share from 35.0% to 37.5%). First versus repeat splits use week-grain data at all 25 stores. Merchants are identified by vertical and average order value band only; the vertical curves in Figure 2 pool the checkouts of each vertical's stores, and reported as ranges where a vertical contains more than one store; the correlations and principal component analysis in Section 6 are computed on the 25 individual stores. Limits: identity is per device, so cross-device paths and any identifier loss restart the clock; the identifier's lifetime of about 400 days bounds the tail; stores with young tracking installs cannot observe long lags and two were excluded on that basis; lag is measured from first site visit, not first ad exposure, so it understates ad-to-purchase lag for shoppers who saw ads before arriving. The Q4 2025 cut in Figure 8 uses week-grain data (checkouts by weeks since first identification, by calendar week) for the 21 panel stores whose tracking was installed before 29 September 2025; two stores were excluded because their first Q4 week was the install week, and two others had no October 2025 data to serve as a base. All of these truncate the tail, so the figures reported here are floors. The panel is not a random sample of ecommerce; it is the set of stores we can measure this precisely. Platform window defaults are as documented by Meta, TikTok, Google Ads and Google Analytics in September 2026.