More of your best customers.

A cohort is a group of buyers tracked together over time. When we can trace a buyer's first order to a real ad click, that group is the ad set or campaign that brought them in — the more useful grouping, since it tells you where to spend. Only when there's no ad-click history do we fall back to grouping by the month a buyer joined, similar to a class of kids who all started school on the same day. The Diagnosis chart (the lines climbing over months) shows a cohort's profit building over time. The quadrant view (the four boxes) shows the same cohorts plotted by LTGP:CAC against buy-again rate, so you can see at a glance which groups fund the business and which drain it — and a group can move between boxes as its numbers change.

Shopify's own reports are revenue reports: what sold, and for how much. They don't know what a product actually cost you, so there's no contribution margin, only top-line sales. They don't follow a customer across months, so there's no cohort or LTV view — no way to see that a channel looks cheap on the first order and expensive a year in, or the other way around. And they tell you a number moved, not why. Upstream sits one layer up: real margin after COGS, an estimated shipping cost, and returns; LTGP:CAC by cohort, not blended ROAS; and a Data Story Feed that names the actual reason a number moved, in plain English, refreshed every night. None of that is a report you could save from Shopify's own analytics.

Not by ROAS (Return On Ad Spending), but by contribution margin. We take every order an “ad click” led to, subtract real product cost, an estimated shipping cost, and returns, and match that against what you actually paid for the click. That's the number that matters, not blended revenue. On top of that, we track it by cohort: a campaign can look unprofitable on the first order and still be your best channel if those buyers come back and spend 3x more over the next six months. That's the whole point of the cohort engine. And you don't have to go looking for it: the “Data Story Feed” flags a campaign automatically the moment its true profit crosses below a healthy line, instead of waiting for you to notice on a dashboard. One thing we deliberately don't do: build a proprietary multi-touch attribution pixel. That's a different, much bigger engineering bet, and it doesn't change the core answer; Which is real cost matched to real orders, not modeled credit across touchpoints.

Every ad click is linked to the real Shopify order it led to, including what the product cost, an estimated shipping cost, and refunds. We add up the lifetime profit from each buyer. We track profit on the first order, profit over 90 days, how long it takes to earn back the ad cost, and how often each group buys again. For every channel and campaign. No made-up 'revenue per ad dollar' numbers from the ad platform: only the real, server-checked profit each buyer actually brought in.

We add a small piece of code to your Shopify theme (an app-embed block) that runs at the edge and saves the ad-click IDs before any browser script can remove them. On our servers, we listen for Shopify's 'new order' notice and join the order: line items, discounts, refunds, etc. With the ad click that started the visit. With your permission, we pull each ad platform's daily spend at the campaign level. No theme code edits, no heavy install.

Yes. Paste in a private API key from your own Klaviyo account — no developer registration, no waiting on anyone — and Upstream writes real LTGP:CAC, payback status, and repeat-purchase signals onto each customer profile, plus one segmentable property for which quadrant they're in: compounding, one-hit, trap, or drain. It also reads back your existing Klaviyo segments and shows the overlap against our own margin-based ranking — where the two agree, and where Klaviyo's revenue-ranked view and our margin-ranked view genuinely disagree. Refreshed nightly, with partial failures reported rather than hidden behind a green checkmark.

No. Because Upstream talks to Shopify's servers directly, the data moves under the browser, where Safari's privacy rules and ad blockers can't see or stop it.

We store the ad-click ID, the matched ad click, and the matching Shopify order details (order ID, profit, product cost, refund status). We never store raw payment details. Shopify and your payment processor handle that. All data is encrypted when stored and when sent.

Yes. You can export or permanently delete all stored ad-click and order data any time from the Upstream settings page. Deletion finishes across our systems within one business day and removes the link between ad clicks and past orders.

You cancel from inside the app — no email, no call, no support ticket. Downgrading to Free or disconnecting your store stops the Shopify subscription at that moment; we don't keep billing an app you've turned off. We don't credit you for the unused days left on the period you already paid for, and we don't claim otherwise. Whatever the button says is what happens, stated before you click it, not buried in a policy page you have to go find.

No, and we label it that way inside the app rather than letting you assume otherwise. The quadrant view and the LTGP:CAC ratios match real profit to real ad spend, which is honest arithmetic, but the link between a customer and a channel is attribution, and attribution is observational. Ad platforms deliberately optimize toward people who were already likely to buy, so a channel can look like it produces excellent customers while producing very few customers you would not have got anyway. Both of those can be true at once, and the attributed number cannot tell them apart. Better matching, longer lookback windows and server-side tracking make the association more precise; none of them make it causal. What would make it causal: withholding that channel from part of your market and comparing those regions against the rest of your store over the same weeks. That is a geo holdout, and it is the only thing in Upstream that produces a causal number. Every other figure carries a badge saying which kind of claim it is making: causal, observed, projection, or measured.

You switch your ads off in a set of states and leave them running everywhere else, then compare the two groups over the same weeks. The rest of your store acts as the control group, so anything that hit the whole business (a season, a competitor, a viral post, a supply problem) hits both sides and cancels out. That is what makes the answer causal rather than a before-and-after: a before-and-after has nothing held back, so seasonality and plain luck are indistinguishable from the effect of whatever you changed. Upstream picks the regions, tells you how long to run it, computes whether the test could detect anything before you start, and analyzes the result from your Shopify order data afterwards. You make the actual exclusion in your ad account. You need one when the question you are asking is 'would these sales have happened anyway', and no dashboard, however detailed, can answer that.

Possibly not, and we would rather tell you that up front than take eight weeks of held-back revenue to get there. For a store around $1M a year, the smallest change a holdout can reliably detect is roughly 18% to 44% of the held-out regions' sales, depending on how much you hold out and for how long. Now compare that to how far a real change actually moves those regions: switching off all paid media there moves them about a third, so that test is just about answerable. Switching off one channel moves them about 13%, and trimming a budget by 20% moves them under 3%. Neither of those is answerable at that size, and no amount of patience or extra weeks fixes it. Single-channel questions become answerable somewhere around $3M a year, and budget-level questions need far more. This matters because underpowered tests do not fail quietly: they hand back confident, inflated numbers that read like answers, and acting on a noisy result has been shown to leave a business worse off than doing nothing a substantial share of the time. Upstream computes your own answer from your own order history, by running the same comparison over past stretches where nothing happened and measuring how far from zero it lands anyway, so the verdict is about your store rather than a rule of thumb.

Judge it on whether the method is published, because that is the part you can actually check. Some vendors do publish a real one, run genuine controlled experiments, and are worth every dollar for the brands they are built for: those brands typically spend seven figures a month on media, and the pricing reflects it. Others use the language of lift and control groups for something built from observational data, and never publish the design, the power calculation, or what happens when the test cannot detect anything. Two questions separate them. Does the tool tell you, before you run, the smallest change it could reliably detect on your data? And does it ever refuse to give you a number? Upstream answers both out loud, publishes the arithmetic, labels its own observed figures as observed rather than only pointing that out about everyone else, and tells most stores our size that a meaningful holdout is out of reach for them. That last one is not a limitation we are hiding; it is the main thing the feature does.

Never automatically, and never in the background. We read your ad spend (read-only, with your permission) to match the exact cost against real profit — that never goes back to the ad platforms. The one exception: the Gateway SKU Bridge can push a list of your best buyers' emails to Meta or Google to build a lookalike audience, and that only happens when you explicitly click Apply on that specific action. Nothing else about your order or customer data is ever shared with an ad platform.