Rising CAC: Find the Broken Funnel Stage

Your CAC went up 27% between June and July 2026 and nobody can tell you why. The media buyer says CPMs are up. The site team says conversion looks fine. Both of them are reading one number each, and the argument goes nowhere, because CAC is not a thing you can pull on. It is the output of six things that each move on their own. This matters because until you know which of the six moved, every fix you ship is a guess. Here is how to find the stage that broke, what to expect when you fix it, and what to watch the following week.

CAC is not a lever. It is six numbers multiplied together.

Write it out as arithmetic and the argument ends:

CAC = CPM / (1,000 x CTR x add-to-cart rate x reach-checkout rate x checkout close rate x new-customer share)

Read left to right, that is: what you pay to put an ad in front of a thousand people, then the share of those people who land on your site, then the share of those who add something to a cart, then the share of carts that make it to checkout, then the share of checkouts that complete, then the share of resulting orders that came from someone new.

Six terms. Every one of them is a number you already have in Shopify and your ad platform. Every one of them moves independently. And CAC, the number everyone argues about, is just what falls out the bottom.

The reason this framing matters is that it converts an argument into a lookup. Nobody has to have an opinion about whether "the site is broken" or "the platform got expensive." You divide six pairs of numbers and the answer is sitting there.

Here is what that looks like for a brand spending roughly $25,000 a week, comparing its first week to its thirteenth.

Funnel stage	Week 1	Week 13	Change
CPM	$25.00	$27.45	+9.8%
Effective CTR	4.38%	4.23%	-3.4%
Add-to-cart rate	7.00%	7.00%	0.0%
Reach-checkout rate	72.0%	48.0%	-33.4%
Checkout close rate	32.0%	32.2%	+0.6%
New-customer share	62.0%	59.0%	-4.8%
CAC	$57.14	$101.82	+78.2%

CAC nearly doubled. Five of the six stages barely moved. One fell by a third.

Check the instrument before you read the dial

Before any of this is worth doing, the numbers have to be real, and often they are not.

The tell is simple: any stage rate above 100%. If your reach-checkout rate prints at 143%, more people started checkout than added to cart, which is impossible. What it actually means is that the numerator and the denominator came from different measurement systems, with different definitions of a session, different attribution windows, or a tag that fires twice.

This is the oldest rule in experimentation. Ron Kohavi and Roger Longbotham, who ran controlled experiments at Microsoft at large scale, open their paper on unexpected experiment results by quoting Twyman's law: any figure that looks interesting or different is usually wrong. One of the cases they walk through is a treatment that looked like it was losing clicks, until the investigation found that the monitoring system's own click action had stopped working and was retrying, depressing the measured click-through rate. The finding was about the instrument, not the users.

The pattern we see again and again is a funnel sheet where one stage has been printing above 100% for months and nobody flinches, because the number lives in a tab nobody opens. Every diagnosis run on top of that sheet was noise. Fix the instrument first. It usually takes an afternoon and it is the highest-return afternoon in the whole exercise.

Once the counts are trustworthy, one more rule keeps you honest: under plus or minus 5% week on week is usually noise, especially below 500 orders a week. Investigate at 10% or more, or when the same direction shows up three weeks running.

Two stages held. One collapsed.

Now look at the same 13 weeks as a picture rather than a table.

Add-to-cart rate is a flat line at 7%. Checkout close rate is a flat line at 32%. Reach-checkout rate walks down from 72% to 48% over three months, and drags CAC from $57 to $102 behind it.

This is the shape that fools teams, and it fools them in a specific way. Site-wide conversion rate is an average across all three site stages, so a collapse in one gets diluted by two that are fine. Overall CVR in this example went from 1.61% to 1.08%, which reads as "conversion is down a bit, we should do some CRO." That framing sends a team off to A/B test button colours on a product page where nothing is wrong.

Meanwhile the actual damage is happening between the cart and the checkout, which is the one stage almost nobody has a dashboard for. Most reporting jumps straight from add-to-cart to purchase. That stage has no owner, no alert, and no weekly review, which is precisely why it is so often the one that breaks.

When I talk to founders running brands at this size, the thing they keep saying is that they have too many dashboards and still cannot answer the one question that matters. That is not a tooling problem. It is a decomposition problem.

Rank the stages by dollars, not by percentage

A 33% fall in one stage and a 10% rise in another are not comparable, because the stages do not carry equal weight in the arithmetic. The only fair comparison is money.

The method is a counterfactual. Take each stage, put it back to where it was four weeks ago, hold every other stage exactly where it is now, and recompute CAC. The difference is what that stage alone is costing you. Because CAC is a straight product of the six terms, this is just multiplication: a stage that fell to 83% of its old value multiplies your CAC by 1/0.83.

Over the last four weeks against the prior four, CAC went from $73.14 to $93.08. Reach-checkout rate accounts for $15.87 of that $19.94. Media cost accounts for $3.15. Everything else is under $1.10, and add-to-cart rate is actually negative, meaning it improved slightly and is helping.

That $3.15 of CPM damage is worth pausing on, because it is where most of the meeting time goes. Meta's own first quarter 2026 results put average price per ad up 12% year-over-year, against ad impressions up 19%. If your CPM is up roughly in that band, you are tracking the platform, not failing. It is real money and it is not nothing, but it is not your constraint, and it is largely not yours to fix. Absorb it and move on to the stage that is five times larger.

The discipline is not "look at more data." It is being able to say out loud, on a Monday morning, the single number that is capping the business right now, and then spending the week on that number and nothing else.

What a broken stage at each level actually means

Once you know which stage broke, the diagnosis narrows fast, because each stage fails for a short and well-documented list of reasons.

Stage	What a decline usually means	First thing to check
CPM	Auction pressure, shrinking audience, or seasonal bidding. Often exogenous.	Frequency, and whether the platform's own reported price per ad moved too
Effective CTR	Creative stopped working or is reaching the wrong people. Fastest to fix, fastest to break again.	Spend-weighted creative age and how much new creative shipped in 30 days
Add-to-cart rate	The click was real but the page did not close it. Price, page, speed, or a stockout.	Out-of-stock on your top 10 SKUs, then mobile load time
Reach-checkout rate	They wanted it, then something in the cart stopped them. Shipping thresholds and cart bugs live here.	The deploy log for the weeks it started sliding
Checkout close rate	They started checkout and bailed. Extra costs, slow delivery, missing payment method, forced account.	Total extra cost shown at the final step versus the product price
New-customer share	Paid spend is increasingly buying orders from people who already knew you.	Share of paid revenue going to retargeting and branded search

The checkout-side rows are where the published research is strongest and where operator attention is weakest. Baymard Institute puts the average documented cart abandonment rate at 70.22%, computed across 50 separate studies, and finds that the average large ecommerce site can gain a 35.26% increase in conversion rate through better checkout design alone. Their breakdown of why people abandon during checkout is the practical part: 40% say extra costs like shipping, tax, and fees were too high, 18% were forced to create an account, and 17% found the checkout too long or complicated.

Notice that the single biggest reason, extra costs, is not a design problem. It is a pricing and shipping-threshold decision that someone made in a spreadsheet, and it shows up as a funnel stage collapsing weeks later. That is the kind of link this decomposition makes visible and a conversion-rate dashboard does not. If you want the full picture on how those numbers translate into working capital and payback, our note on CAC payback benchmarks covers the downstream half.

Do it weekly, and let the constraint move

The reason to run this every week rather than every quarter is not diligence. It is that the constraint moves, and a quarterly cadence means you are always fixing the last one.

This is Eliyahu Goldratt's Theory of Constraints, which most operators now meet through Alex Hormozi rather than through The Goal. The five focusing steps are: identify the constraint, exploit it, subordinate everything else to it, elevate it, and then repeat, without letting inertia become the new constraint. Hormozi's version is blunter: a business grows to its constraint and then stops growing, and most founders work on the wrong thing because they never identified the real one.

Step five is the one people skip. You fix the cart, reach-checkout rate goes back to 70%, and the next week the binding stage is creative fatigue instead. That is not a failure. That is the system working. You have bought yourself the next constraint, which is a better constraint than the one you had.

The loop, in about 20 minutes a week:

1. Paste last week's raw counts into the sheet. Counts only, never rates. Let the rates calculate.
2. Check for any stage above 100%. Fix the instrument before reading the dial.
3. Rank the six stages by dollars of CAC damage.
4. Take the top one. Ignore the other five this week.
5. Run the three checks for that stage, ship the fix, and come back next Monday.

This is not a framework we sketched out for an article. It is how we currently do the work. Every brand we run finance for has this decomposition sitting behind their weekly numbers, and the standing question in the Monday meeting is not "how did last week go," it is "which of the six stages is binding right now, and what are we doing about it this week." The numbers change every week. The question does not.

So we took the sheet we use for it and made it public: the Weekly Funnel Constraint Tracker is a free copy-and-use Google Sheet. You paste raw weekly counts into the blue columns, and it calculates all six stages, flags tracking errors, ranks the stages by dollar damage, names your constraint, and tells you the first three things to check for that specific stage. The worked example in this article is the sample data that ships in it.

When we have struggled with this ourselves, the failure was never the analysis. It was letting three weeks pass between reads, so that by the time we saw the break we could no longer tie it to the change that caused it. Weekly is the cadence that keeps cause and effect close enough together to be useful. If you already run a weekly finance rhythm, this bolts onto it: our weekly cash flow KPIs note covers the cash side of the same meeting.

Do that fifty times a year and you will have broken your top constraint fifty times. That compounds in a way that no single heroic project does.

Sources and methodology

The worked example is synthetic but internally consistent, and every figure in it is reproducible. The 13 weeks of raw counts (spend, impressions, sessions, cart additions, checkouts reached, orders, new customers) are the sample data in the public tracker template. Every rate, CAC figure, and dollar-damage number in this article is calculated from those counts by formula, not asserted. Readers can open the sheet and check the arithmetic.

Cart abandonment and checkout-failure figures come from Baymard Institute's own published research, not from a secondary summary of it. The 70.22% average cart abandonment rate is Baymard's aggregation across 50 separate studies, and the 35.26% conversion uplift from checkout redesign and the abandonment-reason breakdown (40% extra costs, 18% forced account creation, 17% checkout too long) are from the same source.

The media-cost claim is taken from Meta's own investor disclosure rather than press coverage of it. Meta's first quarter 2026 results report average price per ad up 12% year-over-year and ad impressions delivered up 19% year-over-year, for the three months ended March 31, 2026.

The constraint framework is Goldratt's, and the popular framing is Hormozi's. The five focusing steps are published by North River Press, Goldratt's own publisher and the publisher of The Goal. The contemporary operator version, including the point that a business grows only to its constraint, is set out on Acquisition.com.

The data-quality rule comes from experimentation literature, not from marketing convention. Ron Kohavi and Roger Longbotham's paper Unexpected Results in Online Controlled Experiments, published in ACM SIGKDD Explorations, opens with Twyman's law and works through several cases where an apparent result turned out to be a measurement artifact, including the monitoring-system click failure described above.

Stage-level counts can be pulled from your ecommerce platform or from analytics, and the definitions have to match on both sides of every ratio. Google's documentation for the checkout journey report sets out how the cart and checkout steps are defined there. The most common source of a broken ratio is taking the numerator from one system and the denominator from another.

Frequently asked questions

why is my cac going up when my conversion rate looks fine?

Site-wide conversion rate is an average of four separate stages, so one stage can collapse while the average barely moves. Split it into add-to-cart rate, reach-checkout rate, and checkout close rate. In the worked example here, two of the three were flat or improving while the third fell by a third.

what is a reach-checkout rate and why have i never tracked it?

It is the share of sessions that added to cart and then actually started checkout. Most dashboards jump straight from add-to-cart to purchase, so this stage has no owner and no alert on it. That is exactly why it is so often the stage that quietly breaks.

how much does cac have to move before i should actually care?

Under plus or minus 5% week on week is usually noise, especially below 500 orders a week. Investigate at 10% or more. Also investigate whenever the same direction shows up three weeks running, even if no single week looks dramatic.

should i be looking at this weekly or monthly?

Weekly. Monthly review means you find out about a broken cart six weeks after it broke, and by then you cannot tie it to the deploy that caused it. Weekly is frequent enough to connect a change to a cause and slow enough that you are not chasing daily noise.

my cpm is up a lot. isn't that the whole problem?

Usually not. Meta reported average price per ad up 12% year-over-year in Q1 2026, so a CPM rise in that range is you tracking the platform, not you failing. Run the dollar comparison before you accept it as the cause. In the worked example CPM was worth $3.15 of CAC and the checkout stage was worth $15.87.

how do i tell the difference between a site problem and a demand problem?

Site problems show up as a sharp break at one stage on a specific week, usually traceable to a deploy or an app install. Demand problems show up as a slow drift across several stages at once, with click-through rate and add-to-cart rate falling together. If only one stage moved, look at your deploy log first.

does this work if most of my traffic is organic or email?

The site-side stages work identically. The paid-side terms get blurry because you are dividing paid spend by sessions that did not come from paid. Either track the funnel on paid-attributed sessions only, or accept the blended read and watch the trend rather than the absolute number.