Can a funnel test win and still lose money? Yes. The number you tested goes up, the report says “winner”, and the month ends with less money in the bank than before.
It happens because a funnel isn’t a row of separate numbers. It’s one living system. Change one part and the rest of it reacts, sometimes for the better, often in a way nobody measured.
I see this most with coaches who just started reading their numbers. They pick one, the opt-in rate or the checkout rate, and work on it until it climbs. That’s a good habit. The trouble is what they stop looking at while they do it.
Here’s a real one, and an honest one. In 2026 I built a full funnel for Travis Stephenson at PaidCreators. The front-end sales page brought in $249,291 from 6,295 orders, converting 2.35% of visitors on cold Meta traffic, from December 21, 2025 to August 2, 2026 (his funnel dashboard). That number also reflects his offer, his ad spend and his team.
The upsells I built didn’t hold, and I never got the chance to find out why. If I judged that funnel by the front page alone, I’d call it a clean win and stop looking. One strong number told me nothing about the next stage. That’s the habit this article is about.
The short answer
- Every number in a funnel is tied to others. When one goes up, check what went down in the same week.
- More leads can mean fewer buyers. A lower bar at the opt-in lets in people who were less ready to buy.
- More checkouts can mean smaller orders. A cheaper or simpler entry often brings a lower order value and less buying later.
- Judge every test by money, not by the number you tested. Cash collected per visitor, over a window long enough to include what people buy later.
- Pick your guardrails before the test starts. Two or three numbers you agree not to let fall while you push the one you’re working on.
1. Why a “winning” test can leave less money
A test measures one stage. The money gets made across all of them. A change that helps one stage can hurt the stages after it, and the test report rarely follows people that far.
Most testing tools stop counting at the click or the form. They’ll say version B got more opt-ins. They won’t say those opt-ins booked fewer calls three days later, or bought less a month later. That isn’t a flaw in the tool. It’s a question nobody asked it.
So the coach ships version B, sees more leads, and moves on to the next test. The drop shows up weeks later, in a different number, and nobody connects the two. By then it looks like the ads got worse.
This is what I mean by optimizing yourself into a loss. Each change looks like progress on its own. Added together, they leave less money.
2. The opt-in trap: more leads, fewer buyers
There are a handful of common ways to raise an opt-in rate:
- Make the form shorter.
- Make the free promise bigger.
- Remove the questions that filter people out.
- Push harder with a pop-up or a countdown.
Each one works the same way: it lowers the bar. People who weren’t sure now join too. They count as leads exactly like the ready ones, so the lead number goes up and the cost per lead goes down. Everyone is happy for about a week.
What I’d notice
The opt-in rate climbed and the cost per lead dropped, but booked calls per lead fell in the same weeks.
Why it would catch my attention
A lead is only worth what it turns into. If 1,000 leads used to produce 40 calls and 1,500 now produce 40 calls, the funnel paid for 500 extra leads that went nowhere. It also paid to email them, to follow up with them and to sort through them.
The question I’d ask before touching anything
Of the leads from the new version, how many booked or bought, compared with the leads from the old version over the same number of days?
3. The checkout trap: more orders, smaller orders
The same thing happens at the checkout, with different numbers.
To raise the checkout rate, people lower the price, remove the order bump (the small add-on offered on the checkout page), skip the upsell, or cut the page down. More people buy. The buyer count looks great.
Now look at two other numbers. Average order value is what the typical buyer spends on that first visit. Lifetime value is what one buyer spends with you over time, front end plus everything after it. When the entry gets cheaper and lighter, both of these can fall, and they can fall by more than the buyer count rose.
There’s a second effect too. A buyer who paid less and was asked for less is often less committed. That doesn’t make them a bad buyer. It means the step to the next offer is bigger for them, and fewer of them take it.
4. The call trap: more booked calls, fewer real ones
Book-a-call funnels have their own version. To fill the calendar, the application questions come out and the calendar link goes everywhere. Booked calls go up fast.
Then the show-up rate drops, because some of those people booked on a whim. The close rate drops, because some of the ones who show up were never a fit. And the hours on the calendar don’t grow. Whoever takes the calls now spends part of every day on calls that were never going to close, which leaves less time and energy for the ones that could.
The calendar looks full. The bank account doesn’t match it.
5. The pairs to watch together
This is the table I’d keep next to any test. On the left, the number you’re trying to raise. In the middle, the number to check at the same time. On the right, why they tend to move together.
| If this goes up | Check this | Why they move together |
|---|---|---|
| Ad click rate | Opt-in rate and sales per lead | A louder hook can pull clicks from people who are curious, not buying |
| Opt-in rate | Booked calls per lead, sales per lead | A lower bar lets in people who were less ready |
| Cost per lead goes down | Cost per sale | Cheaper leads can take more of them to make one sale |
| Booked calls | Show-up rate and close rate | Fewer filters fill the calendar with people who aren’t a fit |
| Show-up rate | Close rate | Pressure or a gift to show up can bring people who came for the gift |
| Checkout rate | Average order value, bump and upsell take rate | A cheaper or lighter entry brings smaller first orders |
| Front-end buyers | Back-end buyers per 100 buyers, refunds | Buyers who came in at lower commitment can buy less later and refund more often |
| Close rate on calls | Refunds, failed payments, cash collected | A hard close can turn into a cancelled sale a week later |
| Sales this launch (discount) | Sales next launch, full-price buyers | A list that learns to wait for the discount waits for it next time |
You don’t need to watch all of these at once. For any test, pick the row that matches what you’re changing and watch the middle column with it.
6. A worked example: the test that won by 50% and lost money
Illustrative figures. Not a client.
A coach sells a $47 workshop with ads. On the checkout there’s a $17 order bump. After the purchase there’s a $197 upsell. Within 60 days, some buyers move up to a $2,000 program.
The coach tests a new version: the workshop drops to $27 and the bump comes out, to “reduce friction”. Both versions get 10,000 visitors from the same ads, at the same $6,000 in ad spend.
| Number | Version A ($47 + bump) | Version B ($27, no bump) |
|---|---|---|
| Visitors | 10,000 | 10,000 |
| Checkout rate | 2.0% | 3.0% |
| Buyers | 200 | 300 |
| Front-end sales | $9,400 | $8,100 |
| Bump ($17, taken by 30% in A) | $1,020 | $0 |
| Upsell ($197) | 20 buyers (10%), $3,940 | 18 buyers (6%), $3,546 |
| Average order value | $71.80 | $38.82 |
| $2,000 program within 60 days | 6 buyers (3 per 100), $12,000 | 4 buyers (1.3 per 100), $8,000 |
| Total sales in 60 days | $26,360 | $19,646 |
| After $6,000 in ad spend | $20,360 | $13,646 |
The test report says version B won. The checkout rate went from 2.0% to 3.0%. That’s 50% more buyers from the same traffic, and on the testing dashboard it looks like the best change of the quarter.
Now follow the money. Three numbers broke at the same time:
- Average order value fell from $71.80 to $38.82, because the price dropped and the bump disappeared.
- Upsell take rate fell from 10% to 6%. More buyers saw the upsell, and fewer of them took it.
- Program buyers per 100 fell from 3 to 1.3. The buyers who came in cheaper moved up less often.
Over 60 days, version B brought in $6,714 less on the same ad spend. In money per visitor, it went from about $2.64 to about $1.96. The winning test cost more than a quarter of what the funnel used to make.
And the coach would only see it if someone tracked the same buyers for 60 days. Judged on day seven, version B looks like a clear winner. Judged on the cash, it’s a loss.
7. How to test without optimizing into a loss
Pick one money number as the judge
Before any test, agree on the number that decides the winner. My choice is cash collected per visitor, or per dollar of ad spend, over a fixed window. The number you’re testing is the thing you change. The money number is how you judge it.
Set guardrails before the test starts
A guardrail is a number you agree not to let fall while you push another one. If you’re testing the opt-in, the guardrail might be booked calls per lead. Write it down before the test starts, so a good-looking result can’t talk you out of it.
Wait for the slowest stage
If buyers usually move to the program within 30 or 60 days, a test isn’t done at day seven. Call the front-end result early if you have to, but keep watching the same buyers until the slow stage reports.
Follow the same people through
Compare the people who came in through version A with the people who came in through version B, all the way to cash. Mixing everyone from the same month together hides exactly the effect you’re looking for.
Change one thing at a time
If the price, the bump and the page all change in the same week, there’s no way to say which one moved the money. One change, one test, one answer.
FAQ
What’s a guardrail number?
It’s a number you promise not to sacrifice while you work on another one. For an opt-in test, it might be sales per lead. For a checkout test, average order value. If the guardrail drops past the line you set, the test doesn’t count as a win, no matter what the tested number did.
Is a higher opt-in rate ever just good news?
Yes. If the opt-in rate goes up and sales per visitor go up with it, that’s a real win. The point isn’t to distrust every gain. It’s to check the next number before you celebrate.
How long should a test run before I trust it?
Long enough for the slowest stage that matters to report. If most program sales happen within 60 days, judge the money at 60 days. You can read the front end sooner, as long as you treat it as a first look and not the verdict.
What single number should I judge a funnel by?
Cash collected per visitor, or per dollar spent on ads, over a fixed window. It includes every stage at once, so a gain in one place and a loss in another both show up in it.
Does this mean I should stop testing?
No. Testing is how a funnel gets better. The change is in how you judge a test: by money over time, with guardrails, instead of by the one number the test was built to move.
What if my traffic is too small for proper tests?
Then I’d test bigger changes less often and watch the direction of the money number rather than waiting for a perfect result. I’d also be extra careful with changes that lower the price or remove a filter, because those are the ones most likely to look good up front and cost more later.
What to do next
If you want to read these numbers together instead of one at a time, that’s what my workshop covers. Data Scaling Strategies shows you how to read each stage after the click, so you know where the money stops before you raise the budget.
Martin Meneses
I’ve built sales funnels since 2010. More than 100 of them, in 12+ niches.