Funnel Optimization: Find the One Element That Wins, Then Spread It

Funnel optimization starts with the part that already wins

July 1799. French soldiers are rebuilding a fort near the Egyptian town of Rosetta. One of them finds a broken stone slab in an old wall.

It carries one decree written three ways: in hieroglyphs, in a later Egyptian script and in Greek. Scholars could already read Greek. So the Rosetta Stone became the key to reading the other two.

Your funnel has a stone like that. One part that buyers say yes to far more than anything else. It shows what they actually came for.

Most funnel optimization advice says test everything.

The button. The form. The footer.

We do the opposite. Find the one part that already wins. Check that it is real, then spread it through everything else.

A funnel is every step a buyer passes through: the ad, the page, the checkout, the extras, the emails. Here is how to read yours for the winner.

The One-Element Rule: one part that wins on every number

The One-Element Rule is our name for this. When one part of your funnel beats the others on every number, it shows you what buyers want. So you build more of the funnel around it.

Every number. Not just one.

Why does that matter? Because one number can lie. A $1 extra will get taken a lot, and that alone proves nothing.

So we check three numbers. Take rate is the share of people who say yes when they see an offer. We also check money per person who saw it, and refunds.

For example, one music business sold a guitar offer with a few extras. One extra used AI to help buyers write songs. Between 70 and 80 percent of buyers took it.

The rest of the offer did fine. Nothing special.

So they put the songwriting help into the main offer, the upsell, the emails and the ads. Sales took off. We have not seen their numbers, so treat it as an example, not proof.

Seventy percent is more than a nice bonus. It tells you what buyers came for.

Where does the winner hide? Six places to look

The winner usually hides in a spot nobody reads closely. Start with these six.

The checkout extra. A small add-on with a tick box at checkout, often called an order bump. Read its take rate.

The offer right after purchase. This is the upsell. Read its take rate and the money it brings per buyer.

The cheaper offer for people who said no. Some buyers pass on the upsell and take the smaller one. Sometimes the smaller one is the real hit.

One email that gets replies. Seven emails get opened, and one gets answered. Its topic is what readers care about.

One ad that carries the account. You only see it when you read results ad by ad, not in total. We explain how in read results, not leads.

One question on every sales call. Prospects keep asking about the same thing first? That is your headline, handed to you.

How do you find it? Put every part on one scorecard

You find the winner by putting every part on one scorecard. Same time window, same measure for each type.

Step 1: list every part a buyer sees, in order. Ads, main offer, checkout extra, upsell, cheaper offer, emails.

Step 2: give each type one main measure. Extras get take rate, and emails get replies per send. Ads get cost per result.

Compare like with like. An email and an ad do different jobs, so their numbers never compete.

Step 3: write down how many people saw each part. A percent with no count behind it means nothing.

Step 4: run the every-number test. Check the main measure, money per person who saw it, and refunds. Money per person who saw it is the same idea as earnings per visitor, applied to one part.

Step 5: mark any part that beats the next best of its type by double or more. That is our line. Smaller gaps are usually noise.

Is it real or a fluke? Check the count, then check it twice

A big percent from a small number of people can be pure luck. So before you rebuild anything, check two things: how many people saw it, and whether it wins twice.

Example: 7 of 10 buyers take an extra. That is 70 percent. But the true rate could be anywhere from about 40 to 89 percent, using standard confidence interval math.

At 70 of 100 buyers, the range narrows to about 60 to 78 percent. At 280 of 400, it is about 65 to 74.

So our rule is a floor of 100. Each part must be offered to at least 100 people before you call a winner.

Then check it twice. One great week is not a finding. It is a great week.

Results far above normal tend to slide back toward average, which statisticians call regression toward the mean.

So run a second stretch of the same length. Does it still beat the next best by double? Then it is real.

One more trap. Slice the results enough ways and something will look best by pure luck. That is the multiple comparisons problem, so pick your measure before you look.

Ask why it wins before you copy it

A winner is a clue, not a finished answer. Before you spread it, name the reason buyers want it.

Ask five questions.

Is it the topic? The promise? The price?

How little work does it ask of the buyer? What about the format?

In the music example, our read is that the extra took work off the buyer. Writing a song is hard. Getting help writing one is easy to want.

The reason is the part you can reuse. We go deeper on that in copy the rules, not the ad.

Can't name the reason? Ask the people who took it. A two-line email is enough: what made you add this, and what did you hope it would do?

Guessing the reason is how a winner gets copied wrong.

Spread the winner into your main offer, upsell, emails and ads

Once the winner holds up, put more of it everywhere a buyer looks. Four places, in this order.

The main offer. Lead the page headline with it. Name it on the first screen, not in a list of bonuses.

The upsell. After purchase, offer a deeper version of it, or help with the next problem it creates, as in sell the next problem.

The emails. Open the sequence with it. Our lead-first email sequence shows where the first email goes.

The ads. Work it into the hook. Change one thing at a time, as in one variable at a time, so you know what moved.

Sometimes the winner should become the product. Instagram began as Burbn, a check-in app built by Kevin Systrom and Mike Krieger. Photo sharing had become its most popular feature.

So they cut the rest and relaunched as a photo app in October 2010. Facebook bought it in April 2012 for about $1 billion.

Same founders. One feature.

Stop polishing the parts that are not winning

Every hour spent on the eighth email is an hour not spent on the one that works. A few parts usually bring most of the results. This idea is known as the Pareto principle.

Our take: many funnel audits grade every part the same because that fills more pages. Equal attention sounds fair. It is just slow.

How to optimize a funnel into the ground in four easy steps:

1. Test the button color first. It feels like science.

2. Give all eight emails the same hour of polish.

3. Read the totals, never the parts.

4. Find a 70 percent winner, then leave it buried in the bonus list where it belongs.

Polish feels like progress. Mostly it is just polish.

Keep normal split testing (showing two versions to see which wins) for the parts you spread the winner into. Just stop running it on parts that do not matter.

What if nothing in my funnel stands out?

Sometimes every part looks about the same. That usually has one of three causes.

Too few people. Twelve ads with three results each cannot show you a winner. Run fewer things with more people behind each, until each one reaches 100.

Parts too alike. Eight emails making the same point in slightly different words will score alike. Make them truly different.

The funnel is not selling yet. With no sales there is no winner to find. Then the job is getting one thing to work, which we cover in launching ads for a new offer.

And remember, you only see what you tried. Your best part is only the best of what you built. This trap is related to survivorship bias.

So keep testing one new idea at a time, even while you spread the winner. The next stone may still be in the wall.

Find your winner with AI: the copy-paste prompt

Paste the prompt below into your preferred AI. It asks for your parts and their numbers, one at a time. Then it runs the scorecard, the count floor and the second-stretch check.

It never invents a number. If one is missing, it asks you or marks it unknown.

Want a second pair of eyes on the result? You can send it to us.

Run it this week

1. Pull the last two months. Write every part, its main measure and how many people saw it.

2. Run the prompt. Check its math against the 100 floor and the double gap.

3. If one part passes, move it into your headline first. Then the upsell, the emails and the ads.

No winner yet? Check the whole offer with six checks for an offer that is not selling.

The scholar Jean-Francois Champollion did not invent a new language. He used the Greek text, which he could read, to help decode the rest. Your buyers have already shown you what they want, so read their choices.