You can improve a step and still make the funnel worse.

Day 28 on AffiSpark.

Yesterday I wrote that the ask that fails is not always the ask that caused the failure.

Today I think the sharper version is this:

You can improve a step and still make the funnel worse.

I think founders fall into local optimization very easily.

A step underperforms, so we improve that step.

We tighten the copy.

We make the CTA stronger.

We ask earlier.

We remove hesitation.

We push harder on the paid ask.

We make the prompt more visible.

And sometimes the step improves.

But the overall flow gets worse.

Why?

Because if the real problem is upstream, then a stronger local ask just spends the trust budget faster.

That is the risk.

For AffiSpark, I can see that clearly now.

If I made the paid ask more aggressive before the preview had earned enough trust, that would not fix the real problem.

It would just make the pricing friction arrive harder.

If I made the walkthrough CTA louder before the flow had earned enough momentum, that would not fix the real problem.

It would just make one more ask feel heavier.

If I pushed the feedback prompt harder before the user felt understood, that would not create better learning.

It would just create another withdrawal from the same balance.

That is why I think some conversion work backfires.

The local metric can improve while the product gets more exhausting overall.

A stronger step is not always a better step.

Sometimes the best thing you can do for a weak step is improve the step before it.

More proof before pricing.

More clarity before setup.

More completion before handoff.

More context before asking for feedback.

That feels like a better rule.

Not:

how do I make this step convert harder?

But:

what would make this step need less force in the first place?

That question has more leverage.

Because if a step only works when you push it harder, the product may still be compensating for something earlier that has not been solved.

My current rule is simple:

when a step underperforms, I want to optimize upstream before I intensify the step itself.

Curious how others think about this:

what local optimization in your product made the overall flow worse instead of better?