98 of Every 100 Visitors Leave. What Were They Deciding?
A visit that ends without a purchase is not an absence of interest. It is a decision that ended somewhere, and most stores never see where.
Published 2026-09-21 ยท 5 min read
98 of Every 100 Visitors Leave. What Were They Deciding?
The Number Everyone Knows And Nobody Opens
Average conversion in Brazilian e-commerce sits close to 2%. Every team knows this number. Far fewer look at it from the other side: out of every 100 people who arrive, 98 leave without buying.
The usual report gives those 98 a single label. They bounced, or they abandoned. The label is accurate and useless, because it describes where the visit ended and says nothing about what the person was trying to decide when it did.
Leaving Is The End Of A Decision
Nobody leaves a store for no reason. Something in the visit did not close. Four situations cover most of what we see:
Price. The person wants the product and the total is more than they expected.
Shipping. The product price is fine. The freight cost or the delivery date is the problem, and it often appears late in the visit.
Trust. The person has never bought from this store. They look for reviews, return policy, and payment security before committing.
Comparison. They are still choosing between this product and another one, here or in a different tab.
There is also a fifth group that matters just as much: people who were never going to buy today. They are researching, and they will come back or they will not.
Same Exit, Opposite Answers
These situations look identical in a funnel report. They call for opposite responses.
A discount helps the person stuck on price. It does nothing for the person who doubts the store, and it gives away margin to the person who was only comparing and had already chosen you. A trust signal, such as a clear return policy, helps the first-time buyer and is noise to a returning customer. Shipping doubt is answered with a delivery date or a freight option, not with 10% off the product.
Treating all exits as one problem means choosing one answer and being wrong for most of the people who receive it.
What The Visit Already Shows
The information needed to tell these situations apart is usually already in the session. A visitor who scrolls straight to the freight calculator and returns to it three times is showing where the doubt is. One who opens the reviews tab and the returns page is asking a trust question. One who moves between two product pages is comparing.
None of this requires reading what the person typed. It is the pattern of the visit: pauses, returns, scroll depth, and the order of the pages.
No single signal proves intent. Read together, they are enough to ask a better question than "how do we get them back?" The better question is: what was unresolved, and is there anything the store should do about it right now?
Three Decisions, Not One
That question breaks into three smaller ones:
1. Whether to intervene at all. Many visits need nothing.
2. How to intervene: freight information, a trust signal, a comparison aid, a discount, or a human.
3. When to do it: during the visit, on return, or not until the person asks.
This is the approach behind Fluence. It reads behaviour during the visit, diagnoses the most likely reason for the uncertainty, and chooses an intervention, including the choice to do nothing. Each decision is logged with its reason, so the team can see what worked.
In a partner pilot covering more than 3.4 million profiles, this approach was associated with a 2.3x increase in conversion. Those are reported results from one deployment, not a guarantee for another store.
The first step costs nothing: stop calling it abandonment, and start asking which decision ended.
See the six decision problems we work on in e-commerce.