Most stores ask us for traffic. It is the natural request, because traffic is the number you can see going down and the one you can pay to change tomorrow. We nearly always start somewhere else.
The reason is arithmetic rather than preference. Revenue is visitors multiplied by conversion rate multiplied by average order multiplied by repeat purchases. Improving conversion improves the return on every visitor the store gets, from every source, including the ones you already paid for and the ones arriving free. Improving traffic improves nothing except itself, and it charges you again next month for the same result.
So the order is not about ads being bad. It is about which change keeps working. Buy a month of traffic and you own a month. Fix the step where people leave and you own a better version of every month after it, including the months where you buy more traffic.
A conversion fix is retroactive; a traffic fix is not
This is the whole argument in one line, so it is worth sitting with.
When you fix the reason people abandon at the shipping step, that fix applies to paid visitors, organic visitors, email clicks, people who typed your name in directly and people who came back from a story someone shared. It applies to the ones who arrive tomorrow and the ones who arrive in March. You did not buy any of them a second time.
When you add traffic, the extra visitors meet the same store as before. Whatever share of them was going to leave still leaves. You have paid for the whole group to get the buying share, and you will pay again next month at the same ratio, probably at a slightly worse price as the cheap demand runs out.
There is a second effect that is easy to miss. A higher conversion rate raises what you can afford to pay for a click, which means campaigns that were unaffordable become affordable, and the auctions you were losing become ones you can enter. The conversion fix does not just save the traffic you have. It widens the traffic you can buy.
Conversion is the only number that improves the visitors you have already paid for.
The same budget, before and after
Here is a worked example with invented numbers, to show the shape rather than to claim a result.
Say a store gets 40,000 visitors a month and converts 1.5 per cent of them. That is 600 orders. At an average order of $80, the month is $48,000. Say paid brings half the visitors at 90 cents each, so $18,000 of spend.
Now take two routes with the same $18,000.
Route one, buy more traffic. Another $9,000 buys roughly 10,000 more visitors at the same price, which at 1.5 per cent is 150 more orders and $12,000 more revenue. Next month the extra $9,000 is due again, and the price per visitor is unlikely to have improved.
Route two, raise conversion from 1.5 to 1.8 per cent on the traffic you already have. That is 720 orders instead of 600, so $57,600 instead of $48,000, for no extra media spend. The gain repeats every month at no further cost, and the paid half of the traffic now returns more per dollar, which means the budget you already spend goes further.
Swap in your own visitors, conversion rate and average order. The conclusion tends to survive the swap, and where it does not, that itself is the answer about where to start.
Traffic bought against a leak is bought twice
There is a cost to the wrong order that does not show up as a line item.
When a store adds spend on top of a page that loses people, the loss scales with the spend. The abandoned carts, the bounces from a slow product page, the mobile visitors who never see the size guide, all of it grows. You are paying full price to demonstrate the same problem to more people.
Worse, the reporting gets muddier at exactly the moment you need it clear. Cost per order rises, and now there are two candidate explanations sitting on top of each other: the auction got more expensive, or the store is losing a bigger share of a worse traffic mix. Untangling that takes weeks, and the natural response in the meantime is to spend more, which makes the tangle bigger.
Fixing the leak first has a quiet second benefit. It gives you a clean baseline. When the store holds visitors properly and then you turn spend up, any change in cost per order is about the traffic, because the other side of the equation stopped moving. That is a much easier problem to run.
Three times we go to traffic first
The rule is not absolute, and pretending it is would be dishonest.
The first exception is too little traffic to diagnose anything. If a store gets a few hundred visitors a month, there is no conversion rate to read. Two orders more or fewer swings it wildly, and any change you make is unmeasurable. Below a certain volume the first job is genuinely to get people through the door, then look at what they do.
The second is a dated window. If it is six weeks before your season and the store converts acceptably, the conversion work that pays back over a year is the wrong thing to start in the one month where demand is sitting there. We do the small conversion items that ship in days, get the traffic right for the window, and come back to the deeper work in the trough.
The third is a genuinely new product with no demand yet. There is nothing to convert. Someone has to learn the product exists before their behaviour on the page tells you anything.
Outside those, the order holds. And in all three cases we would still fix anything that is outright broken at checkout before spending, because that is not optimisation, it is repair.
Telling a real gain from a change in mix
A conversion rate can rise because the store got better or because the traffic changed, and mistaking one for the other leads to bad decisions later.
So we never read the blended number alone. We look at conversion by device, because mobile and desktop behave differently and a shift in the split moves the average without anything improving. We look at it by source, because branded search and returning email visitors convert differently to cold prospecting, and a week with more branded traffic will look like a win. And we look at the step that was actually fixed, not just the final number: if the change was on the product page, add to cart should move first, and if it did not, something else caused the rise.
We also write down the before figures and the date of the change before we make it. It sounds obvious. It is skipped constantly, and without it every discussion about whether the work paid off comes down to who remembers the old number more confidently.
That discipline is what lets us say in a monthly report that a fix did not work. Some do not. Knowing which is the only way the next choice is better than the last one.