An order was a promise, not a payment

The shop

I worked in-house for an electronics retail chain — phones, laptops, appliances. The website and the physical stores drew on the same stock. It was the middle of a land grab: every large retailer in the country was spending to take customers off the others, and the site was the front line.

The problem

People were leaving before they ordered.

Some of that was ordinary friction. Most sites then made you register, then log in, then check out — three walls before you could buy anything. Connections were slow, machines were slow, hosting was nothing like it is now, and every input field and every second of loading cost orders.

But the deeper problem belonged to that market and that moment. Payment gateways were rare in Vietnam in 2016 and the fees were high, so an order was not a transaction. It was a promise: you said you wanted the thing, and later you went and got it and paid.

There was nothing to pay for. What had to be secured was belief.

That changes what the customer is afraid of. They are not worried about their card. They are worried the thing will not actually be there, at that price, near them, when they turn up.

The constraint

None of the constraints were technical.

Search traffic was the largest. The site ranked, and a restructure that broke the existing URLs would cost positions that had taken years to earn. Whatever changed had to leave that intact.

The manufacturers had rules of their own. Apple and Samsung both dictate how their products may be presented, and a retailer does not negotiate those.

And this was a full in-house operation — marketing, developers, designers. Marketing wanted pages that were attractive and that pushed people toward buying. That is a legitimate goal, and it pulls against a quiet page.

The decision

Order without an account. The registration wall came down.

A speed budget I set myself — a rough one, by judgement rather than measurement. The number mattered less than the principle: nobody should have to cross a dozen input fields and a dozen seconds of loading to buy a phone.

And a block of trust at the top of the product page. Every line in it answered a question the customer was already asking.

What the customer is askingWhat the page answers

01Is this the price I pay?→The final price, with nothing added later

02What else do I get?→Which promotions apply, shown applied

03Does it actually exist?→Stock at the store nearest you

04And if that one is out?→Whether it can be moved from another store

05How do I get it?→Where to collect it, and what delivery costs

Backed by a two-hour hold on the item. The stock was shared with the shops, so the promise had to be real.

Two of those lines are less obvious than they look. In Vietnam a promotion is a genuine draw rather than decoration, so it belongs above the fold and not in the small print. And the stock figure had to be true, not indicative — it was shared with the shops, which is why the hold had to exist behind it.

And then the one I am still pleased with. The call button did not say Call us. It said: free hotline — it's fine if you don't buy.

Everyone in the industry was counting clicks. But the reason people were not calling had nothing to do with taps. Calls cost money then, and a customer who called and did not buy felt they owed something. So the permission not to buy went into the button.

What happened

The order flow was well regarded in the trade. This was the pain point everyone had, and it turned into real orders rather than abandoned ones.

It was instrumented, too: the team tracked customers who opened the order panel and did not finish, and showed a prompt when someone moved to close it. Exit intent, in 2016, in Vietnamese retail.

It also created a conflict I had not designed for. Stock reserved for a customer on the website is stock the customer standing in the shop cannot buy, and the two sales teams were paid on different numbers. Some shop staff held items back; some reported stock they did not have.

What settled it was not design. The company already had the plumbing: a call system that logged who had advised a customer, so an order that began on the hotline was credited to the online team wherever it was finally collected. Shop staff still had to serve that customer, and were not always pleased to. Falsified stock and invented orders were investigated, and people were dismissed over them — managers and shop accountants included.

I do not have the numbers. They existed, and I did not keep them.

Looking back

The two-hour hold was not an interface problem. It was a commission problem wearing an interface.

That is the lesson I keep. A design decision that moves value between two groups inside a company is not finished when the screen is finished. The hold worked because somebody had already built the attribution and the enforcement to make it stick. Without those, the screen would have been quietly defeated by the people who lost by it — and I would have called it a usability problem and looked in the wrong place.

The second thing is simpler. I did not keep the numbers. I can describe exactly what we built and why; I cannot prove what it did. Ten years on, that is the whole distance between a story and a result, and the cost of it is mine.

There is one more thing, which I only noticed recently. Holding stock for two hours across a pool shared by a website and a row of shops — so that the customer online and the customer at the counter see the same truth — is multi-source inventory, seen from the merchant's side. A decade later I build exactly that from the other side, for the companies who sell the software. Same problem, opposite chair.

Written from memory, ten years after the work. Figures are not available and none have been invented.

← All work