Where trade-in forms lose people, and what I changed

For a shop that buys used stock from the public, the trade-in form is the supply chain. Every person who gives up halfway is a guitar, an amplifier or a pair of speakers that goes to a competitor instead.

Audio Trader buys used music and hi-fi gear through a form on its website. In August I took its funnel apart, step by step, to find out where people were going. These are the figures, the changes that followed, and the bug that froze the whole thing for four days.

The funnel

Over the 30 days to 12 August 2026, on the main trade-in page:

Step	Share
Visitors who started the form	55%
Starters who added the required photo	43%
Starters who submitted	about 38%
Visitors who submitted	22%
People who added a photo and then submitted	88%

One caution on the first line. “Started” is measured as the first click into the make and model field, not the first keystroke, so it flatters the start a little. Every form tracker has a definition like this somewhere, and it is worth finding out what yours is before you trust the top of the funnel.

Two more facts shaped everything else: 81% of visitors were on a phone, and about three quarters arrived from paid ads, cold. Paid visitors converted at well under two thirds the rate of people who came to the site directly.

The cliff is the photo

Between picking a condition and adding a photo, roughly half the people left. Once a photo was in, 88% went on to submit. That one step was the biggest leak in the form by a distance.

The obvious reading is that the photo upload was broken. It mostly was not. Recorded errors on that step accounted for under 6% of the people who left. The rest simply stopped. It was effort, or doubt about whether it was worth the bother, not a fault.

Removing the photo was not an option, because the photo is what lets the buying team price the item properly. So the job was to make the step lighter and better explained, not to take it away.

The second leak: phones that lose the connection

There was a smaller leak after the photo. Of people who had added a photo and pressed submit, 4% on desktop never got through. On phones it was 14%, and 18% in the last week measured.

96% of those failures were the same thing: the connection dropped while the photos were uploading, and the request never finished. Not a server error and not a file too big: a phone on a weak signal.

Two details made it worse than it looked:

• Errors are not people. About 800 error events came from about 150 people, because everyone retried and every retry logged another error. Count people, not events, or one frustrated customer looks like a crisis.
• The retries tripped the spam limit. People tapping submit again and again hit the form’s rate limit, which then blocked the retry that would have worked.

What changed, in order

1. Photos are compressed on the phone before upload, and photos from Android phones that do not say what type of file they are, and iPhone HEIC photos, are no longer silently dropped.
2. Finer tracking on the photo step, separating people who opened the picker and gave up from people who never tried. One is an upload problem, the other a motivation problem, and they need different fixes.
3. Submit retries itself when the connection drops, with a key the server recognises, so a retry can never create a second trade-in.
4. Accessibility fixes, so the photo slots work from a keyboard and a screen reader is told when the form has gone through.
5. Name and email come first. The old form asked for the email last, so nobody who gave up could ever be followed up. The new one asks for contact details first, saves progress, and an unfinished form gets a reminder email that opens with the details already filled in.
6. Back closes the form, not the page. On a phone the full-screen form felt like a page, so people pressed back, left the site, and lost everything they had typed and photographed.
7. The condition guide opens over the form instead of sending people to another page in the middle of a quote.

The bug that froze the page

Four days after the new form went live, some visitors found the trade-in page frozen. Nothing they tapped did anything.

The form’s opening code read the id of every element on the page. On a page that contains a form with a field named “id”, the browser hands back the field, not the form’s id, so the code threw an error halfway through opening. The page was left blocked behind a form that never appeared.

More than twenty automated browser tests passed. A real Chrome window, opened the way a visitor would open it, showed the error at once. Now the form opens defensively and undoes itself if anything fails. The lesson I keep relearning: test in the browser your customer uses, along the path they take.

What it changed

I do not know yet, and I will not guess. The new form has been live since late September, and a fair before and after needs a full month on each side, measured the same way. I will add the figures here when there is one.

Checking your own form

If you buy stock from the public, or run any form longer than a contact form:

1. Find your cliff. Track each step, not just start and submit. The step losing the most people is rarely the one you would guess.
2. Split errors from effort. Count the people who hit an error, not the error events, and compare them with everyone who left at that step.
3. Ask for the email first. If you only learn who someone is at the end, you cannot follow up anyone who gives up.
4. Test on a phone, on a poor signal, and press back. That is where your customers are.