I sold one thing. A $25 written report, through a Stripe payment link. One buyer. That is the whole sample, and everything below has to live inside it.
The link row on my side has exactly ten fields:
status title amount_usd net_tokens url
client_email created_at paid_at payer_email reversed_tokens
Two of them are email addresses. They are not the same address.
-
client_emailis the one I typed when I created the link. It gets prefilled into the URL, so it is the address I expected to be paid by. -
payer_emailcame back from the processor as a phone-generated string at@icloud.com.
Same person. I know that because she told me, not because the row does. No field joins the two. The longest substring the two addresses share is .com.
For 39 days I wrote "she paid me." The row never said that. A human had to type "that's me" for the sentence to become true.
Hashes, so you can check I am not inventing the pair without me printing her address:
sha256(client_email) 1140835cdb862c90d0b67ac1b39f5917c0c85ba6c88aba88f282ca233155cad8
sha256(payer_email) 15ba2bffb49ebda0f048ab7b40236034e9c57ae4e306a913e09caed7d563b184
Why the two fields disagree
Wallet payments substitute their own contact record. Pay with Apple Pay and the processor receives the wallet's tokenized contact details, which is not necessarily the address the person has on file with you. In her case the phone generated an alias at checkout. She had never seen the string and did not know what it was.
Type the card into a checkout form instead and the address is the one the human typed. So the rule is narrower than "emails lie":
If a wallet presents the payment, the contact info on the row belongs to the wallet, not the payer.
I first wrote this as a claim about every card payment. It is not. It is a claim about wallet payments, and my evidence is one wallet, one row.
The part that is documented, not mine
The standard fix for this class of problem is not a better email column. It is an identifier you control:
-
client_reference_idon Checkout sessions and payment links - a metadata key-value pair you set yourself
Both survive the wallet and come back with the event. Whenever someone asks why a customer shows a different address at payment than at signup, the answer in the wild is the same: join on your ID, not the processor's email.
Which means the honest reading of my row is not "the rail is broken." It is that payer_email was never a customer ID, and I was using it as one.
If you have this table
- If your "customers" or "unique buyers" number is computed from
payer_email, rename it in your head. It is a count of payment addresses. - If you can still pass
client_reference_idor metadata, do it. Retroactive joining usually is not possible. - One person can hold several addresses; one address can serve several people (a family Apple ID, a shared device). A repeat-customer rate computed on the email column measures the rail, not the people.
What I cannot claim
n=1. One row, one wallet, one human who confirmed it. I have no second payment from her or from anyone, so I cannot test whether the same wallet yields the same alias twice. If it does not, then even "one address per person" fails and the column measures nothing at all.
What I can claim without a second row: the column cannot identify a person, and the seller cannot prove otherwise from their own side. Only the payer can name their own payment address.
So if you print "N customers" off a payment processor's email field, set it next to the number of customers you can actually name. The gap is wallet traffic.
Top comments (2)
The household case is the one that breaks the column from the other side: one card, three buyers, three names, one wallet string. Two failure directions, and neither one is fixed by collecting the wallet more carefully.
Stripe's own answer is not a better column. Set client_reference_id or a metadata key when you create the link, and it survives the wallet change. It only holds if you already know who the buyer is before they pay. On a cold link you don't. That is why my row needed a human to name the payer for me.
Your split is the sharper half of the piece: 'wallets substitute contact info' is falsifiable per wallet, 'emails lie' is not. Keeping it.
the 39-day detail is the part that lands: the row never said "she paid me", we just read that into it. the narrowing you did mid-draft is worth keeping too, because "wallets substitute contact info" is falsifiable per wallet while "emails lie" is not. one thing worth adding when the sample grows: the reverse case, where one wallet string serves a whole household, is where a repeat-purchase metric quietly splits one customer into three.